Study On Chinese Emotion Expression Knowledge Base Construction And Its Application In Emotion Analysis

As human-computer interaction applications are well known and used increasingly, computer is expected to have the same sentiment and emotional processing capacity as human. In recent years, the wide spread use of social media allows users to generate texts on the Internet, especially microblog, blogs and product reviews with personal emotions. The online information promote the study about analysis and tracking of large-scale real individual emotion, it shows important research significance and broad application prospects in social, political, economic and some other fields.This paper studies the construction of Chinese basic resources and its application in text emotion analysis. The research mainly arounds three aspects, namely, emotion model, construction of basic resource for emotion words and multi-label text emotion classification. There are four tasks have been done in this subject, firstly, due to the problem of shortage of Chinese emotion lexicon, a high quality and wide coverage one is constructed automatically based on English emotion lexicon Word Net-Affect, it provides a reliable basic resource for the follow-up text emotion analysis. The main steps are machine translation, noise filtering and semantic based extension. Secondly, the existing Chinese emotion lexicons are generally incomplete and even inaccurate, emotion word only has emotion class and strength information in previous study. In this paper, the word’s emotion type is divided into two kinds, expression and congnitive. The main work here is drilling down into emotion word’s emotional expression information, and then joining How Net’s word concept explanation to solve the problem of polysemy. On the basis of the above work, a fine-grained Chinese emotion expression knowledge base(EEKL) is constructed. Thirdly, faced with the situation that network texts grow rapidly and short sentence length usually carries poor information, a rule based new word detection method is used to expand knowledge base automatically, and word’s concept meanings are used to extend sentence at the same time. Finally, multi-label text emotion classification experiments based on semantic algoritm and machine learning algoritm have been finished, and both of the algoritms apply emotion word resources. Results show that the classification performance of Chinese emotion lexicon and EEKL constructed in this paper are better than traditional emotion word resources. On the other hand, after combining the existing feature representation with EEKL information, the classification performance is improved effectively.The main contribution of this paper are as follows. First, a high quality Chinese emotion lexicon and a finest Chinese emotion expression knowledge base by now have been constructed. Second, the rule based method can expand the scale of knowledge base, the use of word’s concept extension helps to improve the text emotion analysis performance. Finally, compared to traditional Chinese emotion lexicons and the existing feature expression method, the application of new emotion lexicon and fine-grained Chinese emotion expression knowledge base promote the performance of multi- label text emotion classification, which embodies their advantages and effectiveness on text emotion calculation application.