2 ms·

>I used my own metric that is based on jaccard similarity. Which in turn is based on "users who posted to X also posted to Y" metric. Is that really a good met

>I used my own metric that is based on jaccard similarity. Which in turn is based on "users who posted to X also posted to Y" metric. Is that really a good metric of similarity? Just myself, I post in several unrelated subreddits semi-regularly from programming to video games to music, art and even stone masonary, i've posted in subreddits for TV shows i've watched, or just completely random things. I use reddit as a place where I can learn about and interact with people on nearly any subject or topic and I take advantage of that when I can. I'm sure my posting habits aren't that unusual. I'm just not sure that's really an accurate way to gauge similarity.

anvaka 7y ago
Right, I don't think it's 100% accurate either. It gives just some hints what might be related, but like you said, it's not necessary the best possible measure of similarity. Jaccard similarity does not count only the number of people who posted to A and B, it checks how many people posted to A, how many people posted to B, and how many of those people posted TOGETHER to A and B. That togetherness gives us hints what is related, and after it is computed, we can divide by the total number of poster to both A and B (independently), which brings the value to something that we can use to compare against other subreddits. If that value is close to 1, it means that almost all users who posted to A have also posted to B. If it is close to 0, then the overlap is much smaller.