Congrats: a Configurable Granular Trust Scheme for Effective Seller Selection in an E-marketplace

Problem. The e-marketplace of today, with millions of buyers and sellers who never get to meet face to face, is susceptible to the presence of dishonest and fraudulent participants, prowling on unsuspecting trading partners to cheat in transactions, thereby increasing their profit to the detriment of their victims. There is also the multiplicity of goods and services with varying prices and quality, offered by a mix of honest and dishonest vendors. In order to participate in trade without incurring substantial loss, participants rely on intelligent agents using a trust evaluation scheme for partner selection. Making good deals thus depends on the ability of the intelligent agents to evaluate trading partners and picking only trustworthy ones. However, the existing trust evaluation schemes do not adequately protect buyers in the e-marketplace; hence, this study focused on designing a new trust evaluation scheme for buyer agents to use to effectively select sellers. -- Method. To increase the overall performance of intelligent agents and to limit loss for buyers in an e-marketplace, I propose CONGRATS—a configurable granular trust estimation scheme for effective seller selection. The proposed model used historical feedback ratings from multiple sources to estimate trust along multiple dimensions. I simulated a mini e-marketplace to generate the data needed for performance evaluation of the proposed model alongside two existing trust estimation schemes—FIRE and MDT. -- Results. At the peak of performance of CONGRATS, T1 sellers with the highest trust level accounted for about 45% of the total sales as against less than 10% recorded by the least trustworthy (T5) sellers. Compared to FIRE and MDT, CONGRATS had a performance gain of 15% and 30%, respectively, as well as an average earning of 0.89 (out of 1.0) per transaction in contrast to 0.70 and 0.62 per transaction respectively. Cumulative utility gain among buyer groups stood at 612.35 as contrasted to 518.96 and 421.28 for the FIRE and MDT models respectively. -- Conclusions. Modeling trust along multiple dimensions and gathering trust information from many different sources can significantly enhance the trust estimation scheme used by intelligent agents in an e-marketplace. This means that more transactions will occur between buyers and sellers that are more trustworthy. Inarguably, this will reduce loss to an infinitesimal level and consequently boost buyer confidenc