Research On Key Technologies Of Text Sentiment Analysis For Social Media

In recent years, social media has developed rapidly and become the world's largest public data source. As an important information carrier in Web, the text has an exponential growth trend and contains a great research and commercial value. Text sentiment analysis, as one of the main techniques for text analysis, can provide the necessary technical support for social media text analysis. Therefore text sentiment analysis technology based on social media has attracted widely attention. Sentiment analysis for social media contains four tasks, including sentiment information retrieval, sentiment information extraction, sentiment classification, and sentiment summarization. It analyzes and mines the view of expressions and the tendency of the user's subjective text in a variety of network platform. The above mentioned work can assist the decision making of different users, researchers, business organizations, and government agencies. This thesis carries out innovative researches on social media sentiment analysis tasks from the following aspects, and the main contributions of the thesis are as follow:(1) In order to solve the problem of key information extraction in socail media text, how to efficiently mine the product features from the product review corpus is studied. Two kinds of relations between opinion words and product features are defined to achieve the feature extraction work. One relation between the product features and opinion words occur in the same sentence is called Local Context Information (LCI), and the other relation between sentences is called Global Content Information (GCI). After combining the two kinds of information, the algorithm can extract product features accurately. Then product features are grouped and sorted on the basis of the importance, which can help the customers quickly find the important features of products. The experimental results show that the proposed method is suitable for product feature extraction, and it can improve the accuracy of product feature extraction to a certain extent.(2) Due to the complexity and diversity of emotion in social media text, this thesis studies how to introduce cognitive thinking modes into text sentiment analysis, and uses thinking modes to guide sentiment classification. According to the emotion expression difference problems triggered by different thinking modes existed in different nations, areas and language, the specific method of using quantitative thinking modes is proposed, which uses the thinking modes difference to assist sentiment classification on social media text. The experimental results show that considering the characteristics of the thinking modes in the Chinese and English corpus are helpful to improve the precision of sentiment classification on social media texts.(3) Due to hot topics continuously appeared and spreaded in social media platform, this thesis studies how to find the hot topic from the massive microblog information, and use Sina Weibo as the research object. In the micro-blog platform hot topics will cause users' emotion fluctuation, and then users tend to write more emotion words to express their feelings. Base on this emotional fluctuation phenomenon, the thesis proposed emotion distribution language model, which analyzes emotion distribution differences between adjacent time periods to detect the hot topics in the micro-blog platform. Experimental results show that this method can effectively detect the hot topic from the micro-blog platform, and it is helpful to manage and monitor of the hot topic in micro-blog platform.(4). In order to summarize the information flow in the social media platform, this thesis, taking Sina micro-blog as the research object, proposes a new method based on the weights quantization to generate hot topic summary. The method combines micro-blog information entropy with the importance of the topic and a specific micro-blog. Then it uses highest forwarding micro-blog as center to generate topic summary, and also takes the proportion of stop words and edit distance between highest forwarding micro-blog into account to enhance the readability and to reduce information redundancy. Experimental results further show that the method can effectively generate the hot topics summary in the micro-blog platform, which can be used for the summarization and regulation of hot topics in micro-blog platform.