| With the rapid development of mobile internet,more and more people like to use online platforms to express their personal views and share life experiences,resulting in increasingly serious issues of online public opinion.Due to the diversity,complexity,and real-time nature of online public opinion,there are currently main problems in monitoring online public opinion,such as poor real-time performance,low accuracy,and being limited to a single event.Sina Weibo,as the most popular social software in China today,is also a disaster zone for online public opinion.There are still many areas for improvement in research on Sina Weibo public opinion monitoring.In order to monitor the public opinion situation on Sina Weibo and improve the performance and accuracy of the public opinion monitoring system,this article proposes a deep learning based Weibo public opinion monitoring model based on the traditional public opinion monitoring system architecture and sentiment analysis methods.This model aims to monitor and analyze the public opinion information of Weibo hot searches,understand the public’s attitude and response to a certain topic or event,and timely monitor and analyze events with negative public opinion.The main research content and contributions of this article are as follows.(1)We have built a distributed data crawling framework based on Java.Because the real-time public opinion monitoring model needs to constantly obtain massive microblog hot search comment data,in order to meet the performance requirements of the model and preprocess the crawled data at the same time,this paper designed and implemented the distributed data crawling framework,which effectively solved the problem of insufficient performance of other crawlers through key components such as crawler task pool,crawler thread pool.(2)A distributed search storage method based on Elasticsearch was proposed.Considering that when analyzing and monitoring negative public opinion,it is necessary to obtain all comments related to relevant negative words.The performance of the public opinion monitoring system is relatively poor when stored in My SQL.And this article applies Elasticsearch to network public opinion monitoring for the first time,which greatly improves the system performance when analyzing public opinion,and designs data storage and deletion strategies for this purpose.(3)A Bi-LSTM sentiment analysis model integrating improved attention mechanisms has been established.The emotional level carried by Weibo hot search topics can greatly affect the emotional tendency of user comments,thereby affecting the accuracy of sentiment analysis models.In order to pay more attention to the emotional tendency of users’ own comments,this paper establishes a Bi-LSTM emotion analysis model based on two-way short term memory neural network,which integrates improved attention mechanism.Experiments show that the accuracy of user emotion classification is greatly improved.(4)A method for calculating the level of public opinion warning based on sentiment analysis has been proposed.This article first quantifies the public opinion warning levels for hot topics on Sina Weibo,divided into five levels.Combined with numerous data indicators,the sentiment level and public opinion warning level of hot search topics were calculated successively.Finally,based on the results of public opinion warning levels and the advantages of Elasticsearch storage,a careful analysis was conducted on the generation of negative hot search public opinion,effectively achieving public opinion monitoring and analysis of Sina Weibo hot search. |
