Uma abordagem centrada na filtragem colaborativa para redução do custo computacional do método k-Nearest Neighbors.

Abstract

With the emergence of Web 2.0 the volume of information available on the Internet has grown dramatically, becoming increasingly difficult for the user to achieve the desired information. Recommendation systems emerge as an alternative to this problem, suggesting personalized content. Collaborative filtering is one of the most effective approaches in the area of recommendation. Among the collaborative algorithms, latent factors models are the state of the art in the area. However, such models can not provide a justification for the recommended item, which in some areas can make the recommendation uninteresting and easily ignored by the target user. In this context, an interesting alternative is the k-Nearest Neighbors ( kNN ). a simple, popular and very robust method. This technique generates recommendations from ratings of the most similar users (nearest neighbors) to the target user. Despite its efficiency, kNN has a high computational cost when executed over large databases, making its application impractical in some domains. In this work we aim to improve the performance of kNN from the restriction of the search space of the nearest neighbors. The proposed method uses a user heuristic selection based on the choice of most rated items. As a result it was found that using only 15% of the neighbors searching space, it was possible to significantly reduce the computational cost, while maintaining high accuracy level.