Today we're going to talk about the boring topic of product recommendations and the groundbreaking research that's been happening in the academic world.
\nit would have been better if we've got papers on how to make a perfect pancake such as:
1) \"The Role of Pancakes in Modern Society\"
\n2) \"Pancakes: A Comprehensive Review\"
\n3) \"The Efficacy of Various Types of Pancakes\"
\n4) \"How to Make a Perfect Pancake: A Comprehensive Guide\"
But this is not what the competition is about. unfortunately.
\nso, without further ado, here are some of the most renowned and respected papers about product recommendations that have been released in recent months:
Enjoy <3
\nDeep Learning-based Online Alternative Product Recommendations at Scale
\nAlternative recommender systems are critical for ecommerce companies. They guide customers to explore a massive product catalog and assist customers to find the right products among an overwhelming number of options. However, it is a non-trivial task to recommend alternative products that fit customers’ needs. In this paper, we use both textual product information (e.g. product titles and descriptions) and customer behavior data to recommend alternative products. Our results show that the coverage of alternative products is significantly improved in offline evaluations as well as recall and precision. The final A/B test shows that our algorithm increases the conversion rate by 12% in a statistically significant way. In order to better capture the semantic meaning of product information, we build a Siamese Network with Bidirectional LSTM to learn product embeddings. In order to learn a similarity space that better matches the preference of real customers, we use co-compared data from historical customer behavior as labels to train the network. In addition, we use NMSLIB to accelerate the computationally expensive kNN computation for millions of products so that the alternative recommendation is able to scale across the entire catalog of a major ecommerce site.
\nDevelopment of Fashion Product Retrieval and Recommendations Model Based on Deep Learning
\nThe digitization of the fashion industry diversified consumer segments, and consumers now have broader choices with shorter production cycles; digital technology in the fashion industry is attracting the attention of consumers. Therefore, a system that efficiently supports the searching and recommendation of a product is becoming increasingly important. However, the text-based search method has limitations because of the nature of the fashion industry, in which design is a very important factor. Therefore, we developed an intelligent fashion technique based on deep learning for efficient fashion product searches and recommendations consisting of a Sketch-Product fashion retrieval model and vector-based user preference fashion recommendation model. It was found that the “Precision at 5” of the image-based similar product retrieval model was 0.774 and that of the sketch-based similar product retrieval model was 0.445. The vector-based preference fashion recommendation model also showed positive performance. This system is expected to enhance consumers’ satisfaction by supporting users in more effectively searching for fashion products or by recommending fashion products before they begin a search.
\nUsing Deep Learning to Win the Booking.com WSDM WebTour21 Challenge on Sequential Recommendations
\n\n\nTL;DR:- We designed three deep learning architectures based on Multilayer Perceptron (MLP), Gated Recurrent Units (GRU) and XLNet (Transformer) building blocks. - A shared component among the architectures was a Session-based Matrix Factorization head (SMF), which learns a linear mapping between the item (city) embeddings and the session (trip) embeddings to generate recommendations by a dot product operation.
\n
In this paper we present our 1st place solution of the WSDM WebTour 21 Challenge. The competition task was to predict the city of the last booked hotel in a user trip, based on the previously visited cities. For our final solution, we designed three deep learning architectures based on Multilayer Perceptron (MLP), Gated Recurrent Units (GRU) and XLNet (Transformer) building blocks. A shared component among the architectures was a Session-based Matrix Factorization head (SMF), which learns a linear mapping between the item (city) embeddings and the session (trip) embeddings to generate recommendations by a dot product operation. Our final leaderboard result was 0.5939 for precision@4 and scored 2.8% better than the second solution. We published our implementation, using RAPIDS cuDF, TensorFlow and PyTorch, on github.com1.
\nDeep Learning Sentiment Analysis For Recommendations In Social Applications
\nSentiment analysis is a technique for identification of expression, mind-set, or feelings of users and classifies as negative, positive, favorable, unfavorable, etc. from a piece of text in the document. In recent days, deep learning materialized as a valuable way of resolving the sentiment classification problems. The representations are automatically learned in the neural network environment devoid of human labor. But the large-scale data to be trained decides the success of deep learning. In the internet lot of data in the form of comments, opinions are generated by the various websites which gather data about household products of multiple vendors, an incidence that happens in the society in day to day life, etc. Thereby the network users sentiments are derived from their access styles over the web that produces a considerable influence on the other persons who make use of the network to know about a product or political status. Several online sites have grown, and users need suggestions for faster access to the required products with good quality, and hence, recommendation techniques are widely used in various social applications.
\nDetecting Product Adoption Intentions via Multiview Deep Learning
\n\n\nTL;DR:Detecting product adoption intentions on social media could yield significant value in a wide range of applications, such as personalized recommendations and targeted marketing. Existing studies primarily focus on using one type of text representation in target social media posts, ignoring the major yet unexplored potential of fusing different text representations. In this paper, we propose a novel and general multiview deep learning model that simultaneously taps into the capability of multiview learning in leveraging different representations and deep learning in learning latent data representations using a flexible nonlinear transformation. The proposed model leverages three different text representations from a multiview perspective and takes advantage of local and
\n
Detecting product adoption intentions on social media could yield significant value in a wide range of applications, such as personalized recommendations and targeted marketing. In the literature, no study has explored the detection of product adoption intentions on social media, and only a few relevant studies have focused on purchase intention detection for products in one or several categories. Focusing on a product category rather than a specific product is too coarse-grained for precise advertising. Additionally, existing studies primarily focus on using one type of text representation in target social media posts, ignoring the major yet unexplored potential of fusing different text representations. In this paper, we first formulate the problem of product adoption intention mining and demonstrate the necessity of studying this problem and its practical value. To detect a product adoption intention for an individual product, we propose a novel and general multiview deep learning model that simultaneously taps into the capability of multiview learning in leveraging different representations and deep learning in learning latent data representations using a flexible nonlinear transformation. Specifically, the proposed model leverages three different text representations from a multiview perspective and takes advantage of local and long-term word relations by integrating convolutional neural network (CNN) and long short-term memory (LSTM) modules. Extensive experiments on three Twitter datasets demonstrate the effectiveness of the proposed multiview deep learning model compared with the existing benchmark methods. This study also significantly contributes research insights to the literature about intention mining and provides business value to relevant stakeholders such as product providers.
\nImage Based Fashion Product Recommendation with Deep Learning
\nWe develop a two-stage deep learning framework that recommends fashion images based on other input images of similar style. For that purpose, a neural network classifier is used as a data-driven, visually-aware feature extractor. The latter then serves as input for similarity-based recommendations using a ranking algorithm. Our approach is tested on the publicly available Fashion dataset. Initialization strategies using transfer learning from larger product databases are presented. Combined with more traditional content-based recommendation systems, our framework can help to increase robustness and performance, for example, by better matching a particular customer style.
\nMulti-level Deep Learning based e-Commerce Product Categorization
\n\n\nTL;DR:Our proposed classification model is based on a multi-level and multi-class deep learning tree. This model constructs multiple models based on single-label and multi-level label predictions as well as the characteristics of the product tree structure. These models are then combined to generate a new classification model. The proposed classification model is tested on the online test dataset, and the accuracy, recall, and F1 score are 0.8552, 0.8389, 0.8404 in Stage 1 and 0.8397, 0.8428, 0.8379 in Stage 2 respectively
\n
E-commerce product categorization is an important topic, and its quality directly affects subsequent search, recommendations and related personalized services. E-commerce product classification is challenging due to the large scale and complexity of the product information and categories. In the E-Commerce Text Classification Challenge, we combine machine learning, deep learning, and natural language processing to propose a multi-level and multi-class deep learning tree method. Our method constructs multiple models based on single-label and multi-level label predictions as well as the characteristics of the product tree structure and combines the multiple models to generate a new classification model. The proposed classification model is tested on the online test dataset. The accuracy, recall, and F1 score are 0.8552, 0.8389, 0.8404 in leaderboard(Stage 1) and 0.8397, 0.8428, 0.8379 in leaderboard(Stage 2) respectively, ranking among top 3 scorers.
\nDeep Retrieval: Learning A Retrievable Structure for Large-Scale Recommendations
\nOne of the core problems in large-scale recommendations is to retrieve top relevant candidates accurately and efficiently, preferably in sub-linear time. Previous approaches are mostly based on a two-step procedure: first learn an inner-product model, and then use some approximate nearest neighbor (ANN) search algorithm to find top candidates. In this paper, we present Deep Retrieval (DR), to learn a retrievable structure directly with user-item interaction data (e.g. clicks) without resorting to the Euclidean space assumption in ANN algorithms. DR's structure encodes all candidate items into a discrete latent space. Those latent codes for the candidates are model parameters and learnt together with other neural network parameters to maximize the same objective function. With the model learnt, a beam search over the structure is performed to retrieve the top candidates for reranking. Empirically, we first demonstrate that DR, with sub-linear computational complexity, can achieve almost the same accuracy as the brute-force baseline on two public datasets. Moreover, we show that, in a live production recommendation system, a deployed DR approach significantly outperforms a well-tuned ANN baseline in terms of engagement metrics. To the best of our knowledge, DR is among the first non-ANN algorithms successfully deployed at the scale of hundreds of millions of items for industrial recommendation systems.
\nThe COVID-19 pandemic markedly changed the human shopping nature, necessitating a contactless shopping system to curb the spread of the contagious disease efficiently. Consequently, a customer opts for a store where it is possible to avoid physical contacts and shorten the shopping process with extended services such as personalized product recommendations. Automatic age and gender estimation of a customer in a smart store strongly benefit the consumer by providing personalized advertisement and product recommendation; similarly, it aids the smart store proprietor to promote sales and develop an inventory perpetually for the future retail. In our paper, we propose a deep learning-founded enterprise solution for smart store customer relationship management (CRM), which allows us to predict the age and gender from a customer’s face image taken in an unconstrained environment to facilitate the smart store’s extended services, as it is expected for a modern venture. For the age estimation problem, we mitigate the data sparsity problem of the large public IMDB-WIKI dataset by image enhancement from another dataset and perform data augmentation as required. We handle our classification tasks utilizing an empirically leading pre-trained convolutional neural network (CNN), the VGG-16 network, and incorporate batch normalization. Especially, the age estimation task is posed as a deep classification problem followed by a multinomial logistic regression first-moment refinement. We validate our system for two standard benchmarks, one for each task, and demonstrate state-of-the-art performance for both real age and gender estimation.
\nDeep Learning Based Matrix Factorization For Collaborative Filtering
\nCollaborative Filtering based on matrix factorization (MF) has shown tremendous success in the field recommender system. However, MF has difficulty in handling sparsity and scalability. These resulted in low quality of recommendations. In this regard, deep learning has shown immense success in different application areas including recommender systems. To address the limitations, we incorporate deep learning architecture to matrix factorization and develop a novel mode. The core idea of the method is to map users and items input vector to two well-structured deep neural network architectures separately for factorization. Then, we incorporate inner product to the output layers of the network to predict the rating scores. The use of this structure significantly improve the quality of recommendation. The experimental result on real data sets shows that our proposed model outperformed state of the art methods.
\nResearch on Commodity Recommendation System Based on Deep Learning
\nThe traditional matrix factorization model cannot effectively extract the features of users and items, but the feature information can be extracted well based on the deep learning model. At present, the mainstream recommendation algorithms based on deep learning only make recommendation prediction in the form of the product of neural network output or item features and user features, and cannot fully mine the relationship between users and items. Based on this, this paper proposes a recommendation algorithm based on the combination of text convolution neural network and singular value decomposition (Bias SVD) with biased terms. The text convolution neural network (Text CNN) is used to fully extract the feature information of users and items, and then the singular value decomposition method is used to make recommendations to deeply understand the document context information and further improve the accuracy of recommendation. The algorithm is widely evaluated and analyzed on two real data sets of MovieLens, and the accuracy of recommendation is obviously better than that of ConvMF algorithm and mainstream deep learning recommendation algorithm.
\nA Deep Learning Approach of Collaborative Filtering to Recommender System with Opinion Mining
\nTo produce good quality recommendations for large or enterprise scale problems, a competent approach for recommender system is required. This paper presents such an approach which first generates the text score based on users’ reviews with the help of opinion mining. It then feeds ratings corresponding to the text scores to Convolutional Neural Network (CNN). CNN learns and does the dot product of user and product matrices. It is a special kind of feed forward neural network of deep learning technique to get better predictions in a product recommender system. The work done in this paper has improved accuracy and user satisfaction to great extent using CNN. It also helps e-commerce companies to increase the revenue by recommending closest products to users.
