Modelling risk and return awareness for p2p lending recommendation with graph convolutional networks

摘要

P2P lending recommendation (P2PLR) has become an important and popular component of P2P lending platform, which aim to align the right loans with the right investors according to historical interaction. Most of the existing P2PLR models take risk and return as side information into P2PLR to capture the relationship among loans. Although some of them have been proven effective, the following two insights are often neglected. First, risk, the possibility of investors’ loss, is the most important attribute in P2P lending platform, and the implicit capture of risk may lose critical information in the P2PLR scenario. Second, the preference and sensitivity of return on loans is unknown, which is only implicitly reflected in the loans that the investor has invested. These insights motivate us to propose a novel P2PLR model riSk and return Aware Recommendation (STAR), which propagates the influence of risk and return on investor to make the learned investor representations be risk and return aware. Particularly, we specify two STAR methods, named STHCW (riSk and return aware Recommendation with High-order Connectivity with Weight) and STHC (riSk and return aware Recommendation with High-order Connectivity without weight), based on two strategies of embedding and propagation mechanisms, respectively. We conduct extensive experiments on three real-world datasets, demonstrating the effectiveness of our proposed method in recommendation via risk and return. Further analysis reveals that modeling the risk and return awareness is particularly useful for learning the risk and return aware preference of investor.