The Research Of Dynamic Network Public Opinion Warning Based On WEB Mining And Text Analysis

The popularity of the Internet grows fast,and network media is promoted rapidlyas a new media in the masses，especially in the younger generation. Timeliness ofaccess to information is improved than ever,and uncertainty of network publicopinion is also improved greatly.Under the impact of the massiveinformation,people’s minds fluctuate greatly in such background.Cyberspace is easilycarried out as tools by the reactionary forces in the territory and abroad,to spark socialunrest,and make our nation in risk. The war of public opinion launched by reactionaryforces has spread through cyberspace,and make cyberspace a increasingly fiercemain battlefield.So it makes great significance in the network consensus warfare bylaunching the study of internet public opinion to get detection timely and to alarm theinternet public opinion.The research of dynamic network public opinion warning is launched in the wayof Web mining and text analysis in this article.The research includes the acquisitionof the network public opinion,segmentation analysis, clustering and emotionaltendency,so that network public opinion in the specific area can be found out timelyand determine the severity and trends.A model of network public opinion analysis isproposed with web mining as the main method of the analysis of network publicopinion in this article.For characteristics of network public opinion，calculate theemotional tendencies of the web texts by text analysis,so that the basis of networkpublic opinion alert can be provided.The Main Contests are:1.Crawler algorithm.To analyze network public opinion, data on the network isneeded firstly, and crawler plays the role.Topic crawler is suggested with thecombination of Web mining in this article,as the method of network public opiniondata collection. The accuracy of data collection is improved.2.Text segmentation.As the first step of the text analysis,segmentation has itstechnical difficulty: Chinese word segmentation.Comparing the methods of wordsegmentation,and combining the characteristics of the web texts reflecting thenetwork of public opinion,multi-word dictionary is introduced makes theimprovement of the maximum matching word segmentation method,and theaccuracy and efficiency of segmentation is improved. 3.Clustering algorithm.To analyze network public opinion, web text clustering isneeded firstly to classify the web texts primarily and text clustering algorithms arecompared and analyzed in this article.For the common feature of the web textsreflecting the network public opinion,the k-means algorithm is improved by means ofadjunction of flag dictionary,so accuracy and efficiency of the clustering is improved.4.The research of emotional tendencies.To get the emotional tendencies is thefinale conclusion of public opinion analysis,and there are many ways to analyzeemotional tendencies.Naive Bayes Classfier is chosen as a better way after variousmethods of the text emotional tendency are analyzed in this article.Patten Matchingcombines with Naive Bayesian classifier based on property weight is proposed toimprove the efficiencies of emotion classification.