\nDeep Learning Technique for selecting appropriate Beauty Care Products for different skin Type
\nThis research has been undertaken to ease the challenging task of the beauty industry by using Deep Learning Method. Nowadays, the cosmetic product plays a major role in the appearance of personality. Customers are given a number of items with online shopping and e-commerce websites. It's difficult for us to pick the best product for our skin. Over recent years recommendation systems have been commonly used for providing user recommendations in various commercial platforms. The sparsity of the data and the scalability of the method however limit the performance of the algorithms used for recommendation and it is difficult to further improve the quality of the results of the recommendations. Hence we propose a predictive system that provides a precise idea of which product is best for our skin type using the Deep Learning Technique. The suggestion is based on the types of skin that might be Normal, Combination, Dry, Oily, and Sensitive. We have implemented the Deep Neural Network (DNN) model for cosmetic product composition. Finally, it is validated, by comparing with other recommendation algorithms on our generated dataset, that our model can effectively boost the recommendation performance. IndexTerms – Deep Learning, Deep Neural Network, Cosmetic Product Suggestion.
\n\n\nTL;DR:(1) deep learning model is used to extract the product characteristics from sparse inputs, (2) traditional collaborative filtering algorithm is updated with the characteristics obtained by deep learning model for sparse rated inputs, which greatly improves the user and product transaction.
\n
The exponential increase in the volume of online data has generated a confront of overburden of data for online users, which slow down the suitable access to products of pursuit on the Web. This contributed to the need for recommendation systems. Recommender system is a special form of intelligent technique that takes advantage of past user transactions on products to give recommendations of products. Collaborative filtering has turn out to be the commonly adopted method of providing users with customized services, except that it endures the problem of sparsely rated inputs. For collaborative filtering, we introduce a deep learning-based architecture which evaluates a discrete factorisation of vectors from sparse inputs. The characteristics of the products are retrieved using a deep learning model, denoising auto encoders. The traditional collaborative filtering algorithm that predicts and uses the past history of consumer interest and product characteristics are updated with the characteristics obtained by deep learning model for sparse rated inputs. The results of sparse data problem tested on MovieLens data set will greatly enhance the user and product transaction.
\nDerin Öğrenme ile Kıyafet Kombin Önerim Sistemi Cloth Combine Estimation System Using Deep Learning
\n\n\nTL;DR:The goal of this study was to develop a deep neural network (DNN) based algorithm for combining clothes. The dataset used in this study was taken from an international cloth company’s e-commerce website. The algorithm was trained on a dataset of 2,521 cloth combinations. The results of the study showed that the DNN based algorithm was able to successfully combine different clothes together.
\n
The clothing companies are paying more attention on meeting their customers on digital platforms and perform their marketing over the cloth combines which are created by fashion designers. Moreover, the companies on e- commerce platform use cloth recommendation system to draw their customers attention and perform sales. The customers browsing the e-commerce sites are facing some cloth recommendations based on the statistics such as “those looking at this product also looked at this product”. However, beside of statistical recommendations, more intelligence is required to recommend the cloth which combines the cloth that customers own or going to own. In this study, a cloth combine completion system is developed for e-commerce. The system is based deep neural network (DNN) and it is intended to imitate the fashion designer of the brands on digital market. The study is performed over the cloth combines shown on the e-commerce site of an international cloth company. In contrast to applications based on image processing, the clothes are parameterized depending on expert knowledge and a deep neural network is trained to find the best bottom clothing to the selected top clothing. Using the trained network, the brands can create user specific cloth combines as if the combines are performed by fashion-designer.
\nComplementary Outfit Recommendation Using Deep Learning
\n\n\nTL;DR:To recommend complementary outfits to users, we used deep learning methods such as Convolutional Neural Networks and Convolutional Autoencoders. Our system was able to recommend similar apparel items to the user’s input from their wardrobe, and improve the recommendations of e-commerce platforms.
\n
Abstract: Developing a system that provides a complementary outfit based on user queries can be challenging due to its complexity and subjectivity. At present, the methodologies available are used to recommend outfits to the users according to meta-data and user’s location. So we present an empirical study on the application where we make use of Deep Convolutional Neural Networks (DCNN) to the task of classifying apparel and user’s input from their wardrobe along with Convolutional Autoencoder which is capable of finding similar product images with the aim to remove the usage of product tags and improve the recommendations of e-commerce platforms. Choosing the right pair of clothes according to the customer’s choice and preference is of utmost importance. Sometimes people find it difficult to keep up with the trend so in such cases this project will be very helpful for them. Also, this helps the user to save their time and energy which implies that the user need not spend much time worrying about “what goes well with what”. Hence we have come up with a system that is capable of resolving such difficulties and whose primary purpose is to make the user’s work easier and beneficial.
\nRise of Deep Learning for Genomic, Proteomic, and Metabolomic Data Integration in Precision Medicine
\n\n\nTL;DR:Machine learning (ML), deep learning (DL), and artificial neural networks (ANNs) are being ubiquitously incorporated into everyday products such as Internet search, email spam filters, product recommendations, image classification, and speech recognition. Many approaches incorporate complex artificial neural network architectures and are collectively referred to as DL applications. These methods have been shown capable of representing and learning relationships in data in many diverse forms. Omics data pose considerable challenges for DL because of low signal to noise, analytical variance, and complex data integration requirements. However, DL models have already been shown capable of improving the ease of data encoding and predictive model performance over
\n
Abstract Machine learning (ML) is being ubiquitously incorporated into everyday products such as Internet search, email spam filters, product recommendations, image classification, and speech recognition. New approaches for highly integrated manufacturing and automation such as the Industry 4.0 and the Internet of things are also converging with ML methodologies. Many approaches incorporate complex artificial neural network architectures and are collectively referred to as deep learning (DL) applications. These methods have been shown capable of representing and learning predictable relationships in many diverse forms of data and hold promise for transforming the future of omics research and applications in precision medicine. Omics and electronic health record data pose considerable challenges for DL. This is due to many factors such as low signal to noise, analytical variance, and complex data integration requirements. However, DL models have already been shown capable of both improving the ease of data encoding and predictive model performance over alternative approaches. It may not be surprising that concepts encountered in DL share similarities with those observed in biological message relay systems such as gene, protein, and metabolite networks. This expert review examines the challenges and opportunities for DL at a systems and biological scale for a precision medicine readership.
\nProduct Matching Lessons and Recommendations from a Real World Application
\nRetailers rely heavily on product matching to better serve their customers, and to improve their modelling and forecasting. Product matching refers to the process of identifying similar or identical products across different data sources. This is a challenging problem as standardized unique identifiers are not used consistently across retailers, and product descriptions and characteristics vary across data collections. In this paper we present and discuss lessons learned from product matching in a real world application. We propose an evaluation framework where we investigate and compare the use of traditional machine learning methods and deep learning methods on public and proprietary datasets for product matching. Our findings show that traditional machine learning methods perform well on this task and a practitioner should investigate these methods first.
\nModeling Consumer Buying Decision for Recommendation Based on Multi-Task Deep Learning
\nAlthough marketing researchers and sociologists have recognized the importance of buying decision process and its significant influence on consumer's purchasing behaviors, existing recommender systems do not explicitly model the consumer buying decision process or capture the sequential regularities of what happens before and after each purchase. In this paper, we try to bridge the gap and improve recommendation systems by explicitly modeling consumer buying decision process and corresponding stages. In particular, we propose a multi-task learning model with long short-term memory networks (LSTM) to learn consumer buying decision process. It maps items, users, product categories, and the behavior sequences into real valued vectors, with which the probability of purchasing a product can be estimated. In this way, the model can capture user intentions and preferences, predicts the conversion rate of each candidate product, and makes recommendations accordingly. Experiments on real world data demonstrate the effectiveness of the proposed approach.
\nExtraction of Visual Features for Recommendation of Products via Deep Learning
\nIn this paper (The first author is the 1st place winner of the Open HSE Student Research Paper Competition (NIRS) in 2017, Computer Science nomination, with the topic “Extraction of Visual Features for Recommendation of Products”, as alumni of 2017 “Data Science” master program at Computer Science Faculty, HSE, Moscow), we describe a special recommender approach based on features extracted from the clothes’ images. The method of feature extraction relies on pre-trained deep neural network that follows transfer learning on the dataset. Recommendations are generated by the neural network as well. All the experiments are based on the items of category Clothing, Shoes and Jewelry from Amazon product dataset. It is demonstrated that the proposed approach outperforms the baseline collaborative filtering method.
\nLearning An End-to-End Structure for Retrieval in Large-Scale Recommendations
\nOne of the core problems in large-scale recommendations is to retrieve top relevant candidates accurately and efficiently, preferably in sub-linear time. Previous approaches are mostly based on a two-step procedure: first learn an inner-product model, and then use some approximate nearest neighbor (ANN) search algorithm to find top candidates. In this paper, we present Deep Retrieval (DR), to learn a retrievable structure directly with user-item interaction data (e.g. clicks) without resorting to the Euclidean space assumption in ANN algorithms. DR's structure encodes all candidate items into a discrete latent space. Those latent codes for the candidates are model parameters and learnt together with other neural network parameters to maximize the same objective function. With the model learnt, a beam search over the structure is performed to retrieve the top candidates for reranking. Empirically, we first demonstrate that DR, with sub-linear computational complexity, can achieve almost the same accuracy as the brute-force baseline on two public datasets. Moreover, we show that, in a live production recommendation system, a deployed DR approach significantly outperforms a well-tuned ANN baseline in terms of engagement metrics. To the best of our knowledge, DR is among the first non-ANN algorithms successfully deployed at the scale of hundreds of millions of items for industrial recommendation systems.
\nComparative study on traditional recommender systems and deep learning based recommender systems
\n\n\nTL;DR:Recommender systems is a big breakthrough for the field of e-commerce. Product recommendation is a challenging task to e-commerce companies. The traditional recommender systems provide the solutions in recommending the products. This in turn helps companies to generate good revenue. Nowadays, Deep Learning is being used in every domain. Deep Learning techniques in the field of recommender systems can be directly applied. Deep Learning has ample number of algorithms which can be used to give recommendations to users to purchase products. In this paper, the performance of traditional recommender systems and deep learning-based recommender systems are compared
\n
Recommender systems is a big breakthrough for the field of e-commerce. Product recommendation is challenging task to e-commerce companies. Traditional Recommender Systems provided the solutions in recommending the products. This in turn help companies to generate good revenue. Now a day Deep Learning is using in every domain. Deep Learning techniques in the field of Recommender Systems can be directly applied. Deep Learning has ample number of algorithms. These algorithms can be used to give recommendations to users to purchase products. In this paper performance of Traditional Recommender Systems and Deep Learning-based Recommender Systems are compared.
\nFrom Web to Physical and Back: WP User Profiling with Deep Learning
\n\n\nTL;DR:we use supervised machine learning on a web crawler to extract features from web pages (e.g. title, text, links) and then we use these features to predict the pavilion a person is most likely to visit at a fair.
\n
This position paper discusses the definition and implementation of Web-Physical (WP) user profiles, which allow the creation of personalized recommendations and innovative behavioral predictions in particular scenarios, i.e., fairs. The nature of a WP profile builds upon two different worlds: the Web (social networks and web applications) and the Physical one, each one of them being explored through (big) data collection platforms. These two platforms collect radically different information: on the one hand, information of appreciation towards a particular product or service (web domain) together with other metadata; on the other, the leases (x, y) of users in the exhibition space (physical domain). In this scenario, our research idea consists in identifying how the information in the two domains can be merged in a whole entity under a theoretical point of view: this will unleash tangible repercussions in terms of personalized recommendations and effective behavioral predictions, where with personalized recommendation we mean a suggestion to a user in physical terms (eg a pavilion to visit) and / or in web terms (eg a site to visit) and with behavioral prediction a prediction of where a user can go in the future, even in a multimedia perspective (physical + web).
\nBoosting Recommender Systems with Deep Learning
\n\n\nTL;DR:(1) using deep learning methods to extract visual features from images of products, (2) using deep learning to learn relationships between products that are stylistically complementary, and (3) using deep learning for other purposes across the Farfetch platform.
\n
Farfetch is a global fashion marketplace with a catalog that, at any time, has over 200 000 products spanning over 2000 brands from luxury boutiques all around the world. Finding the right product to the right customer is a challenge that, we, as Data Scientists working on the Recommendations team, are trying to solve using state-of-the art algorithms and disruptive technologies. Deep learning (DL) is an area of Machine Learning that has recently been brought to the spotlight for its breakthrough results across several domains. In this talk, we will provide an overview of some ongoing projects in which Deep Learning methods play a major role. A common problem in online marketplaces with large catalogs such as ours is the lack of detailed metadata about the products. Particularly, features such as style, colors, pattern, occasion, sizing, etc., are known to drive customers intent but are hard to catalog manually, in a consistent way. We explain how we use our extensive dataset of normalized product images together with state of the art convolutional neural networks to extract visual features. These can then be used to provide better recommendations and improve other applications across the platform. Another application of DL is to capture relationships between products which are stylistically complementary. We leverage data from thousands of hand-curated outfits to model these intangible fashion concepts only a human specialist can provide, and generalize a method of recommending complementary products using deep siamese neural networks. This an alternative recommendation strategy that can be used to drive cross-sell opportunities. These are merely a sample of problems we are tackling using DL at Farfetch. We believe that there are plenty of opportunities for application of these techniques to recommender systems and we look forward to discussing the potentials of this stream of research with the RecSys community.
