User preference dynamics on evolving social networks: learning, modeling and prediction

Abstract

Modeling users’ preferences and needs is one of the most important personalization tasks in information retrieval domain. User preferences are fairly dynamic, since users tend to exploit a wide range of items and modify their tastes accordingly over time. Moreover, all the time users are facing with others’ opinions and suffering social influence. In our research, we investigate the interplay of User Preferences and Social Networks over time. We define what are user preference dynamics and propose a temporal preference model able to describe how user preferences evolve over time through changes on user profiles. As problem solution, we first investigate temporal networks. By modeling a sample of Twitter network as a temporal social network we perceive how nodes evolve in function of centrality metrics and how different is the evolution when considering static vs. temporal networks. Then, we explore the idea of centrality-based node event detection in evolving networks. The goal is to detect at what points in time a node change its behavior significantly. Our proposal is a node event mining model with two different strategies for detecting change points. Finally, we join our findings and proposals so far and perform an experimental evaluation using two datasets from different domains focused on our main goal: the interplay between user preferences and social networks over time. The discovery is that there is a strong correlation between preference change events and centrality-based node events, specially when considering temporal networks. Moreover, closeness centrality is more suitable when correlating preference changes and node events than betweenness. In the end, we build a complete solution for the preference change prediction problem, taking into account the use of node events detection in continuously evolving networks where the time when edges are active are an explicit element of the representation. Our prediction model is able to forecast changes on user preferences with competitive levels of accuracy.