Extracting morphemes to improve sentiment analysis for Chinese fund reviews

Abstract

Fund reviews on the Web can help people to make investment decisions and their size is relatively small comparing with other products, which lowers PMI algorithm performance. Replacing words by morphemes can improve system capability in Chinese language environment, but defining the SO (sentiment orientation) of morphemes is generally a kind of heavy handwork. We propose a methodology to do this job automatically, whose basic idea is to introduce some outside language and statistical resources to assist polarity calculation. Experiments show that the algorithm is effective and the system get better performance.