Research And Implementation Of Analysis For Hot Events Of Public Opinion Based On Multi-source Data

With the rapid development of online community and social media,people are more keen to share and spread what they see and hear in the Internet world.In fact,it has strong timeliness,large data,fast communication and scattered source,and the composition structure is more complex,usually composed of posts,comments,replies,reposts,etc.Traditional methods use the separated repeated event analysis for different sources of data,because of the lack of overall perspective,the accuracy is low and coverage is incomplete.How to discover and master the evolution process of hot public opinion events in multi-source massive data is one of the hot research hotspots.In view of the above problems,combined with the text characteristics of network community and social media,this paper proposes a multi-source data fusion hot public opinion event analysis technology,and conducts experiments and verification for campus public opinion hot events,as follows:Firstly,the event detection and tracking model based on text chain are proposed for the problem of sparse multi-source text features.Combining the multi-level master-slave relationship of text structure,the text chain is constructed,and the multi-source event detection method is improved to improve the accuracy of event detection;the sentiment characteristics of text are combined and the event tracking method based on sentiment time sequence is adopted to enhance the sensitivity of accurately detecting event changes.Secondly,to solve the problem that the single dimension can not accurately describe the event trend,this paper proposes a method of event popularity calculation based on text popularity,content sensitivity,sentiment fluctuation value and user participation degree,and trains neural network to predict the event popularity.And based on text social attention,text representation and text generalization,extract the summary,and get the accurate abstract of each stage of the event.Thirdly,this paper designs and implements the public opinion event analysis system for campus hot spots,including event detection,event tracking,event popularity prediction,abstract extraction for all-life event analysis.In the experiment,310 thousand texts including text relationship were collected and 60 thousand texts in the event frequency period were selected to construct event set.The experimental results show that compared with the traditional word relation graph model,the F value of the event detection model in this paper is significantly improved,and the event tracking method based on sentiment time series is more sensitive to event changes;the event popularity prediction model of this paper has higher fitting degree than recent thermal prediction algorithm;compared with MMR and textrank,the F value of the abstract extraction model is improved effectively.In conclusion,the methods of event detection,event tracking,event popularity prediction and abstract extraction proposed in this paper can find hot public opinion events in online community and social media more effectively,and analyze the trend of events more accurately and form a concise abstract.