Content-Based Filtering Enhanced by Human Visual Attention Applied to Clothing Recommendation

Recommendation systems (RS) are important applications to help consumers to find interesting products in large databases available on line. A wide range of RS can be easily found such as Netflix and Amazon. In this paper, we propose a novel content-based approach for clothing recommendation, termed CRESA, which combines textual attributes, visual features and human visual attention to compose the clothes profile. Traditionally, the RS uses textual and/or visual features to derive the similarity measure between two products. However some parts of the image may call much more attention of users than others. We believe that the characterization of this phenomenon can improve the quality of the recommendation systems, especially when applied to the clothing recommendation problem. With this purpose, in this work we propose weighting the similarity of visual attention between corresponding parts of products with the measures conventionally used in content-based image recommendation systems. The experimental results showed that our approach reached the best accuracy rates when compared to the baseline approaches.