Research on the Fairness of Cold-start Recommender System Based on Federated Learning Framework
Yuqi Wang, Xiaojun Tang, Ying Lu, Liu Na · 2023
The cold-start problem, in which the embedding for the same item differs for cold-start and non-cold-start due to the unfairness of recommendations, has been a more prevalent challenge in recommender systems in recent years. In this paper, we propose a new model, named FLM, which combines distribution generator and autoencoder to form a joint learning framework based on the concepts of fairness of opportunity and maximum-minimum fairness. Finally, we test the model on the public dataset Movielens 1M, and the experimental results demonstrate that it compares favorably with other models. in four metrics: mean discounted gain (MDG), recall, precision, and normalized discounted cumulative gain (NDCG).