Optimizing recommender system using federated clustering

Abstract

Currently, recommender systems are widely applied in various fields. However, due to the limited needs and special circumstances of users to be recommended, it is difficult for a recommender system to cover all users' interest lists at the same time. In this work, we present a kind of optimized federated clustering scheme (OP-Fed-Clustering) for users' private tendency data. The scheme starts by coding the initial data objects to protect privacy and then optimizes the assignment of data points based on object similarity. We also validates the algorithm's effectiveness on the FoodRecipe dataset and compares the algorithm to initial K-FED. Our tentative data show that the effectiveness of the proposed OP-Fed-Clustering algorithm, demonstrating universally superior performance while preserving user data confidentiality.