Chinese community topic classification method based on graph model

Abstract

Community topic classification is the basis of hot topic discovery. Existing graph models ignore the importance of key information to the text when performing text classification and increase the influence of irrelevant data. To address these problems, we propose a community topic classification model DGAT that incorporates key information as well as information about the topic itself. An integrated algorithm is proposed to extract keywords to avoid the problem of inaccurate keyword extraction. Then a composite complex network model containing both topic and keyword nodes is built. Finally, the graph attention mechanism is used to update node features and incorporate semantic-level attention to learn the effect of different graph structures on the current node classification. An example validation on the Qingdao community topic dataset proves the effectiveness of the method and outperforms the baseline models.