Using context-aware collaborative filtering for POI recommendations in mobile guides

Abstract

Mobile guide is one of the most popular Location Based Services. Currently, providing context-aware services/information is still very challenging in mobile guides. Collaborative filtering (CF), known as "Amazon-like recommendations", is a promising solution for providing context-aware recommendations. The paper investigates how context-aware CF (CaCF) can be introduced into mobile guides. Specifically, we focus on applying CaCF methods on the highly available GPS trajectories to enhance visitors with context-aware POI (Point of Interest) recommendations. After analysing the key issues of CaCF, we present a methodology for addressing them. Firstly, a two-stage method is proposed to identify context parameters which are relevant and thus needed to be modelled in a CaCF application. After identifying relevant context parameters, we explore a statistic-based approach (SBA) to measure similarity between different contexts (situations). In considering two different ways of incorporating context information into the CF process, two CaCF methods are designed: SBA-CP-CaCF (using SBA and contextual pre-filtering), and SBA-CM-CaCF (using SBA and contextual modelling). With these CaCF methods, smart services like "in similar contexts, other people similar to you often⋯" can be provided. Finally, the proposed methods are evaluated with some real GPS trajectories collected from Vienna Zoo (Austria). The results of the experiments show that the proposed CaCF methods are feasible and useful for providing context-aware POI recommendations in mobile guides. More importantly, we show that including context information in the CF process can improve the recommendation performance.