\nA Deep Dive Into Understanding TheRandom Walk-Based Temporal Graph Learning
\n\n\nTL;DR:We study a more scalable graph learning algorithm based on random walks, which operates on dynamic input graphs and has attracted less attention in the architecture community compared to GCN. We propose high-performance CPU and GPU implementations of two important graph learning tasks, that cover a broad class of applications, using random walks on continuous-time dynamic graphs: link prediction and node classification.
\n
Machine learning on graph data has gained significant interest because of its applicability to various domains ranging from product recommendations to drug discovery. While there is a rapid growth in the algorithmic community, the computer architecture community has so far focused on a subset of graph learning algorithms including Graph Convolution Network (GCN), and a few others. In this paper, we study another, more scalable, graph learning algorithm based on random walks, which operates on dynamic input graphs and has attracted less attention in the architecture community compared to GCN. We propose high-performance CPU and GPU implementations of two important graph learning tasks, that cover a broad class of applications, using random walks on continuous-time dynamic graphs: link prediction and node classification. We show that the resulting workload exhibits distinct characteristics, measured in terms of irregularity, core and memory utilization, and cache hit rates, compared to graph traversals, deep learning, and GCN. We further conduct an in-depth performance analysis focused on both algorithm and hardware to guide future software optimization and architecture exploration. The algorithm-focused study presents a rich trade-off space between algorithmic performance and runtime complexity to identify optimization opportunities. We find an optimal hyperparameter setting that strikes balance in this trade-off space. Using this setting, we also perform a detailed microarchitectural characterization to analyze hardware behavior of these applications and uncover execution bottlenecks, which include high cache misses and dependency-related stalls. The outcome of our study includes recommendations for further performance optimization, and open-source implementations for future investigation.
\nA Deep Dive Into Understanding The Random Walk-Based Temporal Graph Learning
\n\n\nTL;DR:We study a scalable graph learning algorithm based on random walks and propose high-performance CPU and GPU implementations of two importantgraph learning tasks on dynamic input graphs. We show that the resulting workload exhibits distinct characteristics, measured in terms of irregularity, core and memory utilization, and cache hit rates. Our microarchitectural characterization identifies execution bottlenecks, which include high cache misses and dependency-related stalls.
\n
Machine learning on graph data has gained significant interest because of its applicability to various domains ranging from product recommendations to drug discovery. While there is a rapid growth in the algorithmic community, the computer architecture community has so far focused on a subset of graph learning algorithms including Graph Convolution Network (GCN), and a few others. In this paper, we study another, more scalable, graph learning algorithm based on random walks, which operates on dynamic input graphs and has attracted less attention in the architecture community compared to GCN. We propose high-performance CPU and GPU implementations of two important graph learning tasks, that cover a broad class of applications, using random walks on continuous-time dynamic graphs: link prediction and node classification. We show that the resulting workload exhibits distinct characteristics, measured in terms of irregularity, core and memory utilization, and cache hit rates, compared to graph traversals, deep learning, and GCN. We further conduct an in-depth performance analysis focused on both algorithm and hardware to guide future software optimization and architecture exploration. The algorithm-focused study presents a rich trade-off space between algorithmic performance and runtime complexity to identify optimization opportunities. We find an optimal hyperparameter setting that strikes balance in this trade-off space. Using this setting, we also perform a detailed microarchitectural characterization to analyze hardware behavior of these applications and uncover execution bottlenecks, which include high cache misses and dependency-related stalls. The outcome of our study includes recommendations for further performance optimization, and open-source implementations for future investigation.
\nReinforcement Learning-based Product Delivery Frequency Control
\nFrequency control is an important problem in modern recommender systems. It dictates the delivery frequency of recommendations to maintain product quality and efficiency. For example, the frequency of delivering promotional notifications impacts daily metrics as well as the infrastructure resource consumption (e.g. CPU and memory usage). There remain open questions on what objective we should optimize to represent business values in the long term best, and how we should balance between daily metrics and resource consumption in a dynamically fluctuating environment. We propose a personalized methodology for the frequency control problem, which combines long-term value optimization using reinforcement learning (RL) with a robust volume control technique we termed “Effective Factor”. We demonstrate statistically significant improvement in daily metrics and resource efficiency by our method in several notification applications at a scale of billions of users. To our best knowledge, our study represents the first deep RL application on the frequency control problem at such an industrial scale.
\nLearning Graph Neural Networks with Deep Graph Library
\n\n\nTL;DR:This tutorial provides an overview of graph neural networks (GNNs), discussing the types of problems that GNNs are well suited for, and introducing some of the most widely used GNN model architectures and problems/applications that are designed to solve. It also introduces the Deep Graph Library (DGL), a new software framework that simplifies the development of efficient GNN-based training and inference programs.
\n
Learning from graph and relational data plays a major role in many applications including social network analysis, marketing, e-commerce, information retrieval, knowledge modeling, medical and biological sciences, engineering, and others. In the last few years, Graph Neural Networks (GNNs) have emerged as a promising new supervised learning framework capable of bringing the power of deep representation learning to graph and relational data. This ever-growing body of research has shown that GNNs achieve state-of-the-art performance for problems such as link prediction, fraud detection, target-ligand binding activity prediction, knowledge-graph completion, and product recommendations. The objective of this tutorial is twofold. First, it will provide an overview of the theory behind GNNs, discuss the types of problems that GNNs are well suited for, and introduce some of the most widely used GNN model architectures and problems/applications that are designed to solve. Second, it will introduce the Deep Graph Library (DGL), a new software framework that simplifies the development of efficient GNN-based training and inference programs. To make things concrete, the tutorial will provide hands-on sessions using DGL. This hands-on part will cover both basic graph applications (e.g., node classification and link prediction), as well as more advanced topics including training GNNs on large graphs and in a distributed setting. In addition, it will provide hands-on tutorials on using GNNs and DGL for real-world applications such as recommendation and fraud detection.
\nA Meta-Learning Perspective on Cold-Start Recommendations for Items
\n\n\nTL;DR:We propose two deep neural network architectures that implement our meta-learning strategy. The first architecture learns a linear classifier whose weights are determined by the item history while the second architecture learns a neural network whose biases are instead adjusted.
\n
Matrix factorization (MF) is one of the most popular techniques for product recommendation, but is known to suffer from serious cold-start problems. Item cold-start problems are particularly acute in settings such as Tweet recommendation where new items arrive continuously. In this paper, we present a meta-learning strategy to address item cold-start when new items arrive continuously. We propose two deep neural network architectures that implement our meta-learning strategy. The first architecture learns a linear classifier whose weights are determined by the item history while the second architecture learns a neural network whose biases are instead adjusted. We evaluate our techniques on the real-world problem of Tweet recommendation. On production data at Twitter, we demonstrate that our proposed techniques significantly beat the MF baseline and also outperform production models for Tweet recommendation.
\nMotivation Factors Analysis and Policy Research of Deep Learning
\nThis article uses Interpretative Structural Modeling Method(ISM) to build deep learning promotion structure diagram of the influence factors, then uses Analytic Hierarchy Process(AHP) to determine the relative importance of various factors, according to the evaluation results it is concluded that the policy guidelines for deep learning to promote the influence degree of the relative maximum conclusion, put forward to promote deep learning better and faster, the government related department should publish relevant policy recommendations, embodied in more research funds, set up special introduction and training of research institutions and researchers, to perfect the theory system. At the same time, the improvement of social awareness will attract more high-tech companies in product research and development, making deep learning applied in more fields.
\nToday we're going to talk about the boring topic of product recommendations and the groundbreaking research that's been happening in the academic world.
\nit would have been better if we've got papers on how to make a perfect pancake such as:
1) \"The Role of Pancakes in Modern Society\"
\n2) \"Pancakes: A Comprehensive Review\"
\n3) \"The Efficacy of Various Types of Pancakes\"
\n4) \"How to Make a Perfect Pancake: A Comprehensive Guide\"
But this is not what the competition is about. unfortunately.
\nso, without further ado, here are some of the most renowned and respected papers about product recommendations that have been released in recent months:
Enjoy <3
\nDeep Learning-based Online Alternative Product Recommendations at Scale
\nAlternative recommender systems are critical for ecommerce companies. They guide customers to explore a massive product catalog and assist customers to find the right products among an overwhelming number of options. However, it is a non-trivial task to recommend alternative products that fit customers’ needs. In this paper, we use both textual product information (e.g. product titles and descriptions) and customer behavior data to recommend alternative products. Our results show that the coverage of alternative products is significantly improved in offline evaluations as well as recall and precision. The final A/B test shows that our algorithm increases the conversion rate by 12% in a statistically significant way. In order to better capture the semantic meaning of product information, we build a Siamese Network with Bidirectional LSTM to learn product embeddings. In order to learn a similarity space that better matches the preference of real customers, we use co-compared data from historical customer behavior as labels to train the network. In addition, we use NMSLIB to accelerate the computationally expensive kNN computation for millions of products so that the alternative recommendation is able to scale across the entire catalog of a major ecommerce site.
\nDevelopment of Fashion Product Retrieval and Recommendations Model Based on Deep Learning
\nThe digitization of the fashion industry diversified consumer segments, and consumers now have broader choices with shorter production cycles; digital technology in the fashion industry is attracting the attention of consumers. Therefore, a system that efficiently supports the searching and recommendation of a product is becoming increasingly important. However, the text-based search method has limitations because of the nature of the fashion industry, in which design is a very important factor. Therefore, we developed an intelligent fashion technique based on deep learning for efficient fashion product searches and recommendations consisting of a Sketch-Product fashion retrieval model and vector-based user preference fashion recommendation model. It was found that the “Precision at 5” of the image-based similar product retrieval model was 0.774 and that of the sketch-based similar product retrieval model was 0.445. The vector-based preference fashion recommendation model also showed positive performance. This system is expected to enhance consumers’ satisfaction by supporting users in more effectively searching for fashion products or by recommending fashion products before they begin a search.
\nUsing Deep Learning to Win the Booking.com WSDM WebTour21 Challenge on Sequential Recommendations
\n\n\nTL;DR:- We designed three deep learning architectures based on Multilayer Perceptron (MLP), Gated Recurrent Units (GRU) and XLNet (Transformer) building blocks. - A shared component among the architectures was a Session-based Matrix Factorization head (SMF), which learns a linear mapping between the item (city) embeddings and the session (trip) embeddings to generate recommendations by a dot product operation.
\n
In this paper we present our 1st place solution of the WSDM WebTour 21 Challenge. The competition task was to predict the city of the last booked hotel in a user trip, based on the previously visited cities. For our final solution, we designed three deep learning architectures based on Multilayer Perceptron (MLP), Gated Recurrent Units (GRU) and XLNet (Transformer) building blocks. A shared component among the architectures was a Session-based Matrix Factorization head (SMF), which learns a linear mapping between the item (city) embeddings and the session (trip) embeddings to generate recommendations by a dot product operation. Our final leaderboard result was 0.5939 for precision@4 and scored 2.8% better than the second solution. We published our implementation, using RAPIDS cuDF, TensorFlow and PyTorch, on github.com1.
\nDeep Learning Sentiment Analysis For Recommendations In Social Applications
\nSentiment analysis is a technique for identification of expression, mind-set, or feelings of users and classifies as negative, positive, favorable, unfavorable, etc. from a piece of text in the document. In recent days, deep learning materialized as a valuable way of resolving the sentiment classification problems. The representations are automatically learned in the neural network environment devoid of human labor. But the large-scale data to be trained decides the success of deep learning. In the internet lot of data in the form of comments, opinions are generated by the various websites which gather data about household products of multiple vendors, an incidence that happens in the society in day to day life, etc. Thereby the network users sentiments are derived from their access styles over the web that produces a considerable influence on the other persons who make use of the network to know about a product or political status. Several online sites have grown, and users need suggestions for faster access to the required products with good quality, and hence, recommendation techniques are widely used in various social applications.
\nDetecting Product Adoption Intentions via Multiview Deep Learning
\n\n\nTL;DR:Detecting product adoption intentions on social media could yield significant value in a wide range of applications, such as personalized recommendations and targeted marketing. Existing studies primarily focus on using one type of text representation in target social media posts, ignoring the major yet unexplored potential of fusing different text representations. In this paper, we propose a novel and general multiview deep learning model that simultaneously taps into the capability of multiview learning in leveraging different representations and deep learning in learning latent data representations using a flexible nonlinear transformation. The proposed model leverages three different text representations from a multiview perspective and takes advantage of local and
\n
Detecting product adoption intentions on social media could yield significant value in a wide range of applications, such as personalized recommendations and targeted marketing. In the literature, no study has explored the detection of product adoption intentions on social media, and only a few relevant studies have focused on purchase intention detection for products in one or several categories. Focusing on a product category rather than a specific product is too coarse-grained for precise advertising. Additionally, existing studies primarily focus on using one type of text representation in target social media posts, ignoring the major yet unexplored potential of fusing different text representations. In this paper, we first formulate the problem of product adoption intention mining and demonstrate the necessity of studying this problem and its practical value. To detect a product adoption intention for an individual product, we propose a novel and general multiview deep learning model that simultaneously taps into the capability of multiview learning in leveraging different representations and deep learning in learning latent data representations using a flexible nonlinear transformation. Specifically, the proposed model leverages three different text representations from a multiview perspective and takes advantage of local and long-term word relations by integrating convolutional neural network (CNN) and long short-term memory (LSTM) modules. Extensive experiments on three Twitter datasets demonstrate the effectiveness of the proposed multiview deep learning model compared with the existing benchmark methods. This study also significantly contributes research insights to the literature about intention mining and provides business value to relevant stakeholders such as product providers.
