Efficient Processing of Alternating Least Squares on a Single Machine

Abstract

Alternating least squares (ALS) is one of the algorithms widely used in recommendation systems. In this paper, we propose a method to perform ALS on a graph engine on a single machine. We employ our graph engine, RealGraph, to handle big graphs and develop ALS efficiently performed on top of it. Real-world graphs in performing ALS follow the power-law degree distribution, specifying that a few nodes have a lot of edges while a lot of nodes do only a few edges. Prior graph engines do not consider this important characteristic, which slows down their performance. According to our extensive performance evaluation, our ALS running on RealGraph significantly outperforms those on other engines up to 2.5 times.