Is anyone doing movie recs based on deep learning + the individual frames in each movie? I think that could be an interesting way to learn nuanced things like "you like heist movies with mixed gender casts" or "you hate action movies that have too many gunfights."
-
Once you start filtering based on ‘good’ movies, you realize the problem isn’t discovery - it’s that the quantity of good movies is very low.
The other problem is that movie scores (like IMDb, 1-10 scale) are tough to read. The worst movies gets a ~3.x and the best movie gets a ~9.3. So a 7.8 sounds good but it’s actually awful.
