An imputation-based matrix factorization method for improving accuracy of collaborative filtering systems. (November 2015)

• Record Type: Journal Article
• Authors: Ranjbar, Manizheh; Moradi, Parham; Azami, Mostafa; Jalili, Mahdi
• Is Part Of: Engineering applications of artificial intelligence. Volume 46:Part A(2015:Oct.)
• Page Start: 58
• Page End: 66
• Publication Date: 2015-11
• Subjects: Matrix factorization -- Collaborative filtering -- Recommender system -- Multiplicative update rules, Impute rates; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2015.08.010
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: Matrix-Factorization (MF) is an accurate and scalable approach for collaborative filtering (CF)-based recommender systems. The performance of matrix MF methods depends on how the system is modeled to mitigate the data sparsity and over-fitting problems. In this paper we aim at improving the performance of MF-based methods through employing imputed ratings of unknown entries. A novel algorithm is proposed based on the classic Multiplicative update rules (MULT), which utilizes imputed ratings to overcome the sparsity problem. Experimental results on three real-world datasets including MovieLens, Jester, and EachMovie reveal the effectiveness of the proposed strategy over state of the art methods. The proposed method is more tolerant against the sparsity of the datasets as compared to other methods including Alternating Least Squares (ALS), Stochastic Gradient Descent (SGD), Regularized Stochastic Gradient Descent (RSGD), Singular Value Decomposition Plus Plus (SVD++) and MULT methods. Highlights: A novel matrix factorization method called IMULT is proposed. IMULT utilizes imputed rating to mitigate the sparsity problem. The convergence proof of IMULT update rules is provided. The results show that IMULT performed better than ALS, SGD, RSGD, SVD++ and MULT.