\nImage Based Fashion Product Recommendation with Deep Learning
\nWe develop a two-stage deep learning framework that recommends fashion images based on other input images of similar style. For that purpose, a neural network classifier is used as a data-driven, visually-aware feature extractor. The latter then serves as input for similarity-based recommendations using a ranking algorithm. Our approach is tested on the publicly available Fashion dataset. Initialization strategies using transfer learning from larger product databases are presented. Combined with more traditional content-based recommendation systems, our framework can help to increase robustness and performance, for example, by better matching a particular customer style.
\nMulti-level Deep Learning based e-Commerce Product Categorization
\n\n\nTL;DR:Our proposed classification model is based on a multi-level and multi-class deep learning tree. This model constructs multiple models based on single-label and multi-level label predictions as well as the characteristics of the product tree structure. These models are then combined to generate a new classification model. The proposed classification model is tested on the online test dataset, and the accuracy, recall, and F1 score are 0.8552, 0.8389, 0.8404 in Stage 1 and 0.8397, 0.8428, 0.8379 in Stage 2 respectively
\n
E-commerce product categorization is an important topic, and its quality directly affects subsequent search, recommendations and related personalized services. E-commerce product classification is challenging due to the large scale and complexity of the product information and categories. In the E-Commerce Text Classification Challenge, we combine machine learning, deep learning, and natural language processing to propose a multi-level and multi-class deep learning tree method. Our method constructs multiple models based on single-label and multi-level label predictions as well as the characteristics of the product tree structure and combines the multiple models to generate a new classification model. The proposed classification model is tested on the online test dataset. The accuracy, recall, and F1 score are 0.8552, 0.8389, 0.8404 in leaderboard(Stage 1) and 0.8397, 0.8428, 0.8379 in leaderboard(Stage 2) respectively, ranking among top 3 scorers.
\nDeep Retrieval: Learning A Retrievable Structure for Large-Scale Recommendations
\nOne of the core problems in large-scale recommendations is to retrieve top relevant candidates accurately and efficiently, preferably in sub-linear time. Previous approaches are mostly based on a two-step procedure: first learn an inner-product model, and then use some approximate nearest neighbor (ANN) search algorithm to find top candidates. In this paper, we present Deep Retrieval (DR), to learn a retrievable structure directly with user-item interaction data (e.g. clicks) without resorting to the Euclidean space assumption in ANN algorithms. DR's structure encodes all candidate items into a discrete latent space. Those latent codes for the candidates are model parameters and learnt together with other neural network parameters to maximize the same objective function. With the model learnt, a beam search over the structure is performed to retrieve the top candidates for reranking. Empirically, we first demonstrate that DR, with sub-linear computational complexity, can achieve almost the same accuracy as the brute-force baseline on two public datasets. Moreover, we show that, in a live production recommendation system, a deployed DR approach significantly outperforms a well-tuned ANN baseline in terms of engagement metrics. To the best of our knowledge, DR is among the first non-ANN algorithms successfully deployed at the scale of hundreds of millions of items for industrial recommendation systems.
\nThe COVID-19 pandemic markedly changed the human shopping nature, necessitating a contactless shopping system to curb the spread of the contagious disease efficiently. Consequently, a customer opts for a store where it is possible to avoid physical contacts and shorten the shopping process with extended services such as personalized product recommendations. Automatic age and gender estimation of a customer in a smart store strongly benefit the consumer by providing personalized advertisement and product recommendation; similarly, it aids the smart store proprietor to promote sales and develop an inventory perpetually for the future retail. In our paper, we propose a deep learning-founded enterprise solution for smart store customer relationship management (CRM), which allows us to predict the age and gender from a customer’s face image taken in an unconstrained environment to facilitate the smart store’s extended services, as it is expected for a modern venture. For the age estimation problem, we mitigate the data sparsity problem of the large public IMDB-WIKI dataset by image enhancement from another dataset and perform data augmentation as required. We handle our classification tasks utilizing an empirically leading pre-trained convolutional neural network (CNN), the VGG-16 network, and incorporate batch normalization. Especially, the age estimation task is posed as a deep classification problem followed by a multinomial logistic regression first-moment refinement. We validate our system for two standard benchmarks, one for each task, and demonstrate state-of-the-art performance for both real age and gender estimation.
\nDeep Learning Based Matrix Factorization For Collaborative Filtering
\nCollaborative Filtering based on matrix factorization (MF) has shown tremendous success in the field recommender system. However, MF has difficulty in handling sparsity and scalability. These resulted in low quality of recommendations. In this regard, deep learning has shown immense success in different application areas including recommender systems. To address the limitations, we incorporate deep learning architecture to matrix factorization and develop a novel mode. The core idea of the method is to map users and items input vector to two well-structured deep neural network architectures separately for factorization. Then, we incorporate inner product to the output layers of the network to predict the rating scores. The use of this structure significantly improve the quality of recommendation. The experimental result on real data sets shows that our proposed model outperformed state of the art methods.
\nResearch on Commodity Recommendation System Based on Deep Learning
\nThe traditional matrix factorization model cannot effectively extract the features of users and items, but the feature information can be extracted well based on the deep learning model. At present, the mainstream recommendation algorithms based on deep learning only make recommendation prediction in the form of the product of neural network output or item features and user features, and cannot fully mine the relationship between users and items. Based on this, this paper proposes a recommendation algorithm based on the combination of text convolution neural network and singular value decomposition (Bias SVD) with biased terms. The text convolution neural network (Text CNN) is used to fully extract the feature information of users and items, and then the singular value decomposition method is used to make recommendations to deeply understand the document context information and further improve the accuracy of recommendation. The algorithm is widely evaluated and analyzed on two real data sets of MovieLens, and the accuracy of recommendation is obviously better than that of ConvMF algorithm and mainstream deep learning recommendation algorithm.
\nA Deep Learning Approach of Collaborative Filtering to Recommender System with Opinion Mining
\nTo produce good quality recommendations for large or enterprise scale problems, a competent approach for recommender system is required. This paper presents such an approach which first generates the text score based on users’ reviews with the help of opinion mining. It then feeds ratings corresponding to the text scores to Convolutional Neural Network (CNN). CNN learns and does the dot product of user and product matrices. It is a special kind of feed forward neural network of deep learning technique to get better predictions in a product recommender system. The work done in this paper has improved accuracy and user satisfaction to great extent using CNN. It also helps e-commerce companies to increase the revenue by recommending closest products to users.
\nDeep Learning Technique for selecting appropriate Beauty Care Products for different skin Type
\nThis research has been undertaken to ease the challenging task of the beauty industry by using Deep Learning Method. Nowadays, the cosmetic product plays a major role in the appearance of personality. Customers are given a number of items with online shopping and e-commerce websites. It's difficult for us to pick the best product for our skin. Over recent years recommendation systems have been commonly used for providing user recommendations in various commercial platforms. The sparsity of the data and the scalability of the method however limit the performance of the algorithms used for recommendation and it is difficult to further improve the quality of the results of the recommendations. Hence we propose a predictive system that provides a precise idea of which product is best for our skin type using the Deep Learning Technique. The suggestion is based on the types of skin that might be Normal, Combination, Dry, Oily, and Sensitive. We have implemented the Deep Neural Network (DNN) model for cosmetic product composition. Finally, it is validated, by comparing with other recommendation algorithms on our generated dataset, that our model can effectively boost the recommendation performance. IndexTerms – Deep Learning, Deep Neural Network, Cosmetic Product Suggestion.
\n\n\nTL;DR:(1) deep learning model is used to extract the product characteristics from sparse inputs, (2) traditional collaborative filtering algorithm is updated with the characteristics obtained by deep learning model for sparse rated inputs, which greatly improves the user and product transaction.
\n
The exponential increase in the volume of online data has generated a confront of overburden of data for online users, which slow down the suitable access to products of pursuit on the Web. This contributed to the need for recommendation systems. Recommender system is a special form of intelligent technique that takes advantage of past user transactions on products to give recommendations of products. Collaborative filtering has turn out to be the commonly adopted method of providing users with customized services, except that it endures the problem of sparsely rated inputs. For collaborative filtering, we introduce a deep learning-based architecture which evaluates a discrete factorisation of vectors from sparse inputs. The characteristics of the products are retrieved using a deep learning model, denoising auto encoders. The traditional collaborative filtering algorithm that predicts and uses the past history of consumer interest and product characteristics are updated with the characteristics obtained by deep learning model for sparse rated inputs. The results of sparse data problem tested on MovieLens data set will greatly enhance the user and product transaction.
\nDerin Öğrenme ile Kıyafet Kombin Önerim Sistemi Cloth Combine Estimation System Using Deep Learning
\n\n\nTL;DR:The goal of this study was to develop a deep neural network (DNN) based algorithm for combining clothes. The dataset used in this study was taken from an international cloth company’s e-commerce website. The algorithm was trained on a dataset of 2,521 cloth combinations. The results of the study showed that the DNN based algorithm was able to successfully combine different clothes together.
\n
The clothing companies are paying more attention on meeting their customers on digital platforms and perform their marketing over the cloth combines which are created by fashion designers. Moreover, the companies on e- commerce platform use cloth recommendation system to draw their customers attention and perform sales. The customers browsing the e-commerce sites are facing some cloth recommendations based on the statistics such as “those looking at this product also looked at this product”. However, beside of statistical recommendations, more intelligence is required to recommend the cloth which combines the cloth that customers own or going to own. In this study, a cloth combine completion system is developed for e-commerce. The system is based deep neural network (DNN) and it is intended to imitate the fashion designer of the brands on digital market. The study is performed over the cloth combines shown on the e-commerce site of an international cloth company. In contrast to applications based on image processing, the clothes are parameterized depending on expert knowledge and a deep neural network is trained to find the best bottom clothing to the selected top clothing. Using the trained network, the brands can create user specific cloth combines as if the combines are performed by fashion-designer.
\nComplementary Outfit Recommendation Using Deep Learning
\n\n\nTL;DR:To recommend complementary outfits to users, we used deep learning methods such as Convolutional Neural Networks and Convolutional Autoencoders. Our system was able to recommend similar apparel items to the user’s input from their wardrobe, and improve the recommendations of e-commerce platforms.
\n
Abstract: Developing a system that provides a complementary outfit based on user queries can be challenging due to its complexity and subjectivity. At present, the methodologies available are used to recommend outfits to the users according to meta-data and user’s location. So we present an empirical study on the application where we make use of Deep Convolutional Neural Networks (DCNN) to the task of classifying apparel and user’s input from their wardrobe along with Convolutional Autoencoder which is capable of finding similar product images with the aim to remove the usage of product tags and improve the recommendations of e-commerce platforms. Choosing the right pair of clothes according to the customer’s choice and preference is of utmost importance. Sometimes people find it difficult to keep up with the trend so in such cases this project will be very helpful for them. Also, this helps the user to save their time and energy which implies that the user need not spend much time worrying about “what goes well with what”. Hence we have come up with a system that is capable of resolving such difficulties and whose primary purpose is to make the user’s work easier and beneficial.
\nRise of Deep Learning for Genomic, Proteomic, and Metabolomic Data Integration in Precision Medicine
\n\n\nTL;DR:Machine learning (ML), deep learning (DL), and artificial neural networks (ANNs) are being ubiquitously incorporated into everyday products such as Internet search, email spam filters, product recommendations, image classification, and speech recognition. Many approaches incorporate complex artificial neural network architectures and are collectively referred to as DL applications. These methods have been shown capable of representing and learning relationships in data in many diverse forms. Omics data pose considerable challenges for DL because of low signal to noise, analytical variance, and complex data integration requirements. However, DL models have already been shown capable of improving the ease of data encoding and predictive model performance over
\n
Abstract Machine learning (ML) is being ubiquitously incorporated into everyday products such as Internet search, email spam filters, product recommendations, image classification, and speech recognition. New approaches for highly integrated manufacturing and automation such as the Industry 4.0 and the Internet of things are also converging with ML methodologies. Many approaches incorporate complex artificial neural network architectures and are collectively referred to as deep learning (DL) applications. These methods have been shown capable of representing and learning predictable relationships in many diverse forms of data and hold promise for transforming the future of omics research and applications in precision medicine. Omics and electronic health record data pose considerable challenges for DL. This is due to many factors such as low signal to noise, analytical variance, and complex data integration requirements. However, DL models have already been shown capable of both improving the ease of data encoding and predictive model performance over alternative approaches. It may not be surprising that concepts encountered in DL share similarities with those observed in biological message relay systems such as gene, protein, and metabolite networks. This expert review examines the challenges and opportunities for DL at a systems and biological scale for a precision medicine readership.
