Reasoning about Advisors for Seller Selection in E-Marketplaces via POMDPs

Abstract. In an e-marketplace populated with a large number of sell-ers, some of which may be dishonest, the selection of good sellers to do business with is crucial but challenging especially when buyers do not have much experience with these sellers. In this paper we introduce the SALE POMDP, a framework for the seller selection problem that allows the decision maker to reason both about the quality of the sellers, as well as the trustworthiness of the advisors. In particular, the framework allows the agent to ask advisors about the trustworthiness of other ad-visors while still offering the benefit of optimally trading off information gathering and exploitation of knowledge as afforded by a POMDP based approach. Via this model, we present a preliminary investigation on the benefit of reasoning about trustworthiness of advisors. We also demon-strate how this enables incorporation of trust propagation as an integral part of the decision making process