An Adaptive Recommendation System in Social Media

Abstract

In a broader sense, news recommendation essentially is to select relevant news by their themes. Identification of topical patterns is critical in this task. Common strategies in the previous studies rely on news entities to extract topic patterns. In such a way, news is recommended solely based on the author's point of view. In this article, we argue that, in social media, the performance of recommendation can be immensely enhanced if user interaction is better utilized. It overcomes the bias of traditional news recommendation by suggesting relevant information with a balanced perspective of authors and readers. This is achieved by identifying and using the topic patterns of the original news posting and its comments, one of the most useful records of user behaviors in social media. In particular, to capture the dynamic concerns of users, hidden topic patterns are extracted by utilizing both textual and structural information of comments. To do so, we model the relationship among comments and that relative to the original posting using an undirected a cyclic graph, where each node is a word, each edge is a structural link between words. Experiments indicate that our proposed solution provides an effective news recommendation service in social media.