Summary
In “Antisocial Behavior in Online Discussion Communities”, Cheng takes a look at what defines anti-social behavior and if it can be predicted. Using 3 different websites, CNN, IGN, and Breitbart, he studies the difference between future banned users (FBU) and never banned users (NBU). He finds that there are noticeable differences between FBUs and NBUs. For instance, FBUs tend to use more negative language, post more often, the quality of their posts are poorer than NBUs, and their posts are more likely to get deleted the longer they are online. He also looked at the different types of anti-social behavior. Some users called Hi-FBUs had their posts deleted more often than Low-FBUs due to the language they use in their posts. He outlined a four ways to identify anti-social behavior:
- Post readability
- How often/where they post
- Community interactions
- If moderators delete the post
Using those features they were able to predict if the user would be banned after the first five posts made by a user with an AUC of 0.8.
Reflection
One thing I found interesting was the difference between the websites themselves. CNN was much quicker to ban a user than IGN. I suppose this could be attributed to the credibility of the website as CNN is slightly more legitimate or serious than IGN or perhaps due to the presence of moderators. I also thought it unsurprising that FBUs wrote poorer quality posts than NBUs since their purpose is probably to incite (and poorly written posts themselves can do that) so they wouldn’t take the time to write more eloquently.
Questions
Is there an alternative to moderators or community feedback that would help prevent anti-social behavior?
Would more crowdsourcing change or affect the outcome of some of their results?
How could we encourage anti-social users to be more social?
