Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.568890
Title: Practical strategies for agent-based negotiation in complex environments
Author: Williams, Colin Richard
Awarding Body: University of Southampton
Current Institution: University of Southampton
Date of Award: 2012
Availability of Full Text:
Access through EThOS:
Access through Institution:
Abstract:
Agent-based negotiation, whereby the negotiation is automated by software programs, can be applied to many different negotiation situations, including negotiations between friends, businesses or countries. A key benefit of agent-based negotiation over human negotiation is that it can be used to negotiate effectively in complex negotiation environments, which consist of multiple negotiation issues, time constraints, and multiple unknown opponents. While automated negotiation has been an active area of research in the past twenty years, existing work has a number of limitations. Specifically, most of the existing literature has considered time constraints in terms of the number of rounds of negotiation that take place. In contrast, in this work we consider time constraints which are based on the amount of time that has elapsed. This requires a different approach, since the time spent computing the next action has an effect on the utility of the outcome, whereas the actual number of offers exchanged does not. In addition to these time constraints, in the complex negotiation environments which we consider, there are multiple negotiation issues, and we assume that the opponents’ preferences over these issues and the behaviour of those opponents are unknown. Finally, in our environment there can be concurrent negotiations between many participants. Against this background, in this thesis we present the design of a range of practical negotiation strategies, the most advanced of which uses Gaussian process regression to coordinate its concession against its various opponents, whilst considering the behaviour of those opponents and the time constraints. In more detail, the strategy uses observations of the offers made by each opponent to predict the future concession of that opponent. By considering the discounting factor, it predicts the future time which maximises the utility of the offers, and we then use this in setting our rate of concession. Furthermore, we evaluate the negotiation agents that we have developed, which use our strategies, and show that, particularly in the more challenging scenarios, our most advanced strategy outperforms other state-of-the-art agents from the Automated Negotiating Agent Competition, which provides an international benchmark for this work. In more detail, our results show that, in one-to-one negotiation, in the highly discounted scenarios, our agent reaches outcomes which, on average, are 2.3% higher than those of the next best agent. Furthermore, using empirical game theoretic analysis we show the robustness of our strategy in a variety of tournament settings. This analysis shows that, in the highly discounted scenarios, no agent can benefit by choosing a different strategy (taken from the top four strategies in that setting) than ours. Finally, in the many-to-many negotiations, we show how our strategy is particularly effective in highly competitive scenarios, where it outperforms the state-of-the-art many-to-many negotiation strategy by up to 45%.
Supervisor: Jennings, Nicholas ; Gerding, Enrico Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.568890  DOI: Not available
Share: