Collaborative Filtering for Item recommendation

Hi, Thanks to FastAi, that I am at-least beginning to think about using DL :slight_smile:

I am trying to use movielens approach for recommending an in app activity to user(item) based on the previous interactions and other user data.

Here are my questions

  1. Lesson 5 sample doesn’t seem to use timestamp for making recommendations, how can I make use of this?
  2. In my case rating aspect is binary ie user picked an activity or did not. So my scale is for y is either 1 or non existent. Has anyone worked with this kind of data with collab_learner?
  3. How can I add more features to item ( like category or hierarchy) while training?
  4. How can I add newer interactions to the model without retraining the model.

I have seen systems like AWS Personalize and Recombee achieving similar things.

Any directions or further reading on the lines of FastAi will definitely help me to take this further.