Abstract:Social network data can facilitate delivering personalized and relevant content to users, but at the same time, poses challenges for designing effective recommendation algorithms that need to handle user preferences, ratings, comments, and other social information. In this study, we leverage social network information to improve the accuracy and efficiency of recommendation systems. We propose two algorithms: Social Network Regularized Kernel Norm Minimization (SNRKNM) based on low- rank matrix factorization techniques and Fuzzy Radial Basis Function Low-Rank Matrix Factorization (FRLRMF), which can utilize social network information more effectively and efficiently. We evaluate and compare these algorithms with two traditional methods: a matrix factorization-based algorithm (the Stochastic Gradient Descent Algorithm Based on Collaborative Filtering (SGD)) and a social net- work algorithm (Social Network Based Collaborative Filtering Recommendation Algorithm (SNCR)), tested on synthetic datasets and the real-world MovieLens dataset. On the MovieLens dataset, our algorithms achieve a root mean square error (RMSE) of 0.74, significantly outperforming the base- line SGD recommendation algorithm's 1.11, statistically significant at the 0.01 level. Compared to the baseline, our algorithms reduce RMSE by 36.28% and exhibit high scalability and robustness across different dataset sizes and rank parameters. SNRKNM utilizes kernel norm smoothing on the latent factors of users and items, capturing their social similarities and diversities. FRLRMF embeds the fuzzy memberships of users and items into the radial basis function (RBF) kernel, handling uncertainty and noise in social network data. In contrast, the SNCR algorithm relies on collaborative filtering within the social network but has limitations in handling large-scale datasets and sparsity.

Low-Rank and Sparse Matrix Completion for Recommendation

A Novel Social Recommendation Method Fusing User’s Social Status and Homophily Based on Matrix Factorization Techniques

A two-step Recommendation Algorithm via Iterative Local Least Squares

Low-Rank and Sparse Matrix Factorization for Scientific Paper Recommendation in Heterogeneous Network

A Survey of Matrix Completion Methods for Recommendation Systems

Adaptive Matrix Completion for the Users and the Items in Tail

Matrix Completion Based on Low-Rank and Local Features Applied to Images Recovery and Recommendation Systems.

A Fast Matrix-Completion-Based Approach for Recommendation Systems

Top-N Recommender System via Matrix Completion

ON SOLUTIONS OF MISSING VALUES IN RECOMMENDATION SYSTEM

Social network recommendation based on low-rank matrix decomposition

Localized matrix factorization for recommendation based on matrix block diagonal forms

RP-LGMC: Rating prediction based on local and global information with matrix clustering

Robust Rank-One Matrix Completion with Rank Estimation.

Social Recommendation Using Low-Rank Semidefinite Program.

Low-Complexity Decentralized Recommendation System With Similarity Constraints

Probabilistic Local Matrix Factorization Based On User Reviews

Incorporating Social Network and User’s Preference in Matrix Factorization for Recommendation

Low-Rank Matrix Recovery Via Efficient Schatten P-Norm Minimization

Improving Top-N Recommendation Performance Using Missing Data

Social Recommendation With Local Low Rank Matrix Approximation.