Internet Public Opinion Analysis For Short Text

With the development of Internet technology, new formal interactive modes of communication such as micro blog provide new platforms for information publishing and sharing. However, spreading of malicious information in Internet has influenced seriously the information catching. How to monitor the objectionable information in micro blog, is a problem demanding prompt solution.This thesis focuses on the public opinion analysis. An algorithm that computes micro blogs’similarities based on nouns’semantics is proposed. First extracts all nouns from micro blogs. Then computes the distance of nouns based on primitive tree of "Hownet" dictionary. Finally finds out the five most similar pairs of nouns and compute the similarity of micro blogs. Based on the above similarity algorithm, an improved dynamic clustering algorithm based on immune is used for clustering in order to make the antibody set from every class of clustering, and the results can describe the class accurately and adapt the increasingly data set. During public opinion analysis, the topic information is given by the antibody set, and new topic can be found by the increasing clustering algorithm.The experimental result on a middle size of manual collected micro blogs shows that the similarity algorithm can get the subject accurately, and more than90%of the similarities of micro blogs from the same class are distributed in the range between0.6and1, which can help the work of clustering. The experimental result also shows that the improved clustering algorithm can cluster effectively, and help for topic detection and alarm. The analysis of clustering result can also help for topic analysis and understanding.