Identifying Dynamic User Roles in Online Health Communities

Online Health Communities (OHCs) have become an effective platform for gaining and sharing health-related information during the past two decades. Users on OHCs may act in different roles. Identifying user roles can help improve user engagement and user retention. However, there have been few works to identify user roles in OHCs using user embeddings to consider the temporal and textual information at the same time. This study is to propose an approach to generate dynamic user embeddings considering the semantic features of users’ posts and the temporal evolution of their interests using a coupled recurrent neural network model. Meanwhile, we use the unsupervised t-SNE to visualize the high-dimensional user embeddings, and LDA to generate top-N words for different clusters, in interpreting distinguished roles in OHCs.