\nProduct Matching Lessons and Recommendations from a Real World Application
\nRetailers rely heavily on product matching to better serve their customers, and to improve their modelling and forecasting. Product matching refers to the process of identifying similar or identical products across different data sources. This is a challenging problem as standardized unique identifiers are not used consistently across retailers, and product descriptions and characteristics vary across data collections. In this paper we present and discuss lessons learned from product matching in a real world application. We propose an evaluation framework where we investigate and compare the use of traditional machine learning methods and deep learning methods on public and proprietary datasets for product matching. Our findings show that traditional machine learning methods perform well on this task and a practitioner should investigate these methods first.
\nModeling Consumer Buying Decision for Recommendation Based on Multi-Task Deep Learning
\nAlthough marketing researchers and sociologists have recognized the importance of buying decision process and its significant influence on consumer's purchasing behaviors, existing recommender systems do not explicitly model the consumer buying decision process or capture the sequential regularities of what happens before and after each purchase. In this paper, we try to bridge the gap and improve recommendation systems by explicitly modeling consumer buying decision process and corresponding stages. In particular, we propose a multi-task learning model with long short-term memory networks (LSTM) to learn consumer buying decision process. It maps items, users, product categories, and the behavior sequences into real valued vectors, with which the probability of purchasing a product can be estimated. In this way, the model can capture user intentions and preferences, predicts the conversion rate of each candidate product, and makes recommendations accordingly. Experiments on real world data demonstrate the effectiveness of the proposed approach.
\nExtraction of Visual Features for Recommendation of Products via Deep Learning
\nIn this paper (The first author is the 1st place winner of the Open HSE Student Research Paper Competition (NIRS) in 2017, Computer Science nomination, with the topic “Extraction of Visual Features for Recommendation of Products”, as alumni of 2017 “Data Science” master program at Computer Science Faculty, HSE, Moscow), we describe a special recommender approach based on features extracted from the clothes’ images. The method of feature extraction relies on pre-trained deep neural network that follows transfer learning on the dataset. Recommendations are generated by the neural network as well. All the experiments are based on the items of category Clothing, Shoes and Jewelry from Amazon product dataset. It is demonstrated that the proposed approach outperforms the baseline collaborative filtering method.
\nLearning An End-to-End Structure for Retrieval in Large-Scale Recommendations
\nOne of the core problems in large-scale recommendations is to retrieve top relevant candidates accurately and efficiently, preferably in sub-linear time. Previous approaches are mostly based on a two-step procedure: first learn an inner-product model, and then use some approximate nearest neighbor (ANN) search algorithm to find top candidates. In this paper, we present Deep Retrieval (DR), to learn a retrievable structure directly with user-item interaction data (e.g. clicks) without resorting to the Euclidean space assumption in ANN algorithms. DR's structure encodes all candidate items into a discrete latent space. Those latent codes for the candidates are model parameters and learnt together with other neural network parameters to maximize the same objective function. With the model learnt, a beam search over the structure is performed to retrieve the top candidates for reranking. Empirically, we first demonstrate that DR, with sub-linear computational complexity, can achieve almost the same accuracy as the brute-force baseline on two public datasets. Moreover, we show that, in a live production recommendation system, a deployed DR approach significantly outperforms a well-tuned ANN baseline in terms of engagement metrics. To the best of our knowledge, DR is among the first non-ANN algorithms successfully deployed at the scale of hundreds of millions of items for industrial recommendation systems.
\nComparative study on traditional recommender systems and deep learning based recommender systems
\n\n\nTL;DR:Recommender systems is a big breakthrough for the field of e-commerce. Product recommendation is a challenging task to e-commerce companies. The traditional recommender systems provide the solutions in recommending the products. This in turn helps companies to generate good revenue. Nowadays, Deep Learning is being used in every domain. Deep Learning techniques in the field of recommender systems can be directly applied. Deep Learning has ample number of algorithms which can be used to give recommendations to users to purchase products. In this paper, the performance of traditional recommender systems and deep learning-based recommender systems are compared
\n
Recommender systems is a big breakthrough for the field of e-commerce. Product recommendation is challenging task to e-commerce companies. Traditional Recommender Systems provided the solutions in recommending the products. This in turn help companies to generate good revenue. Now a day Deep Learning is using in every domain. Deep Learning techniques in the field of Recommender Systems can be directly applied. Deep Learning has ample number of algorithms. These algorithms can be used to give recommendations to users to purchase products. In this paper performance of Traditional Recommender Systems and Deep Learning-based Recommender Systems are compared.
\nFrom Web to Physical and Back: WP User Profiling with Deep Learning
\n\n\nTL;DR:we use supervised machine learning on a web crawler to extract features from web pages (e.g. title, text, links) and then we use these features to predict the pavilion a person is most likely to visit at a fair.
\n
This position paper discusses the definition and implementation of Web-Physical (WP) user profiles, which allow the creation of personalized recommendations and innovative behavioral predictions in particular scenarios, i.e., fairs. The nature of a WP profile builds upon two different worlds: the Web (social networks and web applications) and the Physical one, each one of them being explored through (big) data collection platforms. These two platforms collect radically different information: on the one hand, information of appreciation towards a particular product or service (web domain) together with other metadata; on the other, the leases (x, y) of users in the exhibition space (physical domain). In this scenario, our research idea consists in identifying how the information in the two domains can be merged in a whole entity under a theoretical point of view: this will unleash tangible repercussions in terms of personalized recommendations and effective behavioral predictions, where with personalized recommendation we mean a suggestion to a user in physical terms (eg a pavilion to visit) and / or in web terms (eg a site to visit) and with behavioral prediction a prediction of where a user can go in the future, even in a multimedia perspective (physical + web).
\nBoosting Recommender Systems with Deep Learning
\n\n\nTL;DR:(1) using deep learning methods to extract visual features from images of products, (2) using deep learning to learn relationships between products that are stylistically complementary, and (3) using deep learning for other purposes across the Farfetch platform.
\n
Farfetch is a global fashion marketplace with a catalog that, at any time, has over 200 000 products spanning over 2000 brands from luxury boutiques all around the world. Finding the right product to the right customer is a challenge that, we, as Data Scientists working on the Recommendations team, are trying to solve using state-of-the art algorithms and disruptive technologies. Deep learning (DL) is an area of Machine Learning that has recently been brought to the spotlight for its breakthrough results across several domains. In this talk, we will provide an overview of some ongoing projects in which Deep Learning methods play a major role. A common problem in online marketplaces with large catalogs such as ours is the lack of detailed metadata about the products. Particularly, features such as style, colors, pattern, occasion, sizing, etc., are known to drive customers intent but are hard to catalog manually, in a consistent way. We explain how we use our extensive dataset of normalized product images together with state of the art convolutional neural networks to extract visual features. These can then be used to provide better recommendations and improve other applications across the platform. Another application of DL is to capture relationships between products which are stylistically complementary. We leverage data from thousands of hand-curated outfits to model these intangible fashion concepts only a human specialist can provide, and generalize a method of recommending complementary products using deep siamese neural networks. This an alternative recommendation strategy that can be used to drive cross-sell opportunities. These are merely a sample of problems we are tackling using DL at Farfetch. We believe that there are plenty of opportunities for application of these techniques to recommender systems and we look forward to discussing the potentials of this stream of research with the RecSys community.
\nA Deep Dive Into Understanding TheRandom Walk-Based Temporal Graph Learning
\n\n\nTL;DR:We study a more scalable graph learning algorithm based on random walks, which operates on dynamic input graphs and has attracted less attention in the architecture community compared to GCN. We propose high-performance CPU and GPU implementations of two important graph learning tasks, that cover a broad class of applications, using random walks on continuous-time dynamic graphs: link prediction and node classification.
\n
Machine learning on graph data has gained significant interest because of its applicability to various domains ranging from product recommendations to drug discovery. While there is a rapid growth in the algorithmic community, the computer architecture community has so far focused on a subset of graph learning algorithms including Graph Convolution Network (GCN), and a few others. In this paper, we study another, more scalable, graph learning algorithm based on random walks, which operates on dynamic input graphs and has attracted less attention in the architecture community compared to GCN. We propose high-performance CPU and GPU implementations of two important graph learning tasks, that cover a broad class of applications, using random walks on continuous-time dynamic graphs: link prediction and node classification. We show that the resulting workload exhibits distinct characteristics, measured in terms of irregularity, core and memory utilization, and cache hit rates, compared to graph traversals, deep learning, and GCN. We further conduct an in-depth performance analysis focused on both algorithm and hardware to guide future software optimization and architecture exploration. The algorithm-focused study presents a rich trade-off space between algorithmic performance and runtime complexity to identify optimization opportunities. We find an optimal hyperparameter setting that strikes balance in this trade-off space. Using this setting, we also perform a detailed microarchitectural characterization to analyze hardware behavior of these applications and uncover execution bottlenecks, which include high cache misses and dependency-related stalls. The outcome of our study includes recommendations for further performance optimization, and open-source implementations for future investigation.
\nA Deep Dive Into Understanding The Random Walk-Based Temporal Graph Learning
\n\n\nTL;DR:We study a scalable graph learning algorithm based on random walks and propose high-performance CPU and GPU implementations of two importantgraph learning tasks on dynamic input graphs. We show that the resulting workload exhibits distinct characteristics, measured in terms of irregularity, core and memory utilization, and cache hit rates. Our microarchitectural characterization identifies execution bottlenecks, which include high cache misses and dependency-related stalls.
\n
Machine learning on graph data has gained significant interest because of its applicability to various domains ranging from product recommendations to drug discovery. While there is a rapid growth in the algorithmic community, the computer architecture community has so far focused on a subset of graph learning algorithms including Graph Convolution Network (GCN), and a few others. In this paper, we study another, more scalable, graph learning algorithm based on random walks, which operates on dynamic input graphs and has attracted less attention in the architecture community compared to GCN. We propose high-performance CPU and GPU implementations of two important graph learning tasks, that cover a broad class of applications, using random walks on continuous-time dynamic graphs: link prediction and node classification. We show that the resulting workload exhibits distinct characteristics, measured in terms of irregularity, core and memory utilization, and cache hit rates, compared to graph traversals, deep learning, and GCN. We further conduct an in-depth performance analysis focused on both algorithm and hardware to guide future software optimization and architecture exploration. The algorithm-focused study presents a rich trade-off space between algorithmic performance and runtime complexity to identify optimization opportunities. We find an optimal hyperparameter setting that strikes balance in this trade-off space. Using this setting, we also perform a detailed microarchitectural characterization to analyze hardware behavior of these applications and uncover execution bottlenecks, which include high cache misses and dependency-related stalls. The outcome of our study includes recommendations for further performance optimization, and open-source implementations for future investigation.
\nReinforcement Learning-based Product Delivery Frequency Control
\nFrequency control is an important problem in modern recommender systems. It dictates the delivery frequency of recommendations to maintain product quality and efficiency. For example, the frequency of delivering promotional notifications impacts daily metrics as well as the infrastructure resource consumption (e.g. CPU and memory usage). There remain open questions on what objective we should optimize to represent business values in the long term best, and how we should balance between daily metrics and resource consumption in a dynamically fluctuating environment. We propose a personalized methodology for the frequency control problem, which combines long-term value optimization using reinforcement learning (RL) with a robust volume control technique we termed “Effective Factor”. We demonstrate statistically significant improvement in daily metrics and resource efficiency by our method in several notification applications at a scale of billions of users. To our best knowledge, our study represents the first deep RL application on the frequency control problem at such an industrial scale.
\nLearning Graph Neural Networks with Deep Graph Library
\n\n\nTL;DR:This tutorial provides an overview of graph neural networks (GNNs), discussing the types of problems that GNNs are well suited for, and introducing some of the most widely used GNN model architectures and problems/applications that are designed to solve. It also introduces the Deep Graph Library (DGL), a new software framework that simplifies the development of efficient GNN-based training and inference programs.
\n
Learning from graph and relational data plays a major role in many applications including social network analysis, marketing, e-commerce, information retrieval, knowledge modeling, medical and biological sciences, engineering, and others. In the last few years, Graph Neural Networks (GNNs) have emerged as a promising new supervised learning framework capable of bringing the power of deep representation learning to graph and relational data. This ever-growing body of research has shown that GNNs achieve state-of-the-art performance for problems such as link prediction, fraud detection, target-ligand binding activity prediction, knowledge-graph completion, and product recommendations. The objective of this tutorial is twofold. First, it will provide an overview of the theory behind GNNs, discuss the types of problems that GNNs are well suited for, and introduce some of the most widely used GNN model architectures and problems/applications that are designed to solve. Second, it will introduce the Deep Graph Library (DGL), a new software framework that simplifies the development of efficient GNN-based training and inference programs. To make things concrete, the tutorial will provide hands-on sessions using DGL. This hands-on part will cover both basic graph applications (e.g., node classification and link prediction), as well as more advanced topics including training GNNs on large graphs and in a distributed setting. In addition, it will provide hands-on tutorials on using GNNs and DGL for real-world applications such as recommendation and fraud detection.
