Spencer Kee

Date of Award


First Advisor

Aaron Williams

Second Advisor

Harold Hastings


In this thesis we investigate online communities. In particular we are concerned with online community collapse due to the incursion of new users who, intentionally or not, irreparably change the identity of the group as the content and conversation begins to pander to the lowest common denominator. We analyze this issue by examining Reddit, a social news aggregation platform and the ninth most visited website in the world. Reddit's design, particularly the 'hot algorithm' for sorting content, continuously causes the downfall of online communities by encouraging low quality content and the encroachment of casual users, displacing the members who originally formed the community in a process that mirrors gentrification. We introduce a novel hybrid recommendation system and posit that it can both attract members who care about the integrity of a group to communities that need them and prevent the buildup of low effort content, staving off online gentrification. We also explain why recommendation on sites like Reddit is particularly difficult. The new system utilizes topic modeling, a text analysis tool, in an interesting way on rating information. As a specific application we demonstrate how our system discovered abstract topics in a movie ratings database on its own and outline the future directions of our work.