No abstract

No abstract

Recommender systems (RSs) play a crucial role in helping users quickly find their desired services and promoting sales of service providers in e-commerce. However, service providers intentionally upload cherry-picked service information to mislead RSs for greater profit. Such misleading recommendations pose the risk of degrading user experience and gradually undermine users' confidence in the whole service market over the long term. Worse still, most current expert recommendation methods are more susceptible to such risk because they are heavily dependent on this incomplete service information and assume it is trusted. Therefore, how to satisfy users' service requirements with risks minimized at the same time motivates our work. In this paper, we first discern two risky facets that pose significant challenges to expert recommendation: Sense Drop and Blue Joy. Then we propose a unified framework called RecRisk, which integrates trust, heat equation, and modern portfolio theory to address the above challenges. The main contributions of RecRisk are two-fold: (1) To select the services which satisfy the users' preferences, we design a trust-aware heat equation model (TAHE) that combines heat flow theory with trust elements. (2) We develop a flexible model based on modern portfolio theory to weigh users' satisfaction and services' risk facets, and finally recommend a ranked service list to users. Our experimental results demonstrate that RecRisk simultaneously achieves higher recommendation precision and decreases risks when compared to state-of-the-art approaches.