Hi!
\nI am working on a movie recommendation system and I had a question - After I have trained a CF model in Pytorch, how should I make recommendations for users that are not in the training data i.e new users, assuming I have asked the user to rate the the movies they have watched and I have created a dictionary of movieId and rating the new user would give.
\nThanks a lot!
Hi!
\nI am working on a movie recommendation system and I had a question - After I have trained a CF model in Pytorch, how should I make recommendations for users that are not in the training data i.e new users, assuming I have asked the user to rate the the movies they have watched and I have created a dictionary of movieId and rating the new user would give.
\nThanks a lot!
That's exactly what this competition is doing -- session based recommendations, except you have explict feedback from users, the ratings. Check the public kernels you will get some solutions like co visitition, matrix factorizaiton etc. Since the comp is ending soon, so just stay tuned, plenty of solutions will be released by then.
", "rawMarkdown": "That's exactly what this competition is doing -- session based recommendations, except you have explict feedback from users, the ratings. Check the public kernels you will get some solutions like co visitition, matrix factorizaiton etc. Since the comp is ending soon, so just stay tuned, plenty of solutions will be released by then.", "votes": null }, { "id": "2097951", "postDate": "01/13/2023 05:18:51", "content": "Got it! Thanks a lot!
", "rawMarkdown": "Got it! Thanks a lot!", "votes": null }, { "id": "2098036", "postDate": "01/13/2023 07:13:50", "content": "@buumoo I had a look at a public notebook that used matrix factorization and while predicting that notebook fit a knn on the trained embeddings. Could you please help me understand why that would be helpful?
\nThanks!
During training, the next event is set to target 1, so after training the model is able to predict the next event based on current event, so call \"NEP\" -- next event prediction. And the ajacent events' embeddings will be pushed closer to each other. That's why after training, using KNN is able to find the most posible next events based on last timestamp event embeddings.
", "rawMarkdown": "During training, the next event is set to target 1, so after training the model is able to predict the next event based on current event, so call \"NEP\" -- next event prediction. And the ajacent events' embeddings will be pushed closer to each other. That's why after training, using KNN is able to find the most posible next events based on last timestamp event embeddings.", "votes": null } ], "comments": [ { "id": 2097443, "author_name": "buumoo", "author_url": "", "post_date": "01/12/2023 17:21:46", "content": "That's exactly what this competition is doing -- session based recommendations, except you have explict feedback from users, the ratings. Check the public kernels you will get some solutions like co visitition, matrix factorizaiton etc. Since the comp is ending soon, so just stay tuned, plenty of solutions will be released by then.
", "votes": null, "replies": [ { "id": 2097951, "author_name": "hackingpirate", "author_url": "", "post_date": "01/13/2023 05:18:51", "content": "Got it! Thanks a lot!
", "votes": null, "replies": [] }, { "id": 2098036, "author_name": "hackingpirate", "author_url": "", "post_date": "01/13/2023 07:13:50", "content": "@buumoo I had a look at a public notebook that used matrix factorization and while predicting that notebook fit a knn on the trained embeddings. Could you please help me understand why that would be helpful?
\nThanks!
During training, the next event is set to target 1, so after training the model is able to predict the next event based on current event, so call \"NEP\" -- next event prediction. And the ajacent events' embeddings will be pushed closer to each other. That's why after training, using KNN is able to find the most posible next events based on last timestamp event embeddings.
", "votes": null, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "2096527": "Hi!\nI am working on a movie recommendation system and I had a question - After I have trained a CF model in Pytorch, how should I make recommendations for users that are not in the training data i.e new users, assuming I have asked the user to rate the the movies they have watched and I have created a dictionary of movieId and rating the new user would give.\nThanks a lot!", "2097443": "That's exactly what this competition is doing -- session based recommendations, except you have explict feedback from users, the ratings. Check the public kernels you will get some solutions like co visitition, matrix factorizaiton etc. Since the comp is ending soon, so just stay tuned, plenty of solutions will be released by then.", "2097951": "Got it! Thanks a lot!", "2098036": "buumoo I had a look at a public [notebook](https://www.kaggle.com/code/cpmpml/matrix-factorization-with-gpu) that used matrix factorization and while predicting that notebook fit a knn on the trained embeddings. Could you please help me understand why that would be helpful?\nThanks!", "2098123": "During training, the next event is set to target 1, so after training the model is able to predict the next event based on current event, so call \"NEP\" -- next event prediction. And the ajacent events' embeddings will be pushed closer to each other. That's why after training, using KNN is able to find the most posible next events based on last timestamp event embeddings." }, "source": "meta" }