\nA Meta-Learning Perspective on Cold-Start Recommendations for Items
\n\n\nTL;DR:We propose two deep neural network architectures that implement our meta-learning strategy. The first architecture learns a linear classifier whose weights are determined by the item history while the second architecture learns a neural network whose biases are instead adjusted.
\n
Matrix factorization (MF) is one of the most popular techniques for product recommendation, but is known to suffer from serious cold-start problems. Item cold-start problems are particularly acute in settings such as Tweet recommendation where new items arrive continuously. In this paper, we present a meta-learning strategy to address item cold-start when new items arrive continuously. We propose two deep neural network architectures that implement our meta-learning strategy. The first architecture learns a linear classifier whose weights are determined by the item history while the second architecture learns a neural network whose biases are instead adjusted. We evaluate our techniques on the real-world problem of Tweet recommendation. On production data at Twitter, we demonstrate that our proposed techniques significantly beat the MF baseline and also outperform production models for Tweet recommendation.
\nMotivation Factors Analysis and Policy Research of Deep Learning
\nThis article uses Interpretative Structural Modeling Method(ISM) to build deep learning promotion structure diagram of the influence factors, then uses Analytic Hierarchy Process(AHP) to determine the relative importance of various factors, according to the evaluation results it is concluded that the policy guidelines for deep learning to promote the influence degree of the relative maximum conclusion, put forward to promote deep learning better and faster, the government related department should publish relevant policy recommendations, embodied in more research funds, set up special introduction and training of research institutions and researchers, to perfect the theory system. At the same time, the improvement of social awareness will attract more high-tech companies in product research and development, making deep learning applied in more fields.
\nThanks for sharing😁
", "rawMarkdown": "Thanks for sharing😁", "votes": null } ], "comments": [ { "id": 1770159, "author_name": "", "author_url": "", "post_date": "04/28/2022 03:44:45", "content": "Thanks for sharing😁
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1768988": "Today we're going to talk about the boring topic of product recommendations and the groundbreaking research that's been happening in the academic world.\nit would have been better if we've got papers on how to make a perfect pancake such as:\n\n1) \"The Role of Pancakes in Modern Society\"\n2) \"Pancakes: A Comprehensive Review\"\n3) \"The Efficacy of Various Types of Pancakes\"\n4) \"How to Make a Perfect Pancake: A Comprehensive Guide\"\n\nBut this is not what the competition is about. unfortunately.\nso, without further ado, here are some of the most renowned and respected papers about product recommendations that have been released in recent months:\n\nEnjoy <3\n\n_____\n**[Deep Learning-based Online Alternative Product Recommendations at Scale](https://www.semanticscholar.org/paper/7004638b60301a2b7d5313f4fcb920f4932a2986)**\n\nAlternative recommender systems are critical for ecommerce companies. They guide customers to explore a massive product catalog and assist customers to find the right products among an overwhelming number of options. However, it is a non-trivial task to recommend alternative products that fit customers’ needs. In this paper, we use both textual product information (e.g. product titles and descriptions) and customer behavior data to recommend alternative products. Our results show that the coverage of alternative products is significantly improved in offline evaluations as well as recall and precision. The final A/B test shows that our algorithm increases the conversion rate by 12% in a statistically significant way. In order to better capture the semantic meaning of product information, we build a Siamese Network with Bidirectional LSTM to learn product embeddings. In order to learn a similarity space that better matches the preference of real customers, we use co-compared data from historical customer behavior as labels to train the network. In addition, we use NMSLIB to accelerate the computationally expensive kNN computation for millions of products so that the alternative recommendation is able to scale across the entire catalog of a major ecommerce site.\n_____\n\n_____\n**[Development of Fashion Product Retrieval and Recommendations Model Based on Deep Learning](https://doi.org/10.3390/electronics9030508)**\n\nThe digitization of the fashion industry diversified consumer segments, and consumers now have broader choices with shorter production cycles; digital technology in the fashion industry is attracting the attention of consumers. Therefore, a system that efficiently supports the searching and recommendation of a product is becoming increasingly important. However, the text-based search method has limitations because of the nature of the fashion industry, in which design is a very important factor. Therefore, we developed an intelligent fashion technique based on deep learning for efficient fashion product searches and recommendations consisting of a Sketch-Product fashion retrieval model and vector-based user preference fashion recommendation model. It was found that the “Precision at 5” of the image-based similar product retrieval model was 0.774 and that of the sketch-based similar product retrieval model was 0.445. The vector-based preference fashion recommendation model also showed positive performance. This system is expected to enhance consumers’ satisfaction by supporting users in more effectively searching for fashion products or by recommending fashion products before they begin a search.\n_____\n\n_____\n**[Using Deep Learning to Win the Booking.com WSDM WebTour21 Challenge on Sequential Recommendations](https://scholar.google.com/scholar?q=Using%20Deep%20Learning%20to%20Win%20the%20Booking.com%20WSDM%20WebTour21%20Challenge%20on%20Sequential%20Recommendations)**\n\n\n> **TL;DR:**- We designed three deep learning architectures based on Multilayer Perceptron (MLP), Gated Recurrent Units (GRU) and XLNet (Transformer) building blocks. - A shared component among the architectures was a Session-based Matrix Factorization head (SMF), which learns a linear mapping between the item (city) embeddings and the session (trip) embeddings to generate recommendations by a dot product operation.\n\nIn this paper we present our 1st place solution of the WSDM WebTour 21 Challenge. The competition task was to predict the city of the last booked hotel in a user trip, based on the previously visited cities. For our final solution, we designed three deep learning architectures based on Multilayer Perceptron (MLP), Gated Recurrent Units (GRU) and XLNet (Transformer) building blocks. A shared component among the architectures was a Session-based Matrix Factorization head (SMF), which learns a linear mapping between the item (city) embeddings and the session (trip) embeddings to generate recommendations by a dot product operation. Our final leaderboard result was 0.5939 for precision@4 and scored 2.8% better than the second solution. We published our implementation, using RAPIDS cuDF, TensorFlow and PyTorch, on github.com1.\n_____\n\n_____\n**[Deep Learning Sentiment Analysis For Recommendations In Social Applications](https://scholar.google.com/scholar?q=Deep%20Learning%20Sentiment%20Analysis%20For%20Recommendations%20In%20Social%20Applications)**\n\nSentiment analysis is a technique for identification of expression, mind-set, or feelings of users and classifies as negative, positive, favorable, unfavorable, etc. from a piece of text in the document. In recent days, deep learning materialized as a valuable way of resolving the sentiment classification problems. The representations are automatically learned in the neural network environment devoid of human labor. But the large-scale data to be trained decides the success of deep learning. In the internet lot of data in the form of comments, opinions are generated by the various websites which gather data about household products of multiple vendors, an incidence that happens in the society in day to day life, etc. Thereby the network users sentiments are derived from their access styles over the web that produces a considerable influence on the other persons who make use of the network to know about a product or political status. Several online sites have grown, and users need suggestions for faster access to the required products with good quality, and hence, recommendation techniques are widely used in various social applications.\n_____\n\n_____\n**[Detecting Product Adoption Intentions via Multiview Deep Learning](https://doi.org/10.1287/ijoc.2021.1083)**\n\n\n> **TL;DR:**Detecting product adoption intentions on social media could yield significant value in a wide range of applications, such as personalized recommendations and targeted marketing. Existing studies primarily focus on using one type of text representation in target social media posts, ignoring the major yet unexplored potential of fusing different text representations. In this paper, we propose a novel and general multiview deep learning model that simultaneously taps into the capability of multiview learning in leveraging different representations and deep learning in learning latent data representations using a flexible nonlinear transformation. The proposed model leverages three different text representations from a multiview perspective and takes advantage of local and\n\nDetecting product adoption intentions on social media could yield significant value in a wide range of applications, such as personalized recommendations and targeted marketing. In the literature, no study has explored the detection of product adoption intentions on social media, and only a few relevant studies have focused on purchase intention detection for products in one or several categories. Focusing on a product category rather than a specific product is too coarse-grained for precise advertising. Additionally, existing studies primarily focus on using one type of text representation in target social media posts, ignoring the major yet unexplored potential of fusing different text representations. In this paper, we first formulate the problem of product adoption intention mining and demonstrate the necessity of studying this problem and its practical value. To detect a product adoption intention for an individual product, we propose a novel and general multiview deep learning model that simultaneously taps into the capability of multiview learning in leveraging different representations and deep learning in learning latent data representations using a flexible nonlinear transformation. Specifically, the proposed model leverages three different text representations from a multiview perspective and takes advantage of local and long-term word relations by integrating convolutional neural network (CNN) and long short-term memory (LSTM) modules. Extensive experiments on three Twitter datasets demonstrate the effectiveness of the proposed multiview deep learning model compared with the existing benchmark methods. This study also significantly contributes research insights to the literature about intention mining and provides business value to relevant stakeholders such as product providers.\n_____\n\n_____\n**[Image Based Fashion Product Recommendation with Deep Learning](https://www.semanticscholar.org/paper/02dcec0f36f7b8782b755414f2d9f677576239c6)**\n\nWe develop a two-stage deep learning framework that recommends fashion images based on other input images of similar style. For that purpose, a neural network classifier is used as a data-driven, visually-aware feature extractor. The latter then serves as input for similarity-based recommendations using a ranking algorithm. Our approach is tested on the publicly available Fashion dataset. Initialization strategies using transfer learning from larger product databases are presented. Combined with more traditional content-based recommendation systems, our framework can help to increase robustness and performance, for example, by better matching a particular customer style.\n_____\n\n_____\n**[Multi-level Deep Learning based e-Commerce Product Categorization](https://scholar.google.com/scholar?q=Multi-level%20Deep%20Learning%20based%20e-Commerce%20Product%20Categorization)**\n\n\n> **TL;DR:**Our proposed classification model is based on a multi-level and multi-class deep learning tree. This model constructs multiple models based on single-label and multi-level label predictions as well as the characteristics of the product tree structure. These models are then combined to generate a new classification model. The proposed classification model is tested on the online test dataset, and the accuracy, recall, and F1 score are 0.8552, 0.8389, 0.8404 in Stage 1 and 0.8397, 0.8428, 0.8379 in Stage 2 respectively\n\nE-commerce product categorization is an important topic, and its quality directly affects subsequent search, recommendations and related personalized services. E-commerce product classification is challenging due to the large scale and complexity of the product information and categories. In the E-Commerce Text Classification Challenge, we combine machine learning, deep learning, and natural language processing to propose a multi-level and multi-class deep learning tree method. Our method constructs multiple models based on single-label and multi-level label predictions as well as the characteristics of the product tree structure and combines the multiple models to generate a new classification model. The proposed classification model is tested on the online test dataset. The accuracy, recall, and F1 score are 0.8552, 0.8389, 0.8404 in leaderboard(Stage 1) and 0.8397, 0.8428, 0.8379 in leaderboard(Stage 2) respectively, ranking among top 3 scorers.\n_____\n\n_____\n**[Deep Retrieval: Learning A Retrievable Structure for Large-Scale Recommendations](https://paperswithcode.com/paper/deep-retrieval-an-end-to-end-learnable)**\n\nOne of the core problems in large-scale recommendations is to retrieve top relevant candidates accurately and efficiently, preferably in sub-linear time. Previous approaches are mostly based on a two-step procedure: first learn an inner-product model, and then use some approximate nearest neighbor (ANN) search algorithm to find top candidates. In this paper, we present Deep Retrieval (DR), to learn a retrievable structure directly with user-item interaction data (e.g. clicks) without resorting to the Euclidean space assumption in ANN algorithms. DR's structure encodes all candidate items into a discrete latent space. Those latent codes for the candidates are model parameters and learnt together with other neural network parameters to maximize the same objective function. With the model learnt, a beam search over the structure is performed to retrieve the top candidates for reranking. Empirically, we first demonstrate that DR, with sub-linear computational complexity, can achieve almost the same accuracy as the brute-force baseline on two public datasets. Moreover, we show that, in a live production recommendation system, a deployed DR approach significantly outperforms a well-tuned ANN baseline in terms of engagement metrics. To the best of our knowledge, DR is among the first non-ANN algorithms successfully deployed at the scale of hundreds of millions of items for industrial recommendation systems.\n_____\n\n_____\n**[Deep Learning Based Real Age and Gender Estimation from Unconstrained Face Image towards Smart Store Customer Relationship Management](https://doi.org/10.3390/APP11104549)**\n\nThe COVID-19 pandemic markedly changed the human shopping nature, necessitating a contactless shopping system to curb the spread of the contagious disease efficiently. Consequently, a customer opts for a store where it is possible to avoid physical contacts and shorten the shopping process with extended services such as personalized product recommendations. Automatic age and gender estimation of a customer in a smart store strongly benefit the consumer by providing personalized advertisement and product recommendation; similarly, it aids the smart store proprietor to promote sales and develop an inventory perpetually for the future retail. In our paper, we propose a deep learning-founded enterprise solution for smart store customer relationship management (CRM), which allows us to predict the age and gender from a customer’s face image taken in an unconstrained environment to facilitate the smart store’s extended services, as it is expected for a modern venture. For the age estimation problem, we mitigate the data sparsity problem of the large public IMDB-WIKI dataset by image enhancement from another dataset and perform data augmentation as required. We handle our classification tasks utilizing an empirically leading pre-trained convolutional neural network (CNN), the VGG-16 network, and incorporate batch normalization. Especially, the age estimation task is posed as a deep classification problem followed by a multinomial logistic regression first-moment refinement. We validate our system for two standard benchmarks, one for each task, and demonstrate state-of-the-art performance for both real age and gender estimation.