Modeling Implicit Trust in Matrix Factorization-Based Collaborative Filtering

Yuan Yuan,Yang Yang,Zahir Zahir

doi:10.3390/app9204378

Yuan Yuan, Yang Yang + Show 1 more

Open Access

https://doi.org/10.3390/app9204378

Copy DOI

Abstract

Recommendation systems often use side information to both alleviate problems, such as the cold start problem and data sparsity, and increase prediction accuracy. One such piece of side information, which has been widely investigated in addressing such challenges, is trust. However, the difficulty in obtaining explicit relationship data has led researchers to infer trust values from other means such as the user-to-item relationship. This paper proposes a model to improve prediction accuracy by applying the trust relationship between the user and item ratings. Two approaches to implement trust into prediction are proposed: one involves the use of estimated trust, and the other involves the initial trust. The efficiency of the proposed method is verified by comparing the obtained results with four well-known methods, including the state-of-the-art deep learning-based method of neural graph collaborative filtering (NGCF). The experimental results demonstrate that the proposed method performs significantly better than the NGCF, and the three other matrix factorization methods, namely, the singular value decomposition (SVD), SVD++, and the social matrix factorization (SocialMF).

Highlights

Recommendation systems help users overcome the problem of information overload by providing personalized recommendations
collaborative filtering (CF) can be broadly categorized into memory-based methods, in which the entire rating history is used to determine the relationship among users or items, and model-based methods, in which a model is constructed using the users’ rating matrix
A common approach is to use the Pearson correlation coefficient (PCC) of the co-rated items as trust

Summary

Introduction

Recommendation systems help users overcome the problem of information overload by providing personalized recommendations. The concept behind CF is that users with similar behaviors interact with the same items, and tend to provide similar ratings. CF can be broadly categorized into memory-based methods, in which the entire rating history is used to determine the relationship among users or items, and model-based methods, in which a model is constructed using the users’ rating matrix. The simplest form of memory-based CF involves calculating the similarity between users and making predictions using the similarity of the k-nearest neighbors. Similarity-based methods are simple, they suffer several limitations including the cold start problem, in which not enough user ratings are available for prediction, data sparsity, and scalability. Due to the scalability issues associated with memory-based methods, many large-scale systems in industrial settings use model-based methods to provide recommendations [5,6]

Objectives

Results

Discussion

Conclusion