Design of computational market systems for network information services.

Abstract

One of the goals of an information service network is to provide users with access to a multitude of highly distributed and constantly changing information sources and services. Market price systems constitute one well-studied class of mechanisms for allocating resources among distributed decision makers. By implementing a virtual market system for computational agents, we can hope to realize some of the desirable properties of markets in the distributed computing context. Moreover, the underlying economic theory provides an analytical framework for predicting aggregate behavior and for designing individual agents. This thesis identifies and explores some of the issues involved in designing computational market systems in the context of the University of Michigan Digital Library (UMDL). UMDL provides library services in a distributed environment, where information agents buy and sell information services. The explict realization of this design is to provide a commerce infrastructure that supports the process of describing, locating, and negotiating for a wide variety of information services. One part of this infrastructure is the interaction framework: service description languages, and negotiation and exchange protocols. Another consists of various infrastructure services, which simplify and automate the use of many these languages and protocols. This thesis focuses on the design of negotiation protocols and the integration of negotiation mediators, or auctions. Since the number of potential service offerings and negotiation options is unbounded, UMDL also requires some mechanism to manage the scope of markets actually available to agents. The Auction Manager infrastructure component provides commerce middleware services that simplify and automate both the creation of new markets and the matching of buyers and sellers to existing markets. Online services often allow agents to select various service options. By formally representing two selection operators (buyer and seller choice) and related inference rules, the Auction Manager can more flexibly match agents to markets and detect arbitrage opportunities. The Auction Manager also provides a vehicle to experiment with alternative auction creation policies. Two simple case studies evaluate the welfare effects of different market configurations and auction parameters, and provide a basis for future auction creation policies in the Auction Manager.