联邦推荐系统的协同过滤冷启动解决方法

Cold starts in collaborative filtering for federated recommender systems

基于联邦学习的推荐系统可以在保护用户隐私的情况下,联合多方数据,提升推荐系统的性能,已经成为推荐领域的研究热点之一。联邦协同过滤是联邦推荐系统中最经典及最常用的算法之一。然而,针对联邦协同过滤系统的冷启动问题的研究工作相对较少。针对这一问题,本文提出了一种基于安全内积协议的解决方案。具体地,在系统中添加新用户或新物品时,联合多方评分矩阵,利用安全内积的方法,对多方数据进行相似矩阵的求解,从而完成推荐输出。本文在MovieLens数据集上对所述方法进行了验证。结果表明:本方法能够有效解决基于相似度的协同过滤中的冷启动问题,并且推荐效果也会依据多方数据分布的比例变化。

Recommender systems based on federated learning has become one of the research hotspots in the recommender field.However,few studies focused on the cold start problem of federated recommender systems.Under the framework of federated learning,we propose a novel collaborative filtering algorithm for its solution:through involving more rating matrices,we can get a similarity matrix with secure inner product method,and implement the recommendation for new users to the system.In this work,we verify the performance of our method on MovieLens.The results show that our proposal is effective in solving the cold start problem in similarity-based collaborative filtering,and the recommendation effects vary according to the data distribution among different parties.