\n
![\"Storefront](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4621388%2F73821bf5d1a5bc292459ad6ec531ac8c%2FScreenshot%202022-11-22%20at%2012-13-08%20OTTO%20-%20Mode%20Mbel%20%20Technik%20%20Zum%20Online-Shop.png?generation=1669115616204503&alt=media\")
\nThere, users can find a selection of products personalized in real-time that match their interests. We use user behavior-based recommendations for users with or without login alike, as they perform very well in both cases.
Hi,it may seem obvious to some but can I please ask: what is the use case for such a recommendation algorithm? Imagine we build it tomorrow, how would it be used?
\nI am assuming that we don't have data from logged-in users of the given website, therefore would this model be useful for website personalisation when it comes to anonymous users browsing based on their shopping behaviour during their session? Say I got to the given e-commerce website that uses this model - how would the model change my experience? It'd be great if you could outline a couple of its use-cases to get a better feel of the purpose it's supposed to serve.
\nThank you in advance!
", "messages": [ { "id": "2038235", "postDate": "11/21/2022 09:24:53", "content": "Hi,it may seem obvious to some but can I please ask: what is the use case for such a recommendation algorithm? Imagine we build it tomorrow, how would it be used?
\nI am assuming that we don't have data from logged-in users of the given website, therefore would this model be useful for website personalisation when it comes to anonymous users browsing based on their shopping behaviour during their session? Say I got to the given e-commerce website that uses this model - how would the model change my experience? It'd be great if you could outline a couple of its use-cases to get a better feel of the purpose it's supposed to serve.
\nThank you in advance!
", "rawMarkdown": "Hi,it may seem obvious to some but can I please ask: what is the use case for such a recommendation algorithm? Imagine we build it tomorrow, how would it be used?\n\nI am assuming that we don't have data from logged-in users of the given website, therefore would this model be useful for website personalisation when it comes to anonymous users browsing based on their shopping behaviour during their session? Say I got to the given e-commerce website that uses this model - how would the model change my experience? It'd be great if you could outline a couple of its use-cases to get a better feel of the purpose it's supposed to serve.\n\nThank you in advance!", "votes": null }, { "id": "2038295", "postDate": "11/21/2022 09:57:07", "content": "Hey @ikogias!
\nRecommender systems literally run the world 🙂 If you go to facebook, or Twitter, or YouTube (2nd biggest search engine in the world), or Amazon, or any even moderately big online store, you are bombarded with recommendations 🙂
\nWhy? Because they work. There are millions of dollars riding on incremental increases in the performance of recommender systems, be that directly in sales or ad revenue. Entire companies were built on recommender systems (many would attribute the rise of TikTok to their recommendation engine).
\nHere, you are 100% spot on. The website doesn't need to know anything about you, just the sequence of your actions can be an extremely valuable source of information. Better yet, as a company, you don't have to deal with a lot of the personally identifiable information overhead that comes up when storing certain user data (not that you are entirely off the hook in man jurisdictions, but the situation is likely much easier to handle from a legal and security perspective).
\nPlus, with session-based data you can serve the holy grail of recommender systems, that is real-time recommendations.
\nImagine that between your actions a website is able to display items that you are likely to buy… how much value would be in that? Most likely, enormous!
\nOf course, there is the separate, related question of what value companies get out of organizing Kaggle competitions 🙂 To what extent the solutions can translate to actual software put in production?
\nBut I think that the answer to this question is a resounding yes -- both Kagglers benefit immensely from Kaggling and (smart) companies reap rewards far beyond what the cost of a Kaggle competition might be! Just the employee branding within the ML community goes a really long way, not to mention that internal teams can learn from the solutions of some of the best machine learning practitioners in the world, who would be extremely costly (or impossible) to hire otherwise.
", "rawMarkdown": "Hey @ikogias!\n\nRecommender systems literally run the world 🙂 If you go to facebook, or Twitter, or YouTube (2nd biggest search engine in the world), or Amazon, or any even moderately big online store, you are bombarded with recommendations 🙂\n\nWhy? Because they work. There are millions of dollars riding on incremental increases in the performance of recommender systems, be that directly in sales or ad revenue. Entire companies were built on recommender systems (many would attribute the rise of TikTok to their recommendation engine).\n\nHere, you are 100% spot on. The website doesn't need to know anything about you, just the sequence of your actions can be an extremely valuable source of information. Better yet, as a company, you don't have to deal with a lot of the personally identifiable information overhead that comes up when storing certain user data (not that you are entirely off the hook in man jurisdictions, but the situation is likely much easier to handle from a legal and security perspective).\n\nPlus, with session-based data you can serve the holy grail of recommender systems, that is real-time recommendations.\n\nImagine that between your actions a website is able to display items that you are likely to buy... how much value would be in that? Most likely, enormous!\n\nOf course, there is the separate, related question of what value companies get out of organizing Kaggle competitions 🙂 To what extent the solutions can translate to actual software put in production?\n\nBut I think that the answer to this question is a resounding yes -- both Kagglers benefit immensely from Kaggling and (smart) companies reap rewards far beyond what the cost of a Kaggle competition might be! Just the employee branding within the ML community goes a really long way, not to mention that internal teams can learn from the solutions of some of the best machine learning practitioners in the world, who would be extremely costly (or impossible) to hire otherwise.", "votes": null }, { "id": "2039644", "postDate": "11/22/2022 11:19:45", "content": "Hi @ikogias, the primary use case of a recommendation algorithm would be a \"most interesting products\" recommendation like the one at the storefront of our shop.
