Leveraging User Interest to Improve Thread Recommendation in Online Forum

Abstract

Nowadays thread recommendation is considered to be beneficial to improve the end-user stickiness of an online forum. Given the fact of information overload and the diverse interests of forum users, a recommender system in online forum can satisfy not only forum users' information needs by directing them to what they might be interested in, but also their social needs by connecting them to their friends. Some traditional recommender systems rely on a bipartite graph model to capture users' interests. As an extension, some other content-based methods are proposed to further understand the potential connections between Web users and Web contents. However, due to the prevalence of short and sparse messages in online social media, it is hard for traditional content-based methods to capture Web users' interests. In this paper, we propose a novel graphical model to extract hidden topics from Web contents, cluster Web contents into clusters, and detect users' interests on each cluster. Then we introduce two reran king models which utilize the detected user interest to boost the performance of thread recommendation. Experiment results on a public dataset showed that our proposed methods substantially outperformed the naïve content-based approach. In addition, by testing our approaches with different parameter settings, we observed, to some extent, how forum users' information needs and their social needs interplay to decide which threads they will look for.