MS-SAE: A General Model of Sentiment Analysis Based on Multimode Semantic Extraction and Sentiment Attention Enhancement Mechanism.

Recently, there is a lot of research on sentiment analysis. When existing models extract text features, semantic information can’t be fully obtained, because they ignore context connection between historical texts and current texts. Meanwhile, models cannot self-optimize extraction algorithms, making key sentiment semantics be neglected, because they can’t track and feedback analysis results. Furthermore, models extract insufficient sentiment features, resulting in imperfect semantic extraction and unsatisfactory analysis results, because they only extract POS features. To solve the above problems, we propose a general model of sentiment analysis based on multimode semantic extraction and sentiment attention enhancement mechanism (MS-SAE). The model includes a multimode semantic extraction processing (a multimode semantic extraction module and a sentiment attention enhancement mechanism), an extended dictionary (ExWordNet) and a sentiment analysis module. The multimode semantic extraction module extracts semantic features from multiple perspectives and pays close attention to extracted features, which solves the problem of insufficient semantic extraction. We propose a sentiment attention enhancement mechanism to solve the problem that sentiment semantics is neglected. We construct a general extended dictionary to support MS-SAE in semantic extraction processing. The LSTM-based sentiment analysis module ensures the accuracy of sentiment analysis. We evaluate MS-SAE on SST-2, MR and Subj datasets. Extensive experiments have been conducted and the results demonstrate that MS-SAE could achieve better sentiment analysis performance than the state-of-the-art algorithms in accuracy. It solves the problems including poor understanding of text semantics and errors in analysis results.