M.Sc Thesis

M.Sc StudentKoifman Gabi
SubjectMulti-Agent Negotiation over Database Based Information
DepartmentDepartment of Industrial Engineering and Management
Supervisors DR. Onn Shehory
PROF. Avigdor Gal


We have developed a mechanism that supports trading database tuples in a multiagent system. The mechanism enables negotiation and evaluation of database-based information goods. Our research focuses on two aspects of the mechanism. The first aspect is performing automatic schema-matching between the buyer’s and seller’s databases. The second aspect is dynamic pricing of information goods. We propose methods for schema-matching and different policies for dynamic pricing of information goods. We have developed a test-bed that simulates a multi-agent system where each agent uses the offered mechanism and have evaluated the system performance when sellers use different pricing policies in two market environments, namely competitive and non-competitive. The investigated pricing policies include two novel pricing policies that implement negotiation and price discrimination across consumers and compared them to two policies known in the art, which implement dynamic posted pricing. We offer a schema-matching methodology in which a given set of heuristics is partitioned based on their complexity. Less complex heuristics are utilized in generating a top-K set of possible mappings. These mappings are analyzed for identifying possible points of failure and verified using the more complex heuristics. The complex heuristics uses statistical analysis of the buyer’s and seller’s databases to refine the generated mapping. The evaluation concentrates on two main issues. The first is evaluation of the offered methodology for schema-matching and the second is evaluating the different pricing policies. We have empirically demonstrated the superiority of the offered pricing policies in maximizing sellers' gains. We have additionally identified equilibria profiles of these policies. Evaluating the schema matching methodology, our experiments show a significant increase in the precision of mappings at a relatively low computational cost.