Research On Textual Sentiment Orientation Analysis Of In Network Public Opinion

In general, people tend to be two sides of the emotional things, that is, positive and negative, or commendatory and derogatory. Thus, the usual text sentiment orientation classification is a binary classification problem that eventually will be divided into positive and negative. Text tendency classification is a novel and important research direction, it have commercial and research value, and can be applied to specific network public opinion analysis, product evaluation, information filtering, information filtering, terms of product recommendation and intelligent search.In this paper, the building of Classification model that based Gaussian process as the main line of the text tendency classify, In addition, we also made some other operations, such as emotion corpus acquisition, preprocessing of the text, the emotional dictionary construction, feature extraction, feature vectors as well as key issues to build weight calculation, normalization, etc. we put forward some new ideas and approaches, and verified by experiment. The main research work includes: 1.Use the DOM tree model, analyze the structure of Taobao, Jingdong, Suning and Gome site, we design a multi-threaded crawler, this program can automatically capture people’s evaluation of conditioning products from these four sites. Ultimately collected about 150 million words of text from the Internet as a corpus. The corpus has a reliable source, obviously emotional characteristics, It has certain significance for public opinion research network text emotional tendencies. And It is carried out by the Chinese Word Segmentation and emotion annotation processing.2.Proposed a method for building domain-specific emotional conditioning comments dictionary, the first,we choice a basis emotion dictionary word as a seed, and then use the PMI algorithm calculate the tendency of emotional word, and select the words that have same emotional tendencies written into emotional lexicon. Finally, we based on the dictionary as a feature selection of the target text sentiment classification.3.Constructed a Gaussian process classification model based on PSO optimization algorithm The model is based on the Gaussian process classifiers which combined PSO optimization algorithm. And we using the previous corpus for training and testing, the results showed that after using PSO optimization, classifier performance has significantly improved.4.Use java programming language implement a java web classification system, which has functions like data import, data aggregation and data reports. The program can be run properly under tomcat server, you can visually see the classification results through the system.