Federated Approach to Online Pairwise Comparison Ranking Problem

Abstract

In this work, we propose a federated learning (FL) framework for the online HodgeRank problem to obtain a global ranking based on the pairwise comparison data provided by users while respecting users' data privacy. HodgeRank is a statistical ranking method that views pairwise comparison data as an edge-weighted directed graph, and, vertex weight function can be viewed as a ranking that recovers edge weight using difference. A critical assumption of the HodgeRank is the connectivity of the comparison graph. This assumption relies on the contribution of users' data to form a weakly connected directed graph between items to be ranked. In this paper, we aim to find a framework to compute HodgeRank with the minimum usage of users' data.