Most Chinese opinion mining systems use the specific pattern and nearby approach to combine relevant opinion elements (feature words and opinion words) to express the opinion tendencies of authors. In this paper, we propose a rule-based ad hoc method to study the combination problem of Chinese opinion elements. We extracted the opinion elements of articles based on lexicons and then combined them with the different sentence patterns and grammars to analyze the authors’ opinions. Because the articles on the online communities such as blogs, wikis, online forums, etc. do not have a defined format, there are often opinion comments that do not refer to the topic, resulting in information loss and significantly reduced recall. Therefore, the “default topic” method is proposed to correct this type of problem. Additionally, there might be errors when using the nearby approach to combine opinion elements. Thus, we propose the concept of “clause priority” to increase precision. After 20 months of long-term tracking and analysis, the experimental result indicates that the method proposed in this paper had good precision, recall, and F1 of opinion tendency analysis for review articles.
Journal of Convergence Information Technology 10(2), p.137-144