Sentiment analysis in text mining using ensemble-HMM

Abstract

With a rapid growth of social media, text mining is extensively utilized in fields, and sentiment analysis plays an important role to analyze opinion in texts. Methods in sentiment analysis generally depend on sentiment lexicon that is a set of predefined key words that express sentiment. It requires proper sentiment words to be extracted in advance and suffer from classifying sentences that imply opinion without sentiment key words. This paper presents a new sentiment analysis method, based on ensemble text hidden Markov models (TextHMMs), that uses a sequence of words in texts instead of predefined sentiment lexicon. We attempted to learn patterns of text representing sentiment through ensemble TextHMMs. Our method defines hidden variables in HMM by semantic cluster information in consideration of the co-occurrence of words and calculate a sentiment orientation of sentences by fitted TextHMM. To reflect diverse patterns, we applied an ensemble of TextHMM-based classifiers. In the experiments with benchmark data sets, we show that the method is better than some existing methods, particularly in classifying implicit opinion.