Understanding Communities via Hashtag Engagement: A Clustering Based Approach

Abstract

We develop insight into community use of hashtags on social media and find that hashtags with behavior indicative of real world communities are more engaging. To do this, we study the relationship of hashtag usage with user engagement on Twitter. Hashtag engagement is useful as a surrogate measure of how active community members are. We develop a framework for describing hashtag temporal usage, show the existence of 4 broad classes of hashtags, and show that the engagement of a hashtag varies significantly between classes. Periodically used hashtags, such as for TV shows and weekly community chats, are the most engaging, while hashtags relating to events are the least engaging. Looking at how community dynamics vary within this framework reveals that a hashtag being used more frequently is not positively correlated with it being more engaging. We then explore the periodically used hashtags and find negative correlations with diversity of the user base, which implies concentrated communities are the most engaging. We conclude by studying a set of community conversation-oriented hashtags and find these hashtags to be more engaging than other hashtags, regardless of dynamic type. Our findings support the hypothesis that hashtags with stronger community behavior are more engaging.