Forum User Profiling by Incorporating User Behavior and Social Network Connections

With the rapid development of social media in recent years, user profile inferring becomes crucial to many practical application such as recommendation and customized service. In this paper, we propose an ensemble learning based model, which incorporates user behavior embedding and social network connection embedding, for user profile inference. In which, post content features and user behavior statistics are employed to learn the user behavior embedding. LINE and PUHE are incorporated to learn the user social network connection embedding. The proposed method is evaluated on SMP CUP 2017 user profiling competition dataset. The experiment results demonstrate that leveraging both user behavior embedding and social network connection embedding improves the user profiling efficiently.