A Qualitative Ascending Protocol for Multi-issue One-to-Many Negotiations

Many practical distributed systems environments require novel automatic mechanisms including multi-attribute reverse auctions for efficient partner selection and contracts negotiation. Recent results [6] show that the property of transferable utilities is not of vital importance, as qualitative versions of the standard auctions (e.g. qualitative Vickrey auctions (QVA), qualitative English auctions (QEA)) are proved to exhibit nice efficiency properties as well. Such auctions require that the preferences of the auctioneer are publicly known. However, practical protocols of multi-bilateral closed negotiations between a buyer and multiple sellers are experimentally shown [9] to approximate the Pareto-efficient best-seller QVA outcome, without requesting that any of the parties explicitly reveals their preferences. The only condition is to enable bidders to learn preferences. In this paper we introduce two novel negotiation protocols that approximates a qualitative ascending English auction (QEA) and overcomes some restrictions imposed by the non-transferable utilities environment. Our auction-like protocols are designed for fully automatic environments with learning agents playing the bidders’ roles.