Multi-criteria recommender system based on social relationships and criteria preferences

Multi-criteria recommender systems have garnered considerable interests from researchers and practitioners. In this paper, we study the optimization of the accuracy and scalability of multi-criteria recommendation systems using social relationships and criteria preferences information. We firstly construct a hybrid social recommendation algorithm to investigate the advantages of social relationships, and extend the application scope of the algorithm by an implicit social relationship inference technique. Then the nonlinear aggregate functions are adopted to uncover the relationship between criteria and the overall rating. Besides, we cluster users and train the aggregate function for each user group with a much smaller sample size, which is useful for improving the training efficiency. Finally, we validate the proposed approaches on TripAdvisor multi-criteria rating data sets with different sparsity. The proposed social recommendation model outperforms traditional approaches for both active and cold start users in predicting criteria ratings. Multi-criteria ratings enhance accuracy on the condition that criteria ratings can be accurately predicted. Our results also confirm the benefits from nonlinear aggregate functions and cluster analysis, especially when the data set is extremely sparse.

Automated summary

The paper investigates the optimization of multi-criteria recommender systems, aiming to improve accuracy and scalability using social relationships and criteria preferences information. Results show that utilizing a social recommendation model can enhance accuracy, particularly in sparse datasets.