Beyond Structure: Integrating Sentiment Dynamics into Community Detection on Reddit

Traditional community detection methods focused on structural metrics. Though these methods were capable of showing the communities of users who behave similarly, they failed to depict very key characteristics of similarity such as “emotion”. Sentiment plays a very important role in social cohesiveness. It is evident that communities may evolve and gain strength based on similarity in opinion and are inclined towards positive emotions. The negative emotions symbolizes a conflict or a split within a community. Therefore, the study of community evolution through the context of emotions towards a post is a relevant study under the domain of community detection. This article proposes a novel emotion-aware community detection framework that embeds sentiment dynamics with traditional structural analysis. The new framework is empirically evaluated real-time dataset extracted from Reddit. The user-interaction networks are modelled using the sentiment scores. For tracking community evolution, we apply an overlap-based tracking method which can keep community identity even when the member composition changes. The results clearly exhibit the relationship between emotional tone and the structural change in community formation.