Exploring user historical semantic and sentiment preference for microblog sentiment classification

Microblog text is usually very short, thereby challenging existing sentiment classification methods by providing models with little context. Recently, historical user information has been widely used in many real-world applications, such as recommender systems. However, few research works consider user historical states in the loop of microblog sentiment analysis. In this work, we propose to involve historical user information for microblog sentiment analysis to alleviate the context sparsity problem. In particular, we propose a novel neural microblog sentiment classification method which learns informative representations of microblog posts by exploiting both a user's contextual information and his/her historical state information. The proposed method consists of four components, i.e., a micropost encoder, a user historical sentiment encoder, a User Historical Semantic Encoder, and a micropost sentiment classification component. Extensive experiments are conducted on real-world data collected from Weibo, and experimental results show that the proposed approach achieves superior performance as compared to state-of-the-art baselines.