Hacker News new | past | comments | ask | show | jobs | submit login
The Angry Video Game Model: neural network arch to predict videogame ratings (jgeekstudies.org)
46 points by zdw on April 4, 2024 | hide | past | favorite | 9 comments



Hmm so it's about using neural nets to predict numerical scores from text reviews left by users?

If you have non average tastes for entertainment perhaps you could train something like this to give you custom scores that apply to you personally...


It's the Netflix Prize from 2006 all over again.


That didn't quite work. At least in my Netflix.


I'm not sure it was ever implemented. Netflix had already moved onto streaming, and recommending a show to watch next became irrelevant because the cost to switch was so low.


But they try to recommend you stuff. And can't really figure out what i want, at least in my case.


it is already easy to do this. Unless you are a generally negative person, you will enjoy things that don't do something you actively dislike, right? so then just look at negative reviews only. If there are no issues raised that you would personally not be able to overlook, it's probably a good game.


Someone's negative/positive is another person's positive/negative sometimes. Or someone's negative/positive is another's "i don't care".

I'm not thinking of just selecting any piece of crap that is bearable (that's like 2/3 of netflix), but of selecting stuff i'll really like.


same, but at the same time, positive reviews tend to be low entropy, little is to be gained from positive comments that isn't already captured by the average score. You get a few detailed ones which go into specifics as to why they thought this was thing was good, but negative ones tend to go into way more, and more reliable, detail. I do understand where you're coming from though.


Every developer/publisher has a "neural network" (pun intended) of their own that knows exactly how the game will be rated, however it is always ignored when it comes to greed and probing how much you can milk the client.

A model for this is just a scapegoat.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: