Opinion mining for multiple types of emotion-embedded products/services through evolutionary strategy

Since the advent of blogging, microblogging, and social networking sites, researchers and practitioners have been increasingly concerned with the problem of obtaining useful evaluations from web-based opinion articles in a process known as opinion mining or sentiment analysis. In this study, we focused on reviews based on highly emotion-embedded products/services, such as movies, music, and drama. Furthermore, we tried to solve the multiple polarities problem for the same review word for multiple types of product/service. First, we collected text written in Chinese from a Taiwanese movie forum. In our proposed approach, we applied an evolutionary strategy algorithm to optimize the weight tables corresponding to two different types of movies: horror and drama movies. The experimental results indicated that the proposed method performed better than conventional methods when considering only one generalized type. Further, we employed a new multi-class support vector machine approach for predicting opinions at the document level. We used seven measures to describe the characteristics of an overall document, including the central tendency, dispersion, and shape of the predicted sentence value distribution, where the fluctuations in these values corresponded to their positions in the document. We also demonstrated the effectiveness of this approach for identifying opinions at the document level.