Hacker News new | ask | show | jobs
by dannytatom 2327 days ago
This has a similar issue to the way Reddit recommends subs itself. A sub will be considered similar if it's the exact opposite due to users of one sub going to another to shitpost or brigade or etc.

I'm not sure of a fix for that, but would it be possible/helpful to weigh it by average upvote/downvote of the comments left from users of said sub? Meaning if sub A is about how much baseball sucks, and sub B is about how amazing baseball is, while determining if the 2 are similar you'd find out most posts from sub A to sub B are heavily downvoted and so probably not similar.

1 comments

It seems though relationship can be considered to have a sign: positive, when people mostly align in their upvote intents, and negative when people do the brigading, etc.

I'd still see ability to determine absolute value of relationship as a valuable property of a recommender