Technology Research And Implement On Analysis Of Network Public Opinion Tendency

As the rapid development of Internet, the network media has more important influence in many areas, so analysis of networks public opinion is extremely important. In the existing network public opinion research, acquisition the network information, retrieval, clustering and other aspects of the basic technology have matured and applied in many aspects, but we often overlook users comment on the emotional factors tendency. We analyze public opinion from the perspective of finding hot topics, while research on the distinction based on characteristics of emotional tendency is not deep enough.At the current stage, bottlenecks of the network tendency of public opinion are:(1)As the method based on semantic pattern analysis is limited in natural language processing technology, so it is hard to get practical.(2)The orientation analysis based on machine learning methods in turn depends on the size and quality of the training set and has strong field dependence. So if the knowledge it acquired is not enough in this field, it will decrease the accuracy or increase the complexity.Response to these problems, we in-depth study theory and techniques of analyzing cross-domain public opinion tendency, including the transfer learning, Markov logic networks, uncertain large hypergraph etc.. This paper proposes method of cross-domain network public opinion tendency, which is based on multi-task transfer learning based on Markov logic network, design and implement analysis system of public opinion tendency. This system can make full use of existing knowledge in other areas, and it transfer knowledge based on Markov logic network. Therefore, it can effectively avoid the bottlenecks which feature can not be transferred caused by the different structure. So the system can transfer knowledge from other domains to target domain successfully, and improve analysis accuracy of public opinion domain. We try to use multi-task transfer learning based on Markov logic network to analyze the cross-domain. We establish a model of crossing domains network public opinion tendency, and the model can analyze the tendency rightly and efficiently. Last, we realize the proposed system, and have in-depth analysis of the experimental results. By comparison with other methods, it is proved that our system can make a more accurate analysis of network public opinion and has good noise immunity.