Uso de taxonomias na recomendação de produtos

Abstract

In this work we investigate how to take advantage of valuable information encoded in taxonomies to improve the quality of content-based recommender systems. The use of taxonomies opens the opportunity to incorporate domain-specific and common-senseknowledge compiled by humans. Our investigation is based on a case study over the book domain, in which the recommendation target is a news web page and the items to be recommended are books from an online bookstore. This is a representative real-case application that provides an adequate context to experiment with a number of distinctstrategies. We present a comprehensive study of three strategies to exploit the use of taxonomies in content-based recommender systems. Although some of these strategies have been previously applied to other related IR problems, in this paper we present a fresh perspective for their application in a new scenario. For this, we implement several methods for content-based recommendation that apply these strategie individuallyand in combination, exploring diverse configurations and premises. We perform a comprehensive set of experiments with a collection of 100 news pages (i.e., target pages) from The New York Times, and a collection and a taxonomy of books crawled from Amazon.com. Experimental results indicate that our strategies can be successfullyapplied to improving traditional content-based recommender systems. In particular, when the target page is manually assigned to a category by a user, we obtain gains close to 20% in average precision. On the other hand, if such an assignment is automatic, the gains are still representative, reaching around 13% in average precision