Small-World Reputation Management in E-Commerce Communities: Results on Pivots, Clustering, and Scattering

Successful commerce relies heavily upon the reputations that the different parties acquire through their dealings with each other. We consider a community whose members provide services as well as referrals for services to each other. The search for a service is initiated when a special need or opportunity (e.g., a solicitation from a business) arises. Personal agents assist users in evaluating the services and referrals provided by others, maintaining contact lists, and deciding whom to contact. The agents can communicate with anyone, but may communicate with only a small fraction of the members of the community. Agents and their users have full autonomy in deciding whether or how to respond to a request. We view an e-commerce community as a social network, which supports reputations both for expertise (providing good service) and helpfulness (providing good referrals). We motivate metrics to estimate the effectiveness of a social network to support e-commerce. We use these metrics to compare heuristics through which the agents acquire opinions about and decide whether to help others. We study aggregate phenomena such as the emergence of subcommunities, pivot vertices (which link different subcommunities), and the sensitivity of the social network to the addition or removal of an agent. Our main results are that the users’ reputations reflect their expertise and helpfulness and the social network improves in its support for service location. Importantly, we discovered that the quality of the network improves with the presence of a pivot, has an inverse relationship with clustering, and improves with respect to a concept we term cautious scattering. Topics: Reputation and trust mechanisms and issues. This research was supported by Ericsson and by the National Science Foundation under grant IIS-9624425 (Career Award).