\n
\nThere, users can find a selection of products personalized in real-time that match their interests. We use user behavior-based recommendations for users with or without login alike, as they perform very well in both cases.
We aim to adjust the recommendations to a user's stage in his purchase funnel. This goal led us to develop a multi-objective recommendation algorithm predicting products users want to click on and buy next.
", "rawMarkdown": "Hi @ikogias, the primary use case of a recommendation algorithm would be a \"most interesting products\" recommendation like the one at the storefront of our [shop](https://www.otto.de/). \n\nThere, users can find a selection of products personalized in real-time that match their interests. We use user behavior-based recommendations for users with or without login alike, as they perform very well in both cases.\n\nWe aim to adjust the recommendations to a user's stage in his purchase funnel. This goal led us to develop a multi-objective recommendation algorithm predicting products users want to click on and buy next.", "votes": null }, { "id": "2039654", "postDate": "11/22/2022 11:30:15", "content": "Hey @radek1, you perfectly summarized the impact and use cases of recommendations we deal with on a daily basis. Thanks for your continuous support of the community in general and in this competition in particular. Keep up the great work! 💪
", "rawMarkdown": "Hey @radek1, you perfectly summarized the impact and use cases of recommendations we deal with on a daily basis. Thanks for your continuous support of the community in general and in this competition in particular. Keep up the great work! 💪", "votes": null }, { "id": "2040306", "postDate": "11/22/2022 22:14:30", "content": "Thank you very much, @pnormann! 🙏😊
", "rawMarkdown": "Thank you very much, @pnormann! 🙏😊", "votes": null }, { "id": "2067370", "postDate": "12/16/2022 16:34:54", "content": "Thank you so much for your detailed replies @radek1 and @pnormann
\nI haven't got into the competition yet, but I am curious to know: have people managed to create solutions better than a simple benchmark of showing top selling products? Not sure if there are such simple benchmarks in the leaderboard.
", "rawMarkdown": "Thank you so much for your detailed replies @radek1 and @pnormann \n\nI haven't got into the competition yet, but I am curious to know: have people managed to create solutions better than a simple benchmark of showing top selling products? Not sure if there are such simple benchmarks in the leaderboard.", "votes": null }, { "id": "2067648", "postDate": "12/17/2022 00:21:20", "content": "What people are buying is too varied, the approach that you mention is outcompeted by the co-visitation matrix approach
", "rawMarkdown": "What people are buying is too varied, the approach that you mention is outcompeted by the co-visitation matrix approach", "votes": null } ], "comments": [ { "id": 2038295, "author_name": "radek1", "author_url": "", "post_date": "11/21/2022 09:57:07", "content": "Hey @ikogias!
\nRecommender systems literally run the world 🙂 If you go to facebook, or Twitter, or YouTube (2nd biggest search engine in the world), or Amazon, or any even moderately big online store, you are bombarded with recommendations 🙂
\nWhy? Because they work. There are millions of dollars riding on incremental increases in the performance of recommender systems, be that directly in sales or ad revenue. Entire companies were built on recommender systems (many would attribute the rise of TikTok to their recommendation engine).
\nHere, you are 100% spot on. The website doesn't need to know anything about you, just the sequence of your actions can be an extremely valuable source of information. Better yet, as a company, you don't have to deal with a lot of the personally identifiable information overhead that comes up when storing certain user data (not that you are entirely off the hook in man jurisdictions, but the situation is likely much easier to handle from a legal and security perspective).
\nPlus, with session-based data you can serve the holy grail of recommender systems, that is real-time recommendations.
\nImagine that between your actions a website is able to display items that you are likely to buy… how much value would be in that? Most likely, enormous!
\nOf course, there is the separate, related question of what value companies get out of organizing Kaggle competitions 🙂 To what extent the solutions can translate to actual software put in production?
\nBut I think that the answer to this question is a resounding yes -- both Kagglers benefit immensely from Kaggling and (smart) companies reap rewards far beyond what the cost of a Kaggle competition might be! Just the employee branding within the ML community goes a really long way, not to mention that internal teams can learn from the solutions of some of the best machine learning practitioners in the world, who would be extremely costly (or impossible) to hire otherwise.
