Research On Topic Analysis Method Of Network Public Opinion Based On Text Mining

With the rapid development of the Internet,the network has gradually become the main means for people to obtain information,but it has also brought about an increasing number of network information security issues.Especially in recent years,affected by the epidemic,online public opinion events have frequently occurred,and people pay more and more attention to network public opinion.Therefore,in the face of extremely rich public opinion data on the Internet,how to obtain hot topics and sensitive information related to online public opinion has become an important issue in current research.This paper combines current methods for obtaining and analyzing online public opinion information,conducts in-depth research on topic detection and tracking related technologies,focusing on feature selection,text representation models,text classification algorithms,and similarity measurement.It constructs an efficient online public opinion question analysis method,providing solutions for online public opinion management and monitoring.The main contents of this paper are as follows:1.Propose feature selection and text clustering algorithms based on binary mayfly optimization.To address the issue of low clustering accuracy caused by redundant text features,research has been conducted on feature selection and text clustering methods,and a feature selection algorithm based on binary mayfly optimization has been proposed.The algorithm first uses the inverse document frequency index as the objective function to evaluate the text features,and constructs a vector space model;Secondly,improvements and optimizations were made to the ephemera algorithm,including ephemera position update,mating,and mutation strategies,to further obtain the optimal feature subset;Finally,input the new feature subset into the Kmeans++algorithm to complete text clustering,and obtain the optimal text clustering effect.Through experimental verification on multiple text datasets,this algorithm not only has good clustering performance and optimization ability,but also can significantly reduce feature and convergence time,providing technical support for subsequent topic detection tasks.2.Propose an improved K-nearest neighbor topic tracking algorithm based on the bidirectional quantity model.To solve the problem that the inverse document frequency algorithm in the text representation model lacks Semantic information and the data in the topic is unbalanced,which leads to low tracking efficiency,research on text representation and text classification methods is carried out,and a topic tracking algorithm based on bidirectional quantitative text representation is proposed.Firstly,the algorithm combines the language model with the traditional vector space model to obtain a bidirectional quantitative text representation model;Secondly,combining the distance and direction information of the topic text vector,a multi-objective similarity method is constructed to measure the similarity of the text;Finally,the average similarity of adjacent reports based on the topic is used as the basis for determining the category of tracking reports,replacing the total maximum similarity of texts in the K-nearest neighbor text classification algorithm;Through two datasets validation,this algorithm can effectively improve topic tracking performance and has good classification performance for imbalanced data.3.Complete topic analysis and application verification of online public opinion data.Based on the topic detection and tracking technology proposed in this paper,combined with media platforms such as Weibo,Tieba,forums,and news,practical applications were conducted.Eight topic data were selected for online public opinion topic experiments and analysis to verify the effectiveness of the proposed method.