Scalable Hierarchical Collaborative Filtering Using BSP Trees

Abstract

Collaborative Filtering (CF) technique is used by most of the Recommender Systems (RS) for formulating suggestions of item relevant to users’ interest. However, CF algorithms using the large dataset sometimes become very expensive as the similarity computation among $n$ users is an $O(n^{2} )$ process. In this work, we propose a decomposition-based recommendation algorithm using Binary Space Partitioning (BSP) trees. We divide the entire users’ space into smaller regions based on the location, and then apply the recommendation algorithm separately to each of the regions. Our proposed system recommends item to a user in a specific region only using the rating data of that particular region. This reduces the quadratic complexity of the CF process as we avoid the similarity computation over the entire data. The primary goal of our work is to reduce the running time as well as maintain a good recommendation quality. This ensures scalability, allowing us to tackle bigger datasets. Empirical evaluation of our approach on the MovieLens dataset demonstrates that our method is effective while reducing the running time.