A dictionary-based sentiment classification method considering subject-predicate relation

In order to achieve enhanced credibility of social networking services such as twitter, it is necessary to (1) identify the topic and to (2) check, if the majority of the tweets having the same topic show the same opinion. Therefore, it is indicated to improve the accuracy for analyzing the caller's emotional expression of the “emotional polarity classification”, which is used for opinion classification. For this reason, a semantic orientation acquisition technique is proposed, which integrates “another part-of-speech based semantic orientation dictionary” and a subject predicate relationship considered semantic orientation dictionary of phrases (semantic orientation table). As a feature of the proposed method, a semantic orientation of an adjective or predicate due to its greatest impact on identification semantic orientation of each tweet or document is determined. In addition, the dictionary is created by using tweets. Usual methods have problems with the opinion classification (90 % of the tweets are judged negative), but the proposed method has a more correct to negative judge rate of 60%. An integrated semantic orientation dictionary of phrases is used to improve precision of the classification.