Improving the Scalability of ALS-based Large Recommender Systems with Similar User Index

Alternating Least Squares (ALS) is popular method to compute matrix factorization in the parallel way. However, due to the time complexity in predicting user’s preference, ALS is not scalable to large-scale datasets. In this paper, we propose a similar user index-based parallel matrix factorization approach. Since the group of similar users is indexed in advance, there is no need to compute similarities between all users in datasets. Furthermore, the size of a matrix is reduced because the matrix is only composed of indexed user’s ratings and items. The current advanced cloud computing including Hadoop, MapReduce and Amazon EC2 are employed to implement the proposed approaches. We empirically show that the use of similar user index resolves the scalable issue of ALS and improves the performance of large scale recommender systems in distributed computing environment.