It is my first time dealing with recommender systems, so I don't understand which algorithm should I use. I am given dataset with columns "user_id", "item_id", "like" (0 or 1), 32 features for each item and 32 features for each user. This is how correlation matrix looks like(first 32 features for items, last 32 for users): 
$\begingroup$ $\endgroup$
Add a comment |
1 Answer
$\begingroup$ $\endgroup$
The simplest method would be to generate a feature vector of [[user_features], [item_features]] and send it to a random forest or logistic regression model.
If you want to do a more modern embedding-based approach, you can initialize an embedding for each user/item and train embedding similarity with cosine similarity loss. You can either init a random embedding for each user/item and learn relationships through like data, or have a MLP-type model that predicts embeddings from user/item features.
Embeddings are nice because then you can do things like run a nearest neighbor query between a user's embedding and all items, pick the top-k nearest neighbors and use those as recommendations.
