Recommending missing and suspicious links in multiplex financial networks

Abstract

Many complex systems in finance can be modeled as multiplex networks, or networks which depict multiple types of interactions between entities. We consider the problem of detecting missing and suspicious interactions in multiplex financial networks in a real world context where predictions are provided continuously according to budget limitations. We propose a recommendation system based on a recently proposed heuristic for link prediction and incorporate feedback from previous recommendations to improve the system's performance over time. We provide theoretical conditions which show our approach approximates an (intractable) entropy-minimization solution while remaining computationally efficient and providing recommendations that are explainable. We apply our approach to a real world multiplex financial network and demonstrate its effectiveness at discovering missing and false links.