3 ms·
> Even Netflix "had it all" with the million dollar recommender prize they set out in 2014, and I know their data science team is top notch. But if I scroll thr
by heavyset_go 5y ago
> Even Netflix "had it all" with the million dollar recommender prize they set out in 2014, and I know their data science team is top notch. But if I scroll through their experience with my personal account in 2021, I basically still get "what all others are currently watching in {your_country} now" and I'm less than thrilled.
IMO, Netflix's recommender system suffers from the simple fact that the company itself doesn't want to collect accurate user ratings and make recommendations with them. If they did that, then a not insignificant portion of the content and deals they invested in would get memory-holed due to bad ratings or lack of interest, and that would make the company and its leaders look bad.
- cottager2 5y agoWhy don’t they want to collect accurate user ratings? Also, they know whether or not you finish something, how many episodes you binge, etc. There are many rating-independent signals.
- bllguo 5y agocan you elaborate on this? can't they collect ratings for internal use and not surface them, if that's the issue? they should be able to tune the recommender so that it doesn't completely forget about lower-rated content the way you're saying
- pugets 5y agoNetflix used to do that. You used to be able to rate programs between 1-5 stars and see the average rating. I believe it also recommended other programs to watch based on what you rated highly. They did away with that system and replaced it with their custom recommendation engine. I believe, without any evidence, they did this to make bad shows look better. They once hosted a documentary about sex workers (can't recall the name, sorry) that had an average rating of around 2 out of 5 stars, and after the recommendation system change, it turned into a "75% match" on my account.
- SergeAx 5y agoA movie may have an average rating 2/5 but still be a 75% match for you personally, there's no discrepancy here.
- chefandy 5y agoThat sounds like functionality tunnel vision, to me. Matching users to movies isn’t an abstract metadata networking exercise— the larger goal is recommending things users enjoy so they keep watching. I’m far less likely to suffer a bad movie if it closely aligns with my interests. Poor representations of my passions and crafts frustrate the hell out of me.