An Improved Matrix Approximation for Recommender Systems Based on Context-Information and Transfer Learning

Man Li,Luosheng Wen

doi:10.1109/access.2019.2937896

Man Li, Luosheng Wen

Open Access

https://doi.org/10.1109/access.2019.2937896

Copy DOI

Abstract

As we all known, transfer learning is an effective way to alleviate the sparsity problem in recommender systems by transferring the shared knowledge cross multiple related domains. However, additional related domain is not always available, and auxiliary data may be noisy and this leads to negative transfer. In this paper, we suppose that different parts of one domain also have the shared knowledge and put forward a novel in-domain collaborative filtering framework, which utilizes contextual information to divide an original user-item interaction matrix into some smaller sub-matrices and regards the selected sub-matrices as “multiple domains” to establish transfer learning. The proposed framework no longer needs additional domain information and has a better adaptive ability. Also, considering more actual situation that users may have multiple personalities and items may have diverse attributes, we resort to Rating-Matrix Generative Model (RMGM) to generate the shared cluster-level rating pattern. Experiments on dataset Douban with three different categories demonstrate that the proposed framework can improve the prediction accuracy as well as the top-N recommendation performance.

Highlights

In this fast developing world, the information on the internet is increasing explosively and people’s desire for personalized recommendation service promotes the tremendous development of recommender systems.As the most effective technique in recommender systems, collaborative filtering (CF) [1]–[3] has been applied in many fields
The main contributions of this paper are: (1) Supposing that different parts of one domain have the shared rating pattern, we propose a novel in-domain collaborative filtering framework to achieve extracting the shared rating pattern from different parts of one domain and further transfer it to other parts of the domain
3) SoCo SoCo [19] is a context-aware recommendation model, which partitions the original rating matrix into sub-matrices based on the contextual information firstly and utilizes social PMF method to fill in the missing values in each submatrix

Summary

INTRODUCTION

In this fast developing world, the information on the internet is increasing explosively and people’s desire for personalized recommendation service promotes the tremendous development of recommender systems. Li et al put forward a novel context-aware social matrix factorization method named CSIT [22], which mainly utilized GMM (gaussian mixture model)-based context-aware enhanced model to perform user-item subgrouping and predicted missing ratings by applying their proposed social matrix factorization method to the generated sub-matrices These models can utilize contextual information to improve the recommendation performance, the implicit relationships between all obtained subgroups are ignored. In addition to above models, Pan et al took into account the data heterogeneity and established a principled matrix factorization based transfer learning framework named CST (Coordinate System Transfer) [29], which constructed the coordinate systems for users and items by applying sparse matrix tri-factorization on the auxiliary data and utilized novel regularization technique to adapt the coordinate systems for modeling target domain data This algorithm shows the effectiveness of incorporating two-sided auxiliary knowledge (i.e. user and item side information) into target domain to alleviate the data sparsity problem.

THE PROPOSED FRAMEWORK

EXPERIMENTAL SETTINGS

EXPERIMENTAL RESULTS

CONCLUSION