Abstract

The ubiquity of social media has led to a recent upsurge in the amount of expressive user-generated content, and leveraging this for a fine-grained analysis of the users’ emotions can pave the way for several other downstream tasks. Building on existing approaches for emotion spectrum analysis, we conduct a large-scale study of user emotions on Vent, the largest social media dataset with user-annotated textual posts from 63 distinct emotion categories. Our experiments establish the utility of linguistic homophily and psychological context using community-sensitive and temporal activity-based user-profiles for fine-grained emotion analysis.