FedRecSys: A Federated Learning-Based Recommender System for Concept Drift Adaptation

Abstract

A new field called deep learning can be used to handle large amounts of multidimensional data in different e-commerce platforms. Big Data refers to this type of information, which is accessible in an unstructured and complicated format with respect to volume, velocity, validity, veracity, variety, and variability. To better meet the needs of customers and win their loyalty, it is necessary to examine the vast amount of hidden information present in the volume of data created by online platforms like e-commerce portals. To address this issue and produce the best recommendations depending on user requirements, numerous Recommendation Engines (RE) have been put forth. One such difficulty in large E-commerce based REs is Concept Drift (CD), which mentions about the continuous changing of data in such platforms. Additionally, the data utilised by these e-platforms is made public in several locations, which compromises the security and private information of the user. “Federated Learning (FL)” is a new level of the Deep Learning Approach that has been proposed to address the issue of user privacy security. Federated Learning improves system accuracy by offering security and privacy preservation for user data. This study addresses the CD problem in REs, the overview of REs, and the suggested hybrid method KalADWIN2 for CD detection. Along with the creation of FedRecSys—a mix of RE and FL (for CD adaption & correct recommendations) using Movielens 1M & Book datasets. The paper also focuses on FL, and its various types. The study goes on to detail with the outcomes and a comparative analysis of the Books and Movielens 1M datasets as FedRecSys. It ends with several FL applications, conclusion, and References.

Keywords

• Artificial intelligence
• recommendation engine
• collaborative filtering
• matrix factorization
• concept drift
• ADWIN2