Discovery of Topic Derivative Relationship in Social Networks

Detecting social topics and discovering emergencies are necessary for the detection and control of public opinion. One social topic may derive one and more new topics as information spreads in social networks. This paper proposes the concept of derivative topics to describe the trend of topic change in the process of information dissemination, which benefits to discover public opinion and its evolutionary direction. We aggregate the posts into pseudo-documents and construct subgraphs of pseudo-documents with words as nodes. By extracting the topic words to determine whether there is a derivative relationship between documents, and form a visual derivative relationship graph. First, we group the original dataset into time slices and use paragraph2Vec to train each Microblog post as paragraph vectors. Second, we calculate the similarity between the posts in the same group through their paragraph vectors. The posts with high similarity are aggregated into a pseudo-document. Finally, we extract topic words in each pseudo-document and describe the derivation relationship between the topics by constructing the derivative relationship graph. The experimental results show that the concept of derivative topics we proposed has validity. The structure of the graph shows the derivative relationship between derivative topics and makes the derivative relationship visualization.