Research And Implementation Of The Topic And Propaganda Technique Analysis For Chat Group

With the rapid development of QQ,WeChat and other chat tools,chat group,as a means of communication involving many people,generates information quickly,covers a wide range of content,and brings great social impact,which has become an important way to spread propaganda content.It is of great significance to analyze the content of group chat in order to obtain valuable information and avoid the influence of propaganda information.Compared with Weibo,Twitter and other social platforms,chat group has its own unique characteristics,which brings challenges to the analysis of chat group.On the one hand,most messages in chat group are short,which are updated quickly.Moreover,many people participate in chat group,which leads to the entanglement of conversations and makes it difficult to directly obtain valuable information.On the other hand,there is a large amount of propaganda information in chat groups,but there is no propaganda technique dataset for Chinese at present.How to transfer knowledge from English propaganda technique dataset to Chinese chat group becomes the problem to be solved in this paper.In response to the above challenges,this paper first proposes a conversation extraction algorithm based on graph convolutional neural network clustering.The algorithm is divided into two steps.The first step is to divide chat logs into fragments with appropriate granularity according to message frequency.The second step is to use clustering algorithm based on graph convolutional neural network to cluster each fragment.The clustering algorithm extracts semantic and structural representations between messages through the attention mechanism to improve the clustering effect,so as to solve the task of conversation extraction and provide support for the subsequent topic analysis of chat group.Secondly,a cross-lingual propaganda technique detection algorithm based on multi-dimensional feature fusion is proposed.The algorithm transfers the knowledge of the English propaganda technique dataset to the Chinese scene by integrating the semantic representation and syntactic representation features of the text as well as the features of the English text and the Chinese text,and can judge whether a given text is a propaganda type and the propaganda techniques used,so as to complete the task of propaganda technique detection.The experimental results on the task of session extraction and propaganda technique detection show that the above two algorithms have achieved significant improvement.Finally,based on the above two algorithms,this paper designs and implements a topic analysis and propaganda technique detection system of chat group.The system includes topic analysis and propaganda technique detection functions,and displays the analysis results through a visual interface.