Efficient Low-Rank Matrix Completion Updating Algorithm for Recommender System

Abstract

Recommender systems aim to provide personalized item recommendations to users based on users ratings. However, not all ratings are provided by users, so the rating matrix, which summarizes the user-item interaction data, has many missing values. To fill out these missing values, matrix completion problem is considered. One of the most popular approaches to matrix completion problem is based on low-rank approximation of the rating matrix. To do this, singular value decomposition is required in every iteration, however it is prohibitively expensive when large-scale rating matrices are used. Additionally, recommender systems are frequently updated when a new user or item is added, or existing information is updated. In this paper, we propose a new matrix completion algorithm by recycling the existing information to speed up the computation when a small part of the rating matrix is updated instead of computing the matrix completion from scratch. Experimental results demonstrate that our algorithm is very attractive when the rating matrix is updated frequently by showing that our algorithm has significantly faster execution speed while producing very similar or even better accuracy than the other algorithms.