情報処理学会 第73回全国大会講演要旨

There has been an increased interest in automated negotiation systems for their capabilities in reaching an agreement through negotiation among autonomous software agents. In real life problems, the negotiated contracts consist of multiple and interdependent issues which tends to make the negotiation more complex. In this paper, we propose to define a set of similarity measures used to compare the agents’ constraints, their utilities as well as their certainties over their possible outcomes. Precisely, we define a decision value-structure which gives a reasonable condition under which agents having similar decision structures can form a coalition. We think that a collaborative approach is an efficient way to reason about agents having complex decisional settings, but show similarities in their constraints, preferences or beliefs. Agents will tend to collaborate with agents having the same decisional settings instead of acting selfishly in a highly complex and competitive environment. Therefore, formed coalitions will benefit from the cooperation of its members by satisfying their constraints as well as maximizing their payoffs. Under such criterion, the agents can reach an agreement point more optimally and in a collaborative way. Experiments have been performed to test the existence of the decision value-structure as well as its capability to describe an agent’s decision structure. Moreover, the decision value-structure was used for coalition formation based on measuring the agents similarities.