A recommender approach based on customer emotions

摘要

On electronic commerce websites, the recommender is typically based on a global evaluation of the product by the customer. This preference information is interesting, but not sufficient. Our approach is based on the assumption of the inevitable role of emotions in product appreciation. Our purpose is to evaluate the performance of the approach based on emotions and to compare it with the traditional approach of unique preference. The first step is to collect data. In the first part of the study, we simulate a database of customer/product/evaluation to test our method and to study the influence of matrix sparsity. The second part of the study is applied to real data. The product considered to illustrate the approach is film, a product for which preferences are mostly guided by emotions. We collect the data via an online questionnaire, given a sparse matrix of evaluation (people have not seen all the films). A use case, which consist in making spontaneous recommendations to customers on the basis of their previous ratings, is described. Considering that the behavior of a customer is not random, the notion of user profiles in terms of preference and emotions is introduced. The proposed recommender is tested by removing preference data and predicting it based on the rest of the database. A performance criterion is calculated, measuring the relevance of the proposed products. The results show that (1) the matrix sparsity does not have a significant influence on the results and robustness of our method and (2) consideration of customers' emotional assessments of products improves recommendation performance. (C) 2019 Elsevier Ltd. All rights reserved.