Cross-Domain Transfer Learning Prediction of COVID-19 Popular Topics Based on Knowledge Graph

Abstract

The significance of research on public opinion monitoring of social network emergencies is becoming increasingly important. As a platform for users to communicate and share information online, social networks are often the source of public opinion about emergencies. Considering the relevance and transmissibility of the same event in different social networks, this paper takes the COVID-19 outbreak as the background and selects the platforms Weibo and TikTok as the research objects. In this paper, first, we use the transfer learning model to apply the knowledge obtained in the source domain of Weibo to the target domain of TikTok. From the perspective of text information, we propose an improved TC-LDA model to measure the similarity between the two domains, including temporal similarity and conceptual similarity, which effectively improves the learning effect of instance transfer and makes up for the problem of insufficient sample data in the target domain. Then, based on the results of transfer learning, we use the improved single-pass incremental clustering algorithm to discover and filter popular topics in streaming data of social networks. Finally, we build a topic knowledge graph using the Neo4j graph database and conduct experiments to predict the evolution of popular topics in new emergencies. Our research results can provide a reference for public opinion monitoring and early warning of emergencies in government departments.