Evolução de regras de associação para recomendação de produtos em comércio eletrônico

Abstract

E-commerce has been growing rapidly over the past years. Various products, services, and information are constantly offered to millions of internet users. Defining an adequate strategy to offer a product to a customer is the main goal of a recommender system. To do so, the items to be offered have to take into account the interests of each customer. This association of items is a data mining task, more specifically a task called association rule mining. This dissertation investigated the use of bioinspired algorithms, particularly evolutionary and im-mune algorithms, to build associations among items of a database. Three sets of experiments were performed: an investigation into the influence of different selection and crossover mech-anisms in an evolutionary algorithm for association rule mining; the use of a probabilistic selection in the immune algorithm; and a comparison of the bioinspired algorithms with the standard deterministic algorithm called Apriori. The data bases for comparison were taken from real e-commerce applications. The results allowed the identification of a suitable combi-nation of the selection and crossover mechanisms for the evolutionary algorithm, and to iden-tify the strengths and weaknesses of all approaches when applied to real-world recommender systems.