Emotion Analysis Of Chinese Microblogs Using Extended Emotion Lexicon

With the popularity of social media such as micro-blog in people’s life, micro-blogproduced great influence on people’s life and work. Therefore, micro-blogsentiment analysis has become an important research in Natural Language Processing. Automatic analysis of the emotional content in Chinese micro-blog is at the stage of beginning, emotional dictionary is an important resource micro-blog emotional analysis. But the emotional dictionary construction is still not perfect, this is one of the important reasons of low accuracy of the Chinese micro-blog emotional analysis. This paper propses a method to identify emotions in micro-blog text on the technology of the extended dictionary. In some corpus, emotional dictionary is extended by adding new emotiona words and annotating emotional intensity for each emotional words. Identify six emotions in micro-blog text by extended emotional dictionary.Firstly, this paper proposes a method to exploit emotional words. The method is to detect emotional words, which are not in the existing emotion lexicons but express emotions in the corpus. In order to detect emotional words and identify the emotion a word denotes in a corpus, we make use of a set of seed emotional words and investigate the similarity between candidate words and seed emotional words in the corpus to detect candidate emotional words. After that we use the identified emotional words to extend the existing emotion lexicons, and use unsupervised and supervised support vector machine(SVM) method for sentiment analysis. Compared the results with the existing emotion lexicons, the extending emotion lexiconssubstantially improved the emotion lexicon coverage for weibo, and significantly improved sentiment analysis accuracy.Secondly, we dealt with the emotional words with weight. We investigated the co-occurrence patterns between emotional words and fine-grained emotion category to detect the emotional arousal-nonarousal aroused by the emotional words. The differences of emotional arousal-nonarousal are used as the weight of emotional words. The experimental results show that compared with the emotional strength denoting in the existing emotion lexicon, the method calculating the weight of emotionnal words more accurately illustrated the words on the role of emotional expression and effectively improved the accuracy of sentiment analysis. This method can be effectively integrated the advantages of multiple emotion lexicon, and further enhanced the accuracy of sentiment analysis.Finally, this paper combines the advantage of the two previous works to extend the emotion lexicon.We recognised emotional words and detected the weight of the recognised emotional words based on the weight of seed emotional words.Identify emotions in micro-blog text using a rule-based approach and a supervised approach(SVM). The results show that this dictionary havd better performance in identify six emotions in microblogs than others.