neeraj1987

Recommender systems moved from explicit feedback (like ratings) to implicit feedback precisely because users are less likely to actually rate stuff and also because ratings are subjective; by which I mean your interpretation of 3 stars(good) may not agree with mine(average). I have watched tons of movies/shows on netflix or videos on YT for that matter but have not rated a single video. To address the other part of your suggestion i.e collapse ratings/feedback across media like movies,books,etc. usually it is very difficult to have a dataset that spans multiple media across the same set of users. Even if it is present it would be too sparse (more sparse than usual for a site like YT with a continuously changing content library) to actually help. Though I agree that if anybody can get the recommender right, it is Google with the sheer amount of info it has on each user.

In general what could be a separate source of embeddings? Also, how do these embeddings compare against traditional CF based latent factors?(I ask this in terms of a recommender metric and not complexity)

Absolutely brilliant advice. Sort off in the same boat as OP (though not so much on the depression front), but I am going to give this a shot; write down things that specifically make me feel valuable in a company and see if the list aligns with what my company/boss thinks.

Thanks!

Thanks a lot :-)

hi,very informative talk; especially with those examples for handling cold start and seeding. any pointers on how the multiple entities are incorporated in the interaction matrix? I understand how user/item attributes may be incorporated in the interaction matrix but multiple entities is something that I am struggling to understand. Pointers to associated literature would help too.