Research And Implementation Of Network Public Opinion Analysis System Based On Sentiment Classification

At present,the social Internet has become an important way for people to obtain information.Therefor,the management of network public opinion is becoming more and more important.How to distinguish negative network opinion information is the focus of network public opinion management.Due to the efficiency of manual classification is too low,people must use the network public opinion analysis system to analyse a huge amount of network public opinion data.Common network public opinion analysis systems generally use keyword-based network public opinion analysis technology.When processing the Chinese network public opinion information,the classification accuracy is not high due to the impact of Chinese polysemes and text word order on semantics.With the development of natural language processing technology,machine learning algorithms have gradually become a new hot research direction in the field of public opinion analysis,not only by keywords,but also by identifying the emotional tendency of public opinion to monitor public opinion that may cause public opinion crisis.Different machine learning algorithms have different public opinion sentiment classification capabilities,and inappropriate methods not only cannot effectively analyze public opinion,but also increase the difficulty of analysis.Therefore,this paper designs a public opinion sentiment classification model based on the BERT model to classify negative public opinion,and then uses the DBSCAN-based public opinion topic clustering method to extract hot negative public opinion topics,and finally designs and implements a prototype system to analyze the network public opinion.The main work of this thesis is as follows:(1)A BERT-based network public sentiment and emotion tendency classification model is proposed.The BERT model is used to generate the word vectors of network public opinion data,and an appropriate classification model is selected according to the number of model training data sets to classify the sentiment word vectors of the sentiment word vectors,and finally,the negative sentiment public opinion is classified.(2)A clustering method of network public opinion topics based on DBSCAN is proposed.First,preprocess the negative public opinion information: word segmentation and stop words;at the same time,use the TF-IDF algorithm to generate a feature word weight table,use the Simhash algorithm and the feature word weight table to construct the text features of the negative public opinion information;and finally use the DBSCAN algorithm to text Features are clustered,and hot public opinion topics are extracted from the clusters obtained by clustering.(3)A network public opinion analysis prototype system was designed and implemented.From the perspective of engineering application,introduced a series of functions of the system from the model training module,public opinion classification module,public opinion clustering module and public opinion information analysis module.