I think for smaller subreddit's there's not as much data, so results are more noise. Apparently Jaccard similarity was used, except for the really big subreddit

I think for smaller subreddit's there's not as much data, so results are more noise. Apparently Jaccard similarity was used, except for the really big subreddits: https://github.com/anvaka/sayit https://github.com/anvaka/sayit Unrelated, but had to say it somewhere in this discussion: Thanks @anvaka for your open source graph libs! Way more performant than anything else I've tried.