Research On Public Sentiment Analysis Of Topics In Micro-blog

With the rise and rapid development of Web2.0, a large number of social media sites represented by micro-blog have appeared on the Internet. Micro-blog has features of convenient publishing, rapid propagation, and powerful interaction. It has become a significant platform for obtaining information and exchanging sentiment. The message in micro-blog is usually spread fast and has a wide influence on the society, which makes it convenient for netizen to achieve and share information. However, it also provides a way for the hostile forces and outlaws to spread rumors and trigger public negative emotion. Therefore, effective analysis on public sentiment will greatly contribute to understand public sentiment orientation and make efficient decisions for government departments, which simultaneously has an important significance for public opinion monitoring and guidance. This thesis makes deep research on public sentiment analysis of topics in micro-blog, including micro-blog topics tracking, sentiment analysis of micro-blog, and public sentiment analysis of micro-blog. The main contributions are listed as follows:(1) Traditional methods for micro-blog topics tracking often ignore the semantic information between features, which result in unsatisfactory results. To solve this problem, a method based on word vector for micro-blog topic tracking is proposed. Firstly, the Neural network language model is trained to get the word vector by using a large dataset. Secondly, the semantic information was expanded of eigenvector by using the word vector, and the Fuzzy Sets models for initial topic and micro-blogs are constructed. Finally, the semantic similarities between micro-blogs Fuzzy Sets and the initial topic Fuzzy Sets are computed, and the task of topic tracking is completed according to the decision of pre-defined threshold. Experiments are conducted on ten hot micro-blog topics and the comprehensive F1-Measure improved by 5% than that of the traditional methods reaches 85.71%, which show that the proposed method can make full use of semantic information contained by word vector, which can effectively improve the tracking performance compared with traditional methods.(2) Traditional unsupervised methods of sentiment analysis of micro-blog fail to resolve the problem of feature sparsity of micro-blog corpus, which turns in poor performance. To solve this problem, an unsupervised method based on Biterm Topic Model for sentiment analysis of micro-blog is proposed. Firstly, the co-occurrence words pairs are counted and the BTM model is used to mine the implicit topics in the documents. Secondly, a sentiment dictionary is used to calculate the sentiment distributions of the topics. Finally, the sentiment orientation of the whole micro-blog is obtained on the basis of combined with the sentiment distributions of the topics and the topic distributions of documents. Experiments are conducted on the NLP&CC2012 corpus with the average F1-Measure improved by 15% than that of the traditional methods reaches 75.88%, which show that the proposed method can effectively solve the problem of feature sparsity of micro-blog corpus and accurately identify the sentiment orientation of micro-blog in unsupervised conditions.(3) Existing researches usually ignore or can not accurately describe and analyse the public sentiment, which makes it difficult to meet the needs of micro-blog public opinion monitoring and efficient guidance. To solve this problem, an effective method for micro-blog public sentiment analysis is proposed. Firstly, extract the sentiment summarization of micro-blog topics and describe the public sentiment. Secondly, analyse the public sentiment with three methods proposed above. Finally, lead the public sentiment by generating guide sentences. Experiments are conducted on micro-blog topic corpus and the comprehensive F1-Measure improved by 11% than that of the traditional methods reaches 54.95%, which show that the proposed method can not only effectively improve the comprehensive performance of micro-blog sentiment summarization, but also accurately obtain the public sentiment orientation and effectively lead the public sentiment.