Social recommendation models are attaining more and more attention and relevance in the fast-growing world of social media and social relations. Matrix factorization-based methods and probabilistic model-based methods are used in current social recommendation models. These models are not very suitable for real-world online recommendation applications because the algorithms in these models make use of the user ratings provided in the user-item matrices. To overcome such difficulties, an innovative framework called online graph regularized user preference learning (OGRPL) can be used. OGRPL is a blended graph-based model that utilizes both content feature information from the partially observed user-item matrix and the auxiliary content features for each item. According to the paper, “OGRPL incrementally learns the user preference on the content features of the items” from an all-time dynamic dataset of user ratings and recommends the items based on user preferences in an online manner.

This paper discusses a new framework for social recommendation in social media based on user preferences. The main audience is graph theorists and network theorists, particularly those doing research or working in the social networking area.

Zhao et al. introduce an efficient and further developed iterative model called OGRPL-FW, “which utilizes the Frank-Wolfe algorithm, to solve the online optimization problem.” In addition, according to the authors, OGRPL-FW “incorporates both [a] collaborative user-item relationship as well as item content features into [a] unified preference learning process.” They also present the problem of online social recommendation from a different viewpoint and study the online social recommendation problem by incorporating the user-item relationship and item content features into a unified preference learning process. They mathematically formulate the “objective function for the problem ... from the viewpoint of online user preference learning and then present the objective function in the setting of online learning.”

They also conduct extensive experiments on several large-scale datasets from the largest online recommendation communities, Douban and CIAO. In the experiments, they evaluate the quality of rating prediction, and evaluate the performance and compare online and offline recommendation algorithms using the mean absolute error (MAE) and the root-mean square error (RMSE) criteria. The experiments also include a performance study, an efficiency study, and a parameter analysis.

The authors also mention the possibility of exploring a nonlinear user preference learning function as the user model for the problem of online social recommendation. Based on the results of these experiments, the authors claim that the algorithms proposed in this study obtain significantly lower errors than the other popular and widely used “online recommendation methods while receiving the same amount of training data in the online learning process.”

The paper is very well written and the presentation of the findings of experimental studies is excellent. The authors deserve much appreciation for conducting such a nice study.