Interest Prediction on Multinomial, Time-Evolving Social Graphs
Nozomi Nori, Danushka Bollegala and Mitsuru Ishizuka
We propose a method to predict users’ interests in social media, using time-evolving, multinomial relational data. People’s actions in social media such as Twitter, Delicious, Tumblr and Facebook have two fundamental properties. (a) High-dimensional or multinomial relations - e.g. referring URLs, bookmarking and tagging, clicking a favorite button on a post and etc. (b) Time-varying and user-specific nature: each user has unique preferences that change over time. Consequently, it is appropriate to capture each individual’s action at some instant as multinomial relational data. We propose ActionGraph, a novel graph representation for modeling users’ multinomial, time-evolving actions. Each individual’s action at some instant is represented by an action node. ActionGraph is a bipartite graph whose edges connect an action node to its corresponding object nodes. Using real-world social media data, we justify the proposed graph structure. Our experimental results show that the proposed method significantly outperforms an LDA-based state-of-the-art method and a bipartite graph baseline in prediction tasks.