A machine learning based evaluation of a negotiation between agents involving fuzzy counter-offers

Abstract

Negotiation plays a fundamental role in systems composed of multiple autonomous agents. Some negotiations may require a more elaborated dialogue where agents would explain offer rejections in a general and vague way. We propose that agents would represent their disappointment about an offer through a fuzzy set applied to each attribute of the offer. Fuzziness can also be very useful in order to make user profiles more difficult to acquire. The satisfaction of this intention is evaluated using classification techniques to compare the accuracy of the models that were obtained from the observation of the behaviour of the agents. In order to test how much information may be extracted about the internal preferences of agents, the task of modeling is translated into a classification task solved by a technique that would generate symbolic representations, such as m5.