Design of a Recommender System Based on Customer Preferences: A Comparison Between Two Approaches

This work addresses the design of a preference based system that suggests relevant products to customers. It aims at helping them with their purchase decision (on electronic commerce websites). A use case that consists in making spontaneous recommendations to the customers, on the basis of their previous ratings is described. The product considered to illustrate the approach is a comic. This paper is focused on two recommender approaches. The first approach, “the traditional” approach, is based on the collaborative filtering while the second approach, is based on a new proposed algorithm. Collaborative filtering is a technique to making recommendations by matching people with the same preferences (preferential similarity). The second approach which is proposed is a combination of the traditional collaborative filtering and the perceptual similarities approach between customers (perceptual similarity). Perceptive data include emotional, sensory and semantic ratings of the products. The purpose of this paper is to evaluate the performance of the proposed approach and to compare it with the traditional approach. A test procedure is thus implemented. It consists in simulating customers’ behavior according to a set of products, and to compute a performance criterion of the recommender system, measuring the relevance of the proposed products. The performance of the proposed algorithm is compared with that of the traditional one. The results show that the consideration of perceptual assessments of products by customers generally helps in the relevance of the propositions of the system.