X-FedAvg: An Explainable approach to FedAvg in Link Prediction
Mridula Dwivedi, Babita Pandey, Vipin Saxena · 2024
The extensive accessibility of data in RS has propelled the use of FL. This approach facilitates collaborative model training across distributed data sources while upholding data privacy. This article proposes a novel approach X-FedAvg that incorporates FedAvg with XAI within the domain of LP in RSs. FedAvg employs three fundamental models for Federated Learning: FedGNN, FedMF, and FedCF. Furthermore, we have employed LIME as a means of clarifying predictions. The primary objective of this study is to investigate the extent to which FL is applicable in the LP field, as well as XAI. The experiment conducted using synthetic datasets shows that the proposed method beats FedGNN, FedMF, and FedCF by producing a relatively lower RMSE value of approximately 1.1576 and a relatively higher MAP value of approximately 0.4676.