Research And Implementation On Sentiment Computing For Microblog Hot Topics

With the rapid development of the information socialization, social network, for example, microblog, gets rapid development and becomes a platform for sharing public opinion, communication and sentiment expressing. Meanwhile, these big data also have many information with higher values. The sentiment analysis and the opinion target extraction of the hot topics in the micro-blogs are not only important research aspects in natural language processing and text mining, but also can play an important role in understanding and guiding the public opinions. On the basis of the deep learning and the emotional knowledges, this thesis proposes a sentiment classification approach and the opinion object is extracted by combining the lexical rules and the syntactic features. The main contributions are shown as follows:1) In order to expand the scope of emotional information, this thesis constructs the sentimental lexicon on the basis of the corresponding corpus. After the integration of the existing sentimental lexicon resources, Word2 vec is used to further complete the expansion of the sentimental lexicon terms. Finally, this thesis also proposes the polarity intensity of the sentiment terms, and as a result, to lay the foundation for the further works.2) On the basis of the natural language processing, e.g, word segmentation, part of speech tagging, dependency syntactic parsing and semantic role labeling, this thesis proposes a new approach which is on the basis of the combination rules of part of speech and the dependency relationship of syntax, in order to extract the opinion objects and then to construct the extraction rule library so as to identify the opinion objects of different micro-blogs under the same topic.3) On the basis of the emotion knowledge, this thesis describes the sentiment classification. As the micro-blog has some special characteristics, by containing lots of emoticons, this thesis also proposes an emoticons space model. Combined with a variety of linguistic features in the micro-blog posts, the sentimental tendency and intensity of the micro-blog posts have also been analyzed.Experiment results show that the accuracy of the classification method is higher than 85%, and the result of the extraction method has better performance among the 22 groups in NLP&CC public datasets. It shows that the effectiveness of the proposed approach. Finally, some existing problems and further works are also present in the end.