Research On The Analytical Method Of Public Opinion Facing Hot Topic

Hot topic refers to the majority netizens noticed topics caused by the news on the Internet,the research on public opinion evolution of hot topics contributes to the detection and monitoring of public opinion events,so that the netizens know the current focus of public opinion.In addition,it can help the government to guide the public opinion benignly,prevent the ill-intentioned people from using the extensiveness and convenience of the Internet to lead to the negative public opinion maliciously,thus leading to the social conflicts.At present,the analysis research on public opinion evolution of hot topics has some problems on topic recognition such as the loss of potential semantic information and the low accuracy of clustering.The paper proposes an incremental clustering hot topic evolutionary analysis method based on extended LDA topic model and the feature words and topic relevance strategy.In this method,the traditional LDA topic model is improved for solving the problem of potential semantic loss.The text topic and the body keyword are analyzed to have different degrees of importance on the text topic,the topic distributions of the text and title are respectively calculated.The two distributions are linearly combined to generate the optimal subject distribution.Then,the calculating method of the hybrid similarity is proposed.This method combines the feature words and topic relevance strategy with the extended LDA topic model,improving the problem of topic clustering low accuracy and potential semantic loss.Finally,the evolution of the whole topic from the perspective of subtopics is analyzed,at the same time,introducing the concept of "topic index",thus analyzing the evolution trend of the topic as a whole.Through the designed experiments,the proposed method is compared with the traditional Single-Pass algorithm and the improved Single-Pass algorithm in the literature.The experiments show that the proposed method is optimized for the accuracy and recall of the hot topic recognition and evolution analysis.At the same time,the hybrid similarity calculation method combined with the extended LDA topic model is more effective than the traditional LDA model.Compared with a single calculation using LDA model and a single feature word and topic relevance strategy,the method of using hybrid similarity is more effective on clustering.