PERCEIVE Benchmark Reimagines Social Media Emotion Analysis

Current social media emotion analysis largely ignores that different readers react differently to the same post—it's stuck in an author-centric view. That changes with PERCEIVE, a new bilingual (English and Chinese) benchmark from researchers Jian Liao, Yujin Zheng, Suge Wang, Jianxing Zheng, and Deyu Li. It’s the first to integrate five critical dimensions: author-created content, genuine reader emotional feedback (extracted from their comments), communication behavior, user attributes, and the underlying social graph. By capturing real-world interactions, PERCEIVE enables a paradigm shift toward truly personalized, reader-centric analysis—where each reader’s emotional response to the same content is naturally represented through their actual comments and social connections.

The team evaluated state-of-the-art methods, including large language models (LLMs) with advanced reasoning enhancements, against PERCEIVE’s comprehensive protocol. The results exposed significant shortcomings—current systems fail to handle the multifaceted, user-aware nature of the task. This suggests that existing NLP models lack the social intelligence to understand how emotion and communication behavior intertwine across different individuals. PERCEIVE provides a foundational resource to push future research toward socially-intelligent NLP, bridging the gap between raw sentiment detection and the nuanced, context-dependent emotional experiences of real social media users.