\n_____\n\n_____\n**[Deep Learning Based Matrix Factorization For Collaborative Filtering](https://doi.org/10.1109/ICCWAMTIP53232.2021.9674157)**\n\nCollaborative Filtering based on matrix factorization (MF) has shown tremendous success in the field recommender system. However, MF has difficulty in handling sparsity and scalability. These resulted in low quality of recommendations. In this regard, deep learning has shown immense success in different application areas including recommender systems. To address the limitations, we incorporate deep learning architecture to matrix factorization and develop a novel mode. The core idea of the method is to map users and items input vector to two well-structured deep neural network architectures separately for factorization. Then, we incorporate inner product to the output layers of the network to predict the rating scores. The use of this structure significantly improve the quality of recommendation. The experimental result on real data sets shows that our proposed model outperformed state of the art methods.\n_____\n\n_____\n**[Research on Commodity Recommendation System Based on Deep Learning](https://doi.org/10.1088/1742-6596/1865/4/042011)**\n\nThe traditional matrix factorization model cannot effectively extract the features of users and items, but the feature information can be extracted well based on the deep learning model. At present, the mainstream recommendation algorithms based on deep learning only make recommendation prediction in the form of the product of neural network output or item features and user features, and cannot fully mine the relationship between users and items. Based on this, this paper proposes a recommendation algorithm based on the combination of text convolution neural network and singular value decomposition (Bias SVD) with biased terms. The text convolution neural network (Text CNN) is used to fully extract the feature information of users and items, and then the singular value decomposition method is used to make recommendations to deeply understand the document context information and further improve the accuracy of recommendation. The algorithm is widely evaluated and analyzed on two real data sets of MovieLens, and the accuracy of recommendation is obviously better than that of ConvMF algorithm and mainstream deep learning recommendation algorithm.\n_____\n\n_____\n**[A Deep Learning Approach of Collaborative Filtering to Recommender System with Opinion Mining](https://doi.org/10.1007/978-981-15-6014-9_15)**\n\nTo produce good quality recommendations for large or enterprise scale problems, a competent approach for recommender system is required. This paper presents such an approach which first generates the text score based on users’ reviews with the help of opinion mining. It then feeds ratings corresponding to the text scores to Convolutional Neural Network (CNN). CNN learns and does the dot product of user and product matrices. It is a special kind of feed forward neural network of deep learning technique to get better predictions in a product recommender system. The work done in this paper has improved accuracy and user satisfaction to great extent using CNN. It also helps e-commerce companies to increase the revenue by recommending closest products to users.\n_____\n\n_____\n**[Deep Learning Technique for selecting appropriate Beauty Care Products for different skin Type](https://scholar.google.com/scholar?q=Deep%20Learning%20Technique%20for%20selecting%20appropriate%20Beauty%20Care%20Products%20for%20different%20skin%20Type)**\n\nThis research has been undertaken to ease the challenging task of the beauty industry by using Deep Learning Method. Nowadays, the cosmetic product plays a major role in the appearance of personality. Customers are given a number of items with online shopping and e-commerce websites. It's difficult for us to pick the best product for our skin. Over recent years recommendation systems have been commonly used for providing user recommendations in various commercial platforms. The sparsity of the data and the scalability of the method however limit the performance of the algorithms used for recommendation and it is difficult to further improve the quality of the results of the recommendations. Hence we propose a predictive system that provides a precise idea of which product is best for our skin type using the Deep Learning Technique. The suggestion is based on the types of skin that might be Normal, Combination, Dry, Oily, and Sensitive. We have implemented the Deep Neural Network (DNN) model for cosmetic product composition. Finally, it is validated, by comparing with other recommendation algorithms on our generated dataset, that our model can effectively boost the recommendation performance. IndexTerms – Deep Learning, Deep Neural Network, Cosmetic Product Suggestion.\n_____\n\n_____\n**[Boosting a Hybrid Model Recommendation System for Sparse Data using Collaborative Filtering and Deep Learning](https://scholar.google.com/scholar?q=Boosting%20a%20Hybrid%20Model%20Recommendation%20System%20for%20Sparse%20Data%20using%20Collaborative%20Filtering%20and%20Deep%20Learning)**\n\n\n> **TL;DR:**(1) deep learning model is used to extract the product characteristics from sparse inputs, (2) traditional collaborative filtering algorithm is updated with the characteristics obtained by deep learning model for sparse rated inputs, which greatly improves the user and product transaction.\n\nThe exponential increase in the volume of online data has generated a confront of overburden of data for online users, which slow down the suitable access to products of pursuit on the Web. This contributed to the need for recommendation systems. Recommender system is a special form of intelligent technique that takes advantage of past user transactions on products to give recommendations of products. Collaborative filtering has turn out to be the commonly adopted method of providing users with customized services, except that it endures the problem of sparsely rated inputs. For collaborative filtering, we introduce a deep learning-based architecture which evaluates a discrete factorisation of vectors from sparse inputs. The characteristics of the products are retrieved using a deep learning model, denoising auto encoders. The traditional collaborative filtering algorithm that predicts and uses the past history of consumer interest and product characteristics are updated with the characteristics obtained by deep learning model for sparse rated inputs. The results of sparse data problem tested on MovieLens data set will greatly enhance the user and product transaction.\n_____\n\n_____\n**[Derin Öğrenme ile Kıyafet Kombin Önerim Sistemi Cloth Combine Estimation System Using Deep Learning](https://doi.org/10.1109/ASYU50717.2020.9259804)**\n\n\n> **TL;DR:**The goal of this study was to develop a deep neural network (DNN) based algorithm for combining clothes. The dataset used in this study was taken from an international cloth company’s e-commerce website. The algorithm was trained on a dataset of 2,521 cloth combinations. The results of the study showed that the DNN based algorithm was able to successfully combine different clothes together.\n\nThe clothing companies are paying more attention on meeting their customers on digital platforms and perform their marketing over the cloth combines which are created by fashion designers. Moreover, the companies on e- commerce platform use cloth recommendation system to draw their customers attention and perform sales. The customers browsing the e-commerce sites are facing some cloth recommendations based on the statistics such as “those looking at this product also looked at this product”. However, beside of statistical recommendations, more intelligence is required to recommend the cloth which combines the cloth that customers own or going to own. In this study, a cloth combine completion system is developed for e-commerce. The system is based deep neural network (DNN) and it is intended to imitate the fashion designer of the brands on digital market. The study is performed over the cloth combines shown on the e-commerce site of an international cloth company. In contrast to applications based on image processing, the clothes are parameterized depending on expert knowledge and a deep neural network is trained to find the best bottom clothing to the selected top clothing. Using the trained network, the brands can create user specific cloth combines as if the combines are performed by fashion-designer.\n_____\n\n_____\n**[Complementary Outfit Recommendation Using Deep Learning](https://scholar.google.com/scholar?q=Complementary%20Outfit%20Recommendation%20Using%20Deep%20Learning)**\n\n\n> **TL;DR:**To recommend complementary outfits to users, we used deep learning methods such as Convolutional Neural Networks and Convolutional Autoencoders. Our system was able to recommend similar apparel items to the user’s input from their wardrobe, and improve the recommendations of e-commerce platforms.\n\nAbstract: Developing a system that provides a complementary outfit based on user queries can be challenging due to its complexity and subjectivity. At present, the methodologies available are used to recommend outfits to the users according to meta-data and user’s location. So we present an empirical study on the application where we make use of Deep Convolutional Neural Networks (DCNN) to the task of classifying apparel and user’s input from their wardrobe along with Convolutional Autoencoder which is capable of finding similar product images with the aim to remove the usage of product tags and improve the recommendations of e-commerce platforms. Choosing the right pair of clothes according to the customer’s choice and preference is of utmost importance. Sometimes people find it difficult to keep up with the trend so in such cases this project will be very helpful for them. Also, this helps the user to save their time and energy which implies that the user need not spend much time worrying about “what goes well with what”. Hence we have come up with a system that is capable of resolving such difficulties and whose primary purpose is to make the user’s work easier and beneficial.\n_____\n\n_____\n**[Rise of Deep Learning for Genomic, Proteomic, and Metabolomic Data Integration in Precision Medicine](https://doi.org/10.1089/omi.2018.0097)**\n\n\n> **TL;DR:**Machine learning (ML), deep learning (DL), and artificial neural networks (ANNs) are being ubiquitously incorporated into everyday products such as Internet search, email spam filters, product recommendations, image classification, and speech recognition. Many approaches incorporate complex artificial neural network architectures and are collectively referred to as DL applications. These methods have been shown capable of representing and learning relationships in data in many diverse forms. Omics data pose considerable challenges for DL because of low signal to noise, analytical variance, and complex data integration requirements. However, DL models have already been shown capable of improving the ease of data encoding and predictive model performance over\n\nAbstract Machine learning (ML) is being ubiquitously incorporated into everyday products such as Internet search, email spam filters, product recommendations, image classification, and speech recognition. New approaches for highly integrated manufacturing and automation such as the Industry 4.0 and the Internet of things are also converging with ML methodologies. Many approaches incorporate complex artificial neural network architectures and are collectively referred to as deep learning (DL) applications. These methods have been shown capable of representing and learning predictable relationships in many diverse forms of data and hold promise for transforming the future of omics research and applications in precision medicine. Omics and electronic health record data pose considerable challenges for DL. This is due to many factors such as low signal to noise, analytical variance, and complex data integration requirements. However, DL models have already been shown capable of both improving the ease of data encoding and predictive model performance over alternative approaches. It may not be surprising that concepts encountered in DL share similarities with those observed in biological message relay systems such as gene, protein, and metabolite networks. This expert review examines the challenges and opportunities for DL at a systems and biological scale for a precision medicine readership.\n_____\n\n_____\n**[Product Matching Lessons and Recommendations from a Real World Application](https://doi.org/10.21428/594757db.08c5079e)**\n\nRetailers rely heavily on product matching to better serve their customers, and to improve their modelling and forecasting. Product matching refers to the process of identifying similar or identical products across different data sources. This is a challenging problem as standardized unique identifiers are not used consistently across retailers, and product descriptions and characteristics vary across data collections. In this paper we present and discuss lessons learned from product matching in a real world application. We propose an evaluation framework where we investigate and compare the use of traditional machine learning methods and deep learning methods on public and proprietary datasets for product matching. Our findings show that traditional machine learning methods perform well on this task and a practitioner should investigate these methods first.\n_____\n\n_____\n**[Modeling Consumer Buying Decision for Recommendation Based on Multi-Task Deep Learning](https://doi.org/10.1145/3269206.3269285)**\n\nAlthough marketing researchers and sociologists have recognized the importance of buying decision process and its significant influence on consumer's purchasing behaviors, existing recommender systems do not explicitly model the consumer buying decision process or capture the sequential regularities of what happens before and after each purchase. In this paper, we try to bridge the gap and improve recommendation systems by explicitly modeling consumer buying decision process and corresponding stages. In particular, we propose a multi-task learning model with long short-term memory networks (LSTM) to learn consumer buying decision process. It maps items, users, product categories, and the behavior sequences into real valued vectors, with which the probability of purchasing a product can be estimated. In this way, the model can capture user intentions and preferences, predicts the conversion rate of each candidate product, and makes recommendations accordingly. Experiments on real world data demonstrate the effectiveness of the proposed approach.\n_____\n\n_____\n**[Extraction of Visual Features for Recommendation of Products via Deep Learning](https://doi.org/10.1007/978-3-030-11027-7_20)**\n\nIn this paper (The first author is the 1st place winner of the Open HSE Student Research Paper Competition (NIRS) in 2017, Computer Science nomination, with the topic “Extraction of Visual Features for Recommendation of Products”, as alumni of 2017 “Data Science” master program at Computer Science Faculty, HSE, Moscow), we describe a special recommender approach based on features extracted from the clothes’ images. The method of feature extraction relies on pre-trained deep neural network that follows transfer learning on the dataset. Recommendations are generated by the neural network as well. All the experiments are based on the items of category Clothing, Shoes and Jewelry from Amazon product dataset. It is demonstrated that the proposed approach outperforms the baseline collaborative filtering method.