Vector similarities for offer generation in multi-attribute negotiation

Abstract

This thesis extends the existing approach of using similarity to previous opponent offers for offer generation in automated multi attribute negotiation. Specifically, this thesis focuses on the generality of the offer generation method that allows agents to negotiate in diverse negotiation domains. To achieve this, two vector similarities, cosine and Euclidean distance, are proposed and explored. Results from various domains show that adopting these two vector similarities affects negotiation outcomes positively, as reflected in low negotiation failure rates and close to optimal agreements with high mutual gain. Further, this thesis examines the dynamics of the negotiation and considers whether offers with concession can improve the utility of the opponent. This dynamic has been often overlooked in previous studies. Results show that offers generated with the two vector similarities can minimise lose-lose situations where concession are made, but no positive effect can be observed in term of the opponent's gain. Results from the experiments also show that the majority of offers improve the opponent's payoff. In addition to the proposed vector similarities in generating offers, a contribution of this thesis is the metric called sum of weight difference. This metric identifies the level of opposition between two agents which can affect their potential negotiation outcome. Such a metric can help match two agents in a pre-negotiation stage and determine if they are compatible to negotiate and likely to obtain a mutually high-gain agreement.

Date of Award	2017
Original language	English
Supervisor	Michael Wagner (Supervisor), Masoud MOHAMMADIAN (Supervisor) & Menkes van den Briel (Supervisor)