Context Aware Scalable Collaborative Filtering

Abstract

Traditional Collaborative Filtering (CF) based Recommender Systems (RS) typically do not consider the contextual attributes of users or items while making recommendations. However, there are plenty of applications in day to day life, where the choices made by a person may not only depend on his or her earlier preferences, but more on the context. In this work, we incorporate the contextual attributes to cluster the original useritem rating matrix and then apply CF based recommendation algorithms to the clusters independently. It helps to bring down the runtime as we can bypass computations over the entire data and also generate more accurate recommendations as the partitioned clusters hold similar ratings, which produce greater impacts on each other. In order to cluster the rating matrix, we use both centroid based (k-means) and density based (DBSCAN) clustering techniques and finally compare the results. One of the main objectives of our work is to make it more scalable by reducing the runtime without compromising the recommendation quality much. Experiments performed on two publicly available datasets: MovieLens-1M and MovieLens-100K show that our approach is scalable and accurate.