Research And Implementation Of Multi-source Public Opinion Hot Topic Evolution Analysis Technology

With the vigorous development of the Internet,a large amount of public opinion information has emerged on social media,online communities,and news websites.By understanding the long and short mixed texts in a large number of online corpora and tracking their evolution,public opinion managers can better discover and track the hotspots of massive public opinion data.Based on the data characteristics of multi-source text,this paper proposes a topic evolution model and a keyword based topic generative model integrating multi-source data.At the same time,combining the characteristics of university public opinion information,this paper designs and implements a multi-source public opinion hot topic evolution analysis system.The specific research work is as follows:Firstly,in response to the issue of uneven text vector features from different information sources,this paper proposes a method for text vector feature homogenization,BLTS(BertSum for Long Text and TYCCL for Short Text),which uses a strategy to define the attributes of the current text,and then processes the text with different attributes separately to achieve the goal of uniform text features.Secondly,in order to address the issue of the inability to simultaneously track the evolution process of multiple topics and the inaccurate determination of the same topic in different time slots,this paper proposes a Cross Time Slice Online GSDMM(CO-GSDMM)based on GSDMM,which utilizes online text clustering research methods,combines cross time slice heat analysis strategies,and short text similarity calculation methods,Obtain cross time slice tracing of the same topic and the evolution process of different topics.At the same time,it was verified through experiments that the model has significant advantages in terms of usability and effectiveness compared to the comparative model.Thirdly,to solve the problem that it is difficult to understand topic semantics from individual words,this paper proposes a topic attention FM encoder for Text Generation(TAFM),which integrates the topic weighted attention mechanism.This model can reduce the influence of the order between keywords and focus more on the semantics between keywords for text generation.The experiment shows that the BLEU algorithm score of this model has improved by at least 2%compared to the comparative model.Finally,this project designs and implements a multi-source public opinion hot topic evolution analysis system,which can achieve data collection from multiple data sources,localized storage of data,discovery of hot topics,generation of topic semantics,and visual display and interaction of topic evolution.