A General Matrix Factorization Framework for Recommender Systems in Multi-access Edge Computing Network

Due to the growing number of users and items in recommender system, along with the more complex algorithms for precise recommendation, recommender system in broswer/server architecture will consume more computing cost and more service latency. Besides, the bidirectional transmission of broswer/server architecture requires users to upload necessary user information to the cloud servers. The current manner of recommender systems might cause the leakage of the sensitive information and take the risk of privacy issue. To alleviate the two issues, we integrate the multi-access edge computing network into recommender systems to take fully advantage of base stations and user terminals. Particularlly, we pull the tasks of user-profiles and recommendation algorithms out of the cloud servers and put them to base stations and user terminals. The cloud servers will undertake an independent item-profiles task that will not take any information from users. Considering the differences among user terminals, we propose a general matrix factorization framework that can adopt different matrix factorization-based recommender algorithms with one item-profiles. This framework can allow different terminals to take variety algorithms based on their computing abilities. Experiments are conducted on two real world datasets to validate the proposed methods by comparing them with conventional recommendation methods. Experimental results prove the principle that matrix factorization methods within the proposed framework can enhance the recommendation system’s performance in terms of both prediction and recommendation.