Research And Implementation Of Network Public Opinion Integrating Topic Model And Sentiment Analysis

With the development of the Internet,network platform has become the main way for people to track hot topics and obtain information.Internet public opinion is the opinions and emotions expressed by the public on issues and phenomena in the real society on the Internet platform.Identifying public opinion information such as topics and emotions in the online platform can help public opinion regulators grasp the trend of public opinion.However,there are still some problems in the public opinion analysis:the traditional public opinion analysis methods mostly use a single semantic feature to extract topics,and the feature information extraction of the document is not enough,resulting in low performance of topic detection;In addition,some public opinion analysis methods only focuses on classifying the sentiment of public opinion,fails to analyze the content of public opinion with emotion,neglects to explore the reasons for the generation of public opinion emotion,and lacks a complete understanding of public opinion emotion.For the problems,this paper proposes a specific improvement plan.The main work is as follows:(1)Topic detection algorithm TDTE(Topic Detection algorithm based on Theme Enhancement,TDTE)is proposed.Aiming at the problem of insufficient text feature information extracted by topic detection method,on the basis of semantic features,TDTE algorithm extracts the theme features of the document through the topic model.And a dual-channel feature fusion method is designed to fuse the two features to fully extract the feature information of the document;Then,according to the feature similarity between documents,K-means algorithm performs clustering to obtain topic clusters.(2)The BiLSTM-com-Att(Bidirectional Long Short-Term Memory combined with Attention,BiLSTM-com-Att)model is proposed to judge sentiment polarity of public opinion texts.The model uses the Bi-LSTM network to encode the document.Then the model assigns different weights to different words combined with the attention mechanism,so that the model pays attention to the key information and improves the accuracy of the model in judging the emotional polarity of the document.(3)The ECPFE(Emotion-Cause Pair Feature Extraction,ECPFE)model is proposed,which enhances the learning ability of the model by constructing specific sentence pair representations.The ECPFE model can extract emotional clauses and cause clauses from public opinion,and express the emotion in public opinion and the cause of emotion.This kind of clause pair method explains the generation of public opinion emotion,deeply analyzes public opinion information,and then fully understands public opinion.(4)Based on the above research,this paper designs and implements a public opinion analysis system.The system analyzes multi-source data,extracts topics discussed by the public,analyzes emotional performance from two dimensions of emotional polarity and the causes for the emotion,and visualizes the analysis results of public opinion.This paper verifies the effectiveness of the proposed algorithm model by designing comparative experiments and choosing appropriate evaluation indicators.By the results of public opinion analysis in detail,the feasibility of the public opinion analysis system is demonstrated.