Emotion Detection In Online Social Network Based On Multi-Label Learning

Emotion detection in online social networks benefits many applications such as recommendation systems, personalized advertisement services, etc. Traditional sentiment or emotion analysis mainly address polarity prediction or single label classification, while ignore the co-existence of emotion labels in one instance. In this paper, we address the multiple emotion detection problem in online social networks, and formulate it as a multi-label learning problem. By making observations to an annotated Twitter dataset, we discover that multiple emotion labels are correlated and influenced by social network relationships. Based on the observations, we propose a factor graph model to incorporate emotion labels and social correlations into a unified framework, and solve the emotion detection problem by a multi-label learning algorithm. Performance evaluation shows that the proposed approach outperforms the existing baseline algorithms.