Chinese Network Text Sentiment Analysis And Research Based On EDT

With the deepening of user participation, the text including the user point of view, attitudes and news events, products and other subjective comment quickly filled with the Internet. News, blog, forum, paste and other online media their active users’ number is very big, became the news event propagation main carriers. But these texts different form each other in structures, and have their own characteristics, combined with the huge amount of data, bring challenges to propensity analysis and opinion mining. This paper aims to do propensity analysis on simple sentence on Chinese news texts based.Firstly, made a introduction and summary of the background and the development status of sentiment analysis at home and abroad. Made a brief introduction for these commonly used analytical techniques, such as data cleaning, participle, speech tagging, feature extraction, build emotional ontology library and major research ideas of text sentiment classification methods and techniques. Secondly, considering the analysis of text level theme emotion is very complex, we decomposed the task into topic identification, identification and classification of the emotional theme of subjective and objective three sub tasks. Fusion a variety of methods to extract theme features and put forward a multi-features theme recognition model based on feature space vector distance to identify the text topic. Then, Social media has become an important channel and carrier for the dissemination of information, this paper analyzed the main social media and the difference in their text structure. Focus on the news, blog, forums and other media news text as a research object. These media text are relatively long text, taking the emotion key sentence in the text as the whole chapter’s emotion, and build analysis model based on simple sentence. According to the Chinese expression habits and grammatical features presented use emotional dependency tuples(EDT) as the basic structure of emotional expression, and summarized the rules of extracting emotion expression structure. Established a theme emotion distinguishing model.Finally, through experiments to determine the method parameters and different methods of weighting coefficients, applied this method to the evaluation of COAE2014, achieved an evaluation good result. Conducted comparative experiments with supervised classification algorithms(KNN, SVM) and semi-supervised algorithm(K-MEANS). And analyzed the experimental results if shows that the classification performance of EDT emotion-based have considerably performance with supervised machine learning algorithms, and much higher than the semi-supervised clustering algorithm.