Sentiment Analysis Of Chinese Micro-blog Based On Knowledge Element And Ensemble Learning

Micro-blog is one of the most popular social networking platform, is the freedom to express their views in place, these viewpoints involve product reviews, public opinion, entertainment reviews, at the same time, provides the convenience of easy access for the sentiment analysis of micro-blog data, the sentiment analysis of micro-blog has become the focus of scholars for research. This paper put forward the feature extraction with the help of micro-blog knowledge base, and the use of multiple different classifiers ensemble classifier of micro-blog sentiment analysis task.This paper presents the method to construct the knowledge base of micro-blog, us ing the theory of knowledge element as the representation method of knowledge. Firstl y, construct the initial knowledge element network through accessing to relevant infor mation and collecting data in other news sites, and then through the conditional rando m field by using the method of knowledge element as the initial network based on th e prior knowledge, and automatically extract the knowledge element from micro-blog c orpus, and then through the duplicate removal procedures to obtain the final micro-blo g domain knowledge element, finally store the knowledge in XML format documents i n order to use next. Through the experiment for different domain of micro-blog corpu s, which show the feasibility and effectiveness of our method, so, it can provide a su pport for the sentiment analysis of micro-blog next.The paper also presents the method for sentiment analysis of Micro-blog based o n the ensemble learning, according to the reason for diversification of sources and inv olving so many domain in micro-blog data, we select four kinds of common classifier together to finish the task about sentiment analysis of micro-blog, which integration method has been improved in the basis of the original simple voting method, the Bay esian decision method was introduced to vote, the use of the confusion matrix of eac h classifier is trained as a priori knowledge classification. The problem of sparse featu re of micro-blog make it difficult to sentiment analysis, for this reason, we expand u se the feature of micro-blog based on knowledge base, taking full advantage of micro-blog socializing, increase the feature about micro-blog structure. In order to study the effect of multiple combinations of features and different classifiers of micro-blog senti ment analysis, this paper adopts the "control variables" idea of cross experiment, the r esults show that the proposed expansion of micro-blog characteristics and integrated ap proach is effective and feasible to micro-blog sentiment analysis.