\n_____\n\n_____\n**[Learning An End-to-End Structure for Retrieval in Large-Scale Recommendations](https://doi.org/10.1145/3459637.3482362)**\n\nOne of the core problems in large-scale recommendations is to retrieve top relevant candidates accurately and efficiently, preferably in sub-linear time. Previous approaches are mostly based on a two-step procedure: first learn an inner-product model, and then use some approximate nearest neighbor (ANN) search algorithm to find top candidates. In this paper, we present Deep Retrieval (DR), to learn a retrievable structure directly with user-item interaction data (e.g. clicks) without resorting to the Euclidean space assumption in ANN algorithms. DR's structure encodes all candidate items into a discrete latent space. Those latent codes for the candidates are model parameters and learnt together with other neural network parameters to maximize the same objective function. With the model learnt, a beam search over the structure is performed to retrieve the top candidates for reranking. Empirically, we first demonstrate that DR, with sub-linear computational complexity, can achieve almost the same accuracy as the brute-force baseline on two public datasets. Moreover, we show that, in a live production recommendation system, a deployed DR approach significantly outperforms a well-tuned ANN baseline in terms of engagement metrics. To the best of our knowledge, DR is among the first non-ANN algorithms successfully deployed at the scale of hundreds of millions of items for industrial recommendation systems.\n_____\n\n_____\n**[Comparative study on traditional recommender systems and deep learning based recommender systems](https://doi.org/10.18280/AMA_B.610202)**\n\n\n> **TL;DR:**Recommender systems is a big breakthrough for the field of e-commerce. Product recommendation is a challenging task to e-commerce companies. The traditional recommender systems provide the solutions in recommending the products. This in turn helps companies to generate good revenue. Nowadays, Deep Learning is being used in every domain. Deep Learning techniques in the field of recommender systems can be directly applied. Deep Learning has ample number of algorithms which can be used to give recommendations to users to purchase products. In this paper, the performance of traditional recommender systems and deep learning-based recommender systems are compared\n\nRecommender systems is a big breakthrough for the field of e-commerce. Product recommendation is challenging task to e-commerce companies. Traditional Recommender Systems provided the solutions in recommending the products. This in turn help companies to generate good revenue. Now a day Deep Learning is using in every domain. Deep Learning techniques in the field of Recommender Systems can be directly applied. Deep Learning has ample number of algorithms. These algorithms can be used to give recommendations to users to purchase products. In this paper performance of Traditional Recommender Systems and Deep Learning-based Recommender Systems are compared.\n_____\n\n_____\n**[From Web to Physical and Back: WP User Profiling with Deep Learning](https://doi.org/10.1007/978-3-030-03056-8_11)**\n\n\n> **TL;DR:**we use supervised machine learning on a web crawler to extract features from web pages (e.g. title, text, links) and then we use these features to predict the pavilion a person is most likely to visit at a fair.\n\nThis position paper discusses the definition and implementation of Web-Physical (WP) user profiles, which allow the creation of personalized recommendations and innovative behavioral predictions in particular scenarios, i.e., fairs. The nature of a WP profile builds upon two different worlds: the Web (social networks and web applications) and the Physical one, each one of them being explored through (big) data collection platforms. These two platforms collect radically different information: on the one hand, information of appreciation towards a particular product or service (web domain) together with other metadata; on the other, the leases (x, y) of users in the exhibition space (physical domain). In this scenario, our research idea consists in identifying how the information in the two domains can be merged in a whole entity under a theoretical point of view: this will unleash tangible repercussions in terms of personalized recommendations and effective behavioral predictions, where with personalized recommendation we mean a suggestion to a user in physical terms (eg a pavilion to visit) and / or in web terms (eg a site to visit) and with behavioral prediction a prediction of where a user can go in the future, even in a multimedia perspective (physical + web).\n_____\n\n_____\n**[Boosting Recommender Systems with Deep Learning](https://doi.org/10.1145/3109859.3109926)**\n\n\n> **TL;DR:**(1) using deep learning methods to extract visual features from images of products, (2) using deep learning to learn relationships between products that are stylistically complementary, and (3) using deep learning for other purposes across the Farfetch platform.\n\nFarfetch is a global fashion marketplace with a catalog that, at any time, has over 200 000 products spanning over 2000 brands from luxury boutiques all around the world. Finding the right product to the right customer is a challenge that, we, as Data Scientists working on the Recommendations team, are trying to solve using state-of-the art algorithms and disruptive technologies. Deep learning (DL) is an area of Machine Learning that has recently been brought to the spotlight for its breakthrough results across several domains. In this talk, we will provide an overview of some ongoing projects in which Deep Learning methods play a major role. A common problem in online marketplaces with large catalogs such as ours is the lack of detailed metadata about the products. Particularly, features such as style, colors, pattern, occasion, sizing, etc., are known to drive customers intent but are hard to catalog manually, in a consistent way. We explain how we use our extensive dataset of normalized product images together with state of the art convolutional neural networks to extract visual features. These can then be used to provide better recommendations and improve other applications across the platform. Another application of DL is to capture relationships between products which are stylistically complementary. We leverage data from thousands of hand-curated outfits to model these intangible fashion concepts only a human specialist can provide, and generalize a method of recommending complementary products using deep siamese neural networks. This an alternative recommendation strategy that can be used to drive cross-sell opportunities. These are merely a sample of problems we are tackling using DL at Farfetch. We believe that there are plenty of opportunities for application of these techniques to recommender systems and we look forward to discussing the potentials of this stream of research with the RecSys community.\n_____\n\n_____\n**[A Deep Dive Into Understanding TheRandom Walk-Based Temporal Graph Learning](https://doi.org/10.5281/ZENODO.5555384)**\n\n\n> **TL;DR:**We study a more scalable graph learning algorithm based on random walks, which operates on dynamic input graphs and has attracted less attention in the architecture community compared to GCN. We propose high-performance CPU and GPU implementations of two important graph learning tasks, that cover a broad class of applications, using random walks on continuous-time dynamic graphs: link prediction and node classification.\n\nMachine learning on graph data has gained significant interest because of its applicability to various domains ranging from product recommendations to drug discovery. While there is a rapid growth in the algorithmic community, the computer architecture community has so far focused on a subset of graph learning algorithms including Graph Convolution Network (GCN), and a few others. In this paper, we study another, more scalable, graph learning algorithm based on random walks, which operates on dynamic input graphs and has attracted less attention in the architecture community compared to GCN. We propose high-performance CPU and GPU implementations of two important graph learning tasks, that cover a broad class of applications, using random walks on continuous-time dynamic graphs: link prediction and node classification. We show that the resulting workload exhibits distinct characteristics, measured in terms of irregularity, core and memory utilization, and cache hit rates, compared to graph traversals, deep learning, and GCN. We further conduct an in-depth performance analysis focused on both algorithm and hardware to guide future software optimization and architecture exploration. The algorithm-focused study presents a rich trade-off space between algorithmic performance and runtime complexity to identify optimization opportunities. We find an optimal hyperparameter setting that strikes balance in this trade-off space. Using this setting, we also perform a detailed microarchitectural characterization to analyze hardware behavior of these applications and uncover execution bottlenecks, which include high cache misses and dependency-related stalls. The outcome of our study includes recommendations for further performance optimization, and open-source implementations for future investigation.\n_____\n\n_____\n**[A Deep Dive Into Understanding The Random Walk-Based Temporal Graph Learning](https://doi.org/10.1109/IISWC53511.2021.00019)**\n\n\n> **TL;DR:**We study a scalable graph learning algorithm based on random walks and propose high-performance CPU and GPU implementations of two importantgraph learning tasks on dynamic input graphs. We show that the resulting workload exhibits distinct characteristics, measured in terms of irregularity, core and memory utilization, and cache hit rates. Our microarchitectural characterization identifies execution bottlenecks, which include high cache misses and dependency-related stalls.\n\nMachine learning on graph data has gained significant interest because of its applicability to various domains ranging from product recommendations to drug discovery. While there is a rapid growth in the algorithmic community, the computer architecture community has so far focused on a subset of graph learning algorithms including Graph Convolution Network (GCN), and a few others. In this paper, we study another, more scalable, graph learning algorithm based on random walks, which operates on dynamic input graphs and has attracted less attention in the architecture community compared to GCN. We propose high-performance CPU and GPU implementations of two important graph learning tasks, that cover a broad class of applications, using random walks on continuous-time dynamic graphs: link prediction and node classification. We show that the resulting workload exhibits distinct characteristics, measured in terms of irregularity, core and memory utilization, and cache hit rates, compared to graph traversals, deep learning, and GCN. We further conduct an in-depth performance analysis focused on both algorithm and hardware to guide future software optimization and architecture exploration. The algorithm-focused study presents a rich trade-off space between algorithmic performance and runtime complexity to identify optimization opportunities. We find an optimal hyperparameter setting that strikes balance in this trade-off space. Using this setting, we also perform a detailed microarchitectural characterization to analyze hardware behavior of these applications and uncover execution bottlenecks, which include high cache misses and dependency-related stalls. The outcome of our study includes recommendations for further performance optimization, and open-source implementations for future investigation.\n_____\n\n_____\n**[Reinforcement Learning-based Product Delivery Frequency Control](https://www.semanticscholar.org/paper/375806294739ef53917215ba5d487a20acaed774)**\n\nFrequency control is an important problem in modern recommender systems. It dictates the delivery frequency of recommendations to maintain product quality and efficiency. For example, the frequency of delivering promotional notifications impacts daily metrics as well as the infrastructure resource consumption (e.g. CPU and memory usage). There remain open questions on what objective we should optimize to represent business values in the long term best, and how we should balance between daily metrics and resource consumption in a dynamically fluctuating environment. We propose a personalized methodology for the frequency control problem, which combines long-term value optimization using reinforcement learning (RL) with a robust volume control technique we termed “Effective Factor”. We demonstrate statistically significant improvement in daily metrics and resource efficiency by our method in several notification applications at a scale of billions of users. To our best knowledge, our study represents the first deep RL application on the frequency control problem at such an industrial scale.\n_____\n\n_____\n**[Learning Graph Neural Networks with Deep Graph Library](https://doi.org/10.1145/3366424.3383111)**\n\n\n> **TL;DR:**This tutorial provides an overview of graph neural networks (GNNs), discussing the types of problems that GNNs are well suited for, and introducing some of the most widely used GNN model architectures and problems/applications that are designed to solve. It also introduces the Deep Graph Library (DGL), a new software framework that simplifies the development of efficient GNN-based training and inference programs.\n\nLearning from graph and relational data plays a major role in many applications including social network analysis, marketing, e-commerce, information retrieval, knowledge modeling, medical and biological sciences, engineering, and others. In the last few years, Graph Neural Networks (GNNs) have emerged as a promising new supervised learning framework capable of bringing the power of deep representation learning to graph and relational data. This ever-growing body of research has shown that GNNs achieve state-of-the-art performance for problems such as link prediction, fraud detection, target-ligand binding activity prediction, knowledge-graph completion, and product recommendations. The objective of this tutorial is twofold. First, it will provide an overview of the theory behind GNNs, discuss the types of problems that GNNs are well suited for, and introduce some of the most widely used GNN model architectures and problems/applications that are designed to solve. Second, it will introduce the Deep Graph Library (DGL), a new software framework that simplifies the development of efficient GNN-based training and inference programs. To make things concrete, the tutorial will provide hands-on sessions using DGL. This hands-on part will cover both basic graph applications (e.g., node classification and link prediction), as well as more advanced topics including training GNNs on large graphs and in a distributed setting. In addition, it will provide hands-on tutorials on using GNNs and DGL for real-world applications such as recommendation and fraud detection.\n_____\n\n_____\n**[A Meta-Learning Perspective on Cold-Start Recommendations for Items](https://scholar.google.com/scholar?q=A%20Meta-Learning%20Perspective%20on%20Cold-Start%20Recommendations%20for%20Items)**\n\n\n> **TL;DR:**We propose two deep neural network architectures that implement our meta-learning strategy. The first architecture learns a linear classifier whose weights are determined by the item history while the second architecture learns a neural network whose biases are instead adjusted.\n\nMatrix factorization (MF) is one of the most popular techniques for product recommendation, but is known to suffer from serious cold-start problems. Item cold-start problems are particularly acute in settings such as Tweet recommendation where new items arrive continuously. In this paper, we present a meta-learning strategy to address item cold-start when new items arrive continuously. We propose two deep neural network architectures that implement our meta-learning strategy. The first architecture learns a linear classifier whose weights are determined by the item history while the second architecture learns a neural network whose biases are instead adjusted. We evaluate our techniques on the real-world problem of Tweet recommendation. On production data at Twitter, we demonstrate that our proposed techniques significantly beat the MF baseline and also outperform production models for Tweet recommendation.\n_____\n\n_____\n**[Motivation Factors Analysis and Policy Research of Deep Learning](https://doi.org/10.2991/MMEBC-16.2016.255)**\n\nThis article uses Interpretative Structural Modeling Method(ISM) to build deep learning promotion structure diagram of the influence factors, then uses Analytic Hierarchy Process(AHP) to determine the relative importance of various factors, according to the evaluation results it is concluded that the policy guidelines for deep learning to promote the influence degree of the relative maximum conclusion, put forward to promote deep learning better and faster, the government related department should publish relevant policy recommendations, embodied in more research funds, set up special introduction and training of research institutions and researchers, to perfect the theory system. At the same time, the improvement of social awareness will attract more high-tech companies in product research and development, making deep learning applied in more fields.\n_____", "1770159": "Thanks for sharing😁" }, "source": "meta" }