Awesome data. Really interesting to look at, and great presentation.
But there are a few things that kinda bother me with this:
The problem I can find with this data is that it isn't a representation of the reddit hidden communities as a whole, just the hidden communities of those who actually post (only 20% of Reddit).
A question I have is whether these are two-way connections with the groups. It's not clear exactly how the analysis is done 100% (perhaps I missed this portion), but could connections between subreddits be generated by there being a lot of people who post in a very tiny subreddit also posting in a larger subreddit? This means that though someone may like Large Subreddit A, they may not like the more specific Subreddit B. But a lot who like Subreddit B like Subreddit A.
I combatted this in two ways - first, I only looked at the top 433 reddits.
There are always going to be the same people cross-subscribed between A and B as between B and A. This graph is not of the number of people cross-subscribed between two reddits - it's of the sum (number of people cross-subscribed)/(users in A) + (number of people cross-subscribed)/(users in B). So if a lot of people in a tiny subreddit are cross-subscribed, they get a big boost from the first term, but almost no boost if they make up a tiny sliver of subscribers to reddit B.
Comments
Awesome data. Really interesting to look at, and great presentation.
But there are a few things that kinda bother me with this:
The problem I can find with this data is that it isn't a representation of the reddit hidden communities as a whole, just the hidden communities of those who actually post (only 20% of Reddit).
A question I have is whether these are two-way connections with the groups. It's not clear exactly how the analysis is done 100% (perhaps I missed this portion), but could connections between subreddits be generated by there being a lot of people who post in a very tiny subreddit also posting in a larger subreddit? This means that though someone may like Large Subreddit A, they may not like the more specific Subreddit B. But a lot who like Subreddit B like Subreddit A.
I combatted this in two ways - first, I only looked at the top 433 reddits.
There are always going to be the same people cross-subscribed between A and B as between B and A. This graph is not of the number of people cross-subscribed between two reddits - it's of the sum (number of people cross-subscribed)/(users in A) + (number of people cross-subscribed)/(users in B). So if a lot of people in a tiny subreddit are cross-subscribed, they get a big boost from the first term, but almost no boost if they make up a tiny sliver of subscribers to reddit B.