Introducing Contextual Information in an Ensemble Recommendation System for Fashion Domains.

In online marketing environments, we have seen strong growth in the fashion domain, allowing consumers to access a worldwide network of brands. Despite the significant advances of the so-called Recommender Systems in more traditional scenarios, they still fail to offer a personalized and reliable fashion shopping experience that allows customers to discover products that suit their style and products that complement their choices or challenge them with new ideas. In this work, we propose a new ensemble recommendation system that combines different context information (customer-product interaction, item characteristics and user behaviour) with the predictions (recommendations) of different state-of-the-art traditional Recommender Systems to recognize new patterns in user-item interaction and to ensure a desirable level of personalization for fashion domains. Specifically, in the present work, we present a first instantiation that combines a collaborative filtering neural network method, a non-customized classical method and domain context information. In our experimental evaluation, considering two Amazon data collections, the instantiation of our proposal presented significant gains of up to 80% of MRR, 70% of NDCG and 108% of Hits compared with the methods considered state-of-the-art for the fashion recommendation scenario.