A Derivative Topic Dissemination Model Based on Representation Learning and Topic Relevance
Qian Li, Yunpeng Xiao, Xinming Zhou, Rong Wang, Sirui Duan, Xiang Long Yu · IEEE Transactions on Knowledge and Data Engineering · 2024
In social networks, topics often demonstrate a “fission” trend, where new topics arise from existing ones. Effectively predicting collective behavioral patterns during the dissemination of derivative topics is crucial for public opinion management. Addressing the symbiotic, antagonistic nature of “native-derived” topics, a derivative topic propagation model based on representation learning, topic relevance is proposed herein. First, considering the transition in user interest levels, cognitive accumulation at different evolutionary stages of native-derivative topics, a user content representation method, namely DTR2vec, is introduced, based on topic-related feature associations, for learning user content features. Then, evolutionary game theory is introduced by recognizing the symbiotic, antagonistic nature of “native-derived” topics during their propagation. Moreover, implicit relationships between users are explored, user influence is quantified for learning user structural features. Finally, considering the graph convolutional network’s ability to process non-euclidean structured data, the proposed model integrates user content, structural features to predict user forwarding behavior. Experimental results indicate that the proposed model not only effectively predicts the dissemination trends of derivative topics but also more authentically reflects the association, game relationships between native, derivative topics during their dissemination.