A Comparative Analysis of Various Approaches for Incorporating Contextual Information into Recommender Systems

Recommender systems are being widely applied in many fields, such as e-commerce, e-documents, places and travel, multimedia, news and advertising and transportation. These systems are similar to an information filtering system that helps to identify a set of items that best satisfy the users’ demands based on their preference profiles. The integration of contextual information (e.g., location, weather conditions and user mood) into recommender systems to improve their performance has recently received considerable attention in the research literature. Studies in the relevant literature have focused on incorporating contextual information into conventional recommender systems by employing three approaches: Pre-filtering, post-filtering and modeling. In this paper, we conduct a systematic comparison of various approaches and show how to integrate contextual information into recommender systems. Additionally, we provide an in-depth analysis of the most notable studies to date and point out the strengths, weaknesses and application scenarios for each of the approaches. We also empirically evaluate the real-world datasets, analyzing distinct recommendation quality metrics and characteristics of the datasets. An important result is that accuracy-based comparisons show no clear winner among the approaches.