7 ms·
Kind of off topic, but I've been thinking about building a site like HN/reddit, but instead of ranking stuff based on how many votes it has, use machine learnin
by Houshalter 13y ago
Kind of off topic, but I've been thinking about building a site like HN/reddit, but instead of ranking stuff based on how many votes it has, use machine learning to predict how likely a specific individual is to upvote it.
That way everyone's front page is based off what they tend to like, and what people who like similar stuff to them liked. That way you don't have to fracture stuff into 50 different communities and try to figure out what kind of content is acceptable in each, and what ones you want to subscribe to.
- helloTree 13y agoBut probably you get stuck in local maxima. E.g. if I am on Hacker News I want to find new and exciting stuff and not things I already know ...
- Houshalter 13y agoYa you'd have to take that into account somehow. If 50 other people that are also interested in "technology" or whatever vote for something, then it's likely you will like it as well. There is no reason you have to select content purely based on the website it's from, or the words it contains, but it's a really good filter for getting rid of stuff you definitely don't like. I.e. if you filtered reddit this way, most images and memes would get ranked way lower for me. Someone else might have articles or news stories ranked lower. But mostly I'm looking to automatically find people who have similar interests or voting patterns, and then selecting based on what those people voted for to see what you might like.