Algorithmic recommendations, such as recommended videos on Youtube, are widely cited as problematic. But what about recommendations generated by humans? In this interactive data visualization, we map the ways in which links on reddit can lead users down "rabbit holes" from mainstream communities to more niche, sometimes toxic ones.
The network structure of sidebar links
On Reddit, every subreddit community has a sidebar where the moderators often put information and links for their users to read. The links may go to other subreddits, sometimes to promote them or to suggest further reading. These links form a directed graph, where nodes are subreddits and edges are links between them. The links make it possible to take a path from one community to many others in a few short steps.
While this property may allow users to explore their diverse interests, it may also lead them to smaller communities that harbor more extreme or toxic content. They may also become stuck in an echo chamber with limited access to opposing views.
Sentiment Analysis
"Toxicity" is a difficult property to measure. As a proxy for toxicity, this visualization uses sentiment intensity. By averaging the strength of the positive or negative tone of 100 prominent posts on a subreddit, we get a single sentiment score between 1 and -1, where 1 is most positive and -1 is most negative.
More Positive Tone
More Negative Tone
This sentiment analysis has limitations: for example, it cannot detect sarcasm and it does not recognize coded terms used by communities in ways that differ from their original meaning. Along with each subreddit's sentiment score, we provide example posts to give context to the score.
