Design And Implementation Of Network Public Opinion Early Warning System For Enterprise Risk Control

With the continuous development of Internet technology,the analysis of online public opinion of enterprises has attracted much attention.The regulatory authorities of the relevant enterprises can obtain the online public opinion information of the regulated enterprises through the basic information inquiry platform of the enterprises and some more authoritative news release platforms.This system provides intuitive real-time visual analysis results for the relevant enterprise supervision departments,enabling them to make timely,accurate and effective analysis and treatment of the public opinions of the supervised enterprises,so as to minimize the risk of enterprises and reduce economic losses.The network public opinion early warning system for enterprise risk control designed in this paper adopts the design idea of layering and modularization,and at the same time,it extracts the functional module requirements of the system through relevant research and analysis in the early stage.The system mainly includes four subsystems of data acquisition,data cleaning,public opinion analysis and data visualization.The author independently designed and implemented the following four subsystems:(1)Data acquisition subsystem: including crawler configuration,anti-crawler,URL deduplication,data capture,data storage functions,to achieve the crawling and storage tasks of enterprise metadata and related news data.(2)Data cleaning subsystem: the crawling enterprise metadata and related news data are preliminarily processed,and the dirty data is cleaned into clean data by means of missing value processing,error value processing,data deduplication and unified format,and then stored in non-relational database Mongo DB.(3)Public opinion analysis subsystem: after cleaning the news text data in the data as being processed object,through the participle,eliminating stop words,Word2 vec and TF-IDF technology to do text processing for news text,the processed word segmentation results are input into LDA model for topic analysis,and the text vectors are input into SVM and LR model for emotion polarity analysis.Finally,the results of emotion polarity analysis were quantified and compared with the risk threshold.They were divided into different levels of warning states,at the same time,the public opinion alarm was triggered.(4)Data visualization subsystems: this subsystem is designed for the convenience of relevant enterprise supervision department users to intuitively view the enterprise information and various public opinion analysis results,through the word cloud,pie charts,line charts,and other forms of visual graphics the user can analysis enterprise public opinion,at the same time,users can also through the statistical reports to view business related information and system log data.Through relevant testing and deployment,the system has been successfully launched,with stable functions of each module,and it can provide real-time public opinion information of supervised enterprises to relevant enterprise supervision departments,and also provide reference for them to formulate corresponding preventive measures.