Exploiting context information to improve the precision of recommendation systems in retailing

Abstract

In the retailing industry, recommendation systems analyze historical purchasing information with the purpose of predicting user product preferences. Nevertheless, despite the increasing use of these applications, their results still lack precision with respect to the real needs and preferences of customers. This is in part because the user’s purchase history is insufficient to identify the products that a user would need to buy, given that user preferences are highly affected by changes in contextual situations (e.g., geographical location, special dates, activities of interest) over time. This paper presents a recommendation system that exploits context information to improve the precision of recommendations. Our system relies on the collaborative filtering approach, and the post-filtering paradigm as the mechanism to include context information into the recommendation algorithm. We tested our system using data provided by a Colombian retailing company finding that our recommendations are successful for a greater number of customers, compared to the baseline approach.