", "votes": null, "replies": [ { "id": 2039654, "author_name": "pnormann", "author_url": "", "post_date": "11/22/2022 11:30:15", "content": "Hey @radek1, you perfectly summarized the impact and use cases of recommendations we deal with on a daily basis. Thanks for your continuous support of the community in general and in this competition in particular. Keep up the great work! 💪
", "votes": null, "replies": [] }, { "id": 2040306, "author_name": "radek1", "author_url": "", "post_date": "11/22/2022 22:14:30", "content": "Thank you very much, @pnormann! 🙏😊
", "votes": null, "replies": [] } ] }, { "id": 2039644, "author_name": "pnormann", "author_url": "", "post_date": "11/22/2022 11:19:45", "content": "Hi @ikogias, the primary use case of a recommendation algorithm would be a \"most interesting products\" recommendation like the one at the storefront of our shop.
\n
\nThere, users can find a selection of products personalized in real-time that match their interests. We use user behavior-based recommendations for users with or without login alike, as they perform very well in both cases.
We aim to adjust the recommendations to a user's stage in his purchase funnel. This goal led us to develop a multi-objective recommendation algorithm predicting products users want to click on and buy next.
", "votes": null, "replies": [] }, { "id": 2067370, "author_name": "ikogias", "author_url": "", "post_date": "12/16/2022 16:34:54", "content": "Thank you so much for your detailed replies @radek1 and @pnormann
\nI haven't got into the competition yet, but I am curious to know: have people managed to create solutions better than a simple benchmark of showing top selling products? Not sure if there are such simple benchmarks in the leaderboard.
", "votes": null, "replies": [ { "id": 2067648, "author_name": "radek1", "author_url": "", "post_date": "12/17/2022 00:21:20", "content": "What people are buying is too varied, the approach that you mention is outcompeted by the co-visitation matrix approach
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2038235": "Hi,it may seem obvious to some but can I please ask: what is the use case for such a recommendation algorithm? Imagine we build it tomorrow, how would it be used?\n\nI am assuming that we don't have data from logged-in users of the given website, therefore would this model be useful for website personalisation when it comes to anonymous users browsing based on their shopping behaviour during their session? Say I got to the given e-commerce website that uses this model - how would the model change my experience? It'd be great if you could outline a couple of its use-cases to get a better feel of the purpose it's supposed to serve.\n\nThank you in advance!", "2038295": "Hey @ikogias!\n\nRecommender systems literally run the world 🙂 If you go to facebook, or Twitter, or YouTube (2nd biggest search engine in the world), or Amazon, or any even moderately big online store, you are bombarded with recommendations 🙂\n\nWhy? Because they work. There are millions of dollars riding on incremental increases in the performance of recommender systems, be that directly in sales or ad revenue. Entire companies were built on recommender systems (many would attribute the rise of TikTok to their recommendation engine).\n\nHere, you are 100% spot on. The website doesn't need to know anything about you, just the sequence of your actions can be an extremely valuable source of information. Better yet, as a company, you don't have to deal with a lot of the personally identifiable information overhead that comes up when storing certain user data (not that you are entirely off the hook in man jurisdictions, but the situation is likely much easier to handle from a legal and security perspective).\n\nPlus, with session-based data you can serve the holy grail of recommender systems, that is real-time recommendations.\n\nImagine that between your actions a website is able to display items that you are likely to buy... how much value would be in that? Most likely, enormous!\n\nOf course, there is the separate, related question of what value companies get out of organizing Kaggle competitions 🙂 To what extent the solutions can translate to actual software put in production?\n\nBut I think that the answer to this question is a resounding yes -- both Kagglers benefit immensely from Kaggling and (smart) companies reap rewards far beyond what the cost of a Kaggle competition might be! Just the employee branding within the ML community goes a really long way, not to mention that internal teams can learn from the solutions of some of the best machine learning practitioners in the world, who would be extremely costly (or impossible) to hire otherwise.", "2039644": "Hi @ikogias, the primary use case of a recommendation algorithm would be a \"most interesting products\" recommendation like the one at the storefront of our [shop](https://www.otto.de/). \n\nThere, users can find a selection of products personalized in real-time that match their interests. We use user behavior-based recommendations for users with or without login alike, as they perform very well in both cases.\n\nWe aim to adjust the recommendations to a user's stage in his purchase funnel. This goal led us to develop a multi-objective recommendation algorithm predicting products users want to click on and buy next.", "2039654": "Hey @radek1, you perfectly summarized the impact and use cases of recommendations we deal with on a daily basis. Thanks for your continuous support of the community in general and in this competition in particular. Keep up the great work! 💪", "2040306": "Thank you very much, @pnormann! 🙏😊", "2067370": "Thank you so much for your detailed replies @radek1 and @pnormann \n\nI haven't got into the competition yet, but I am curious to know: have people managed to create solutions better than a simple benchmark of showing top selling products? Not sure if there are such simple benchmarks in the leaderboard.", "2067648": "What people are buying is too varied, the approach that you mention is outcompeted by the co-visitation matrix approach" }, "source": "meta" }