Contextual Sentiment Topic Model for Adaptive Social Emotion Classification

Yanghui Rao

Abstract: Social emotion classification is important for numerous applications, such as public opinion measurement, corporate reputation estimation, and customer preference analysis. However, topics that evoke a certain emotion in the general public are often context-sensitive, making it difficult to train a universal classifier for all collections. A multilabeled sentiment topic model, namely, the contextual sentiment topic model (CSTM), can be used for adaptive social emotion classification. The CSTM distinguishes context-independent topics from both a background theme, which characterizes nondiscriminative information, and a contextual theme, which characterizes context-dependent information across different collections. Experimental results demonstrated the effectiveness of this model for the adaptive classification of social emotions.