Research And Design Of Sentiment Analysis Of Police-related Network Public Opinion Based On Deep Learning

With the rapid development of new media such as Weibo and live streaming,and the exponential growth of the number of users,the position of social public opinion has gradually shifted to new online media,and a new cyberspace "public opinion field" has gradually formed.Because public security work is closely related to people’s daily life,some inappropriate handling methods can easily arouse the attention of netizens,resulting in police-related online public opinion incidents.At present,in order to detect and deal with public opinions as soon as possible,although the public security organs have adopted certain public opinion monitoring methods,there are problems such as inaccurate analysis of relevant information and insufficient early warning,which lead to the continuous fermentation of individual police-related public opinion incidents,which have negative effects.In view of the above problems,combined with the work of public security,this paper studies the application of the deep learning-based sentiment analysis model in the police-related network public opinion,and designs and implements the police-related network public opinion monitoring system.The main contents are as follows:(1)This article first obtains 7052 Weibo text data of police-related online public opinions through the public opinion secretary platform,and then manually marks the text data with emotional tendencies(positive,neutral,negative)in the form of tags,and integrates them into a suitable dataset for sentiment analysis of police-related network public opinion.In order to avoid the influence of personal subjective cognition on the judgment of emotional tendency,this paper ensures the relative accuracy of artificial emotional annotation through the joint participation of five practitioners in emotional annotation.(2)This paper cites two kinds of TextCNN and LSTM deep learning models that are frequently used in natural language processing and have good analysis effects,and on this basis,the Bert and Attention mechanisms are introduced respectively to form two Bert-TextCNN and LSTM-Attention combined deep learning models.Through the analysis and comparison on the police-related network public opinion data set and the Weibo standard data set,the accuracy rate,recall rate and the harmonic mean F1 value of the two are used as evaluation indicators.Experiments show that the above models have good results in the analysis of police-related network public opinion.Among them,the combined deep learning models BERT-TextCNN and LSTM-Attention have an improvement of 3%～4%on the original model,and finally the one with the highest F1 value is selected.The Bert-TextCNN combined deep learning model is applied to the sentiment analysis module of the police-related network public opinion monitoring system.(3)This paper designs and implements a police-related network public opinion monitoring system.Through the analysis and investigation of the requirements,the core requirements of the system in data mining,data storage and data analysis are determined.Using a distributed and micro-service architecture,the overall system is divided into a public opinion monitoring module,a data processing and analysis module,and a data visualization module which realizes distributed design,implementation,and deployment.Through the design and implementation of the system,it solves the dynamic monitoring of police-related network public opinion,realizes real-time data storage,mining,query,and dynamic tracking of public opinion,and provides functions such as negative information analysis and early warning,which can better improve the current response to police-related public opinions.This can better improve the current analysis and early warning capabilities for dealing with negative information on the police-related network.