Abstract
Negotiation has been widely applied in computational systems, particularly in e-marketplaces (e-markets), to facilitate transactions between buyers and sellers. In existing e-markets, participants must repeatedly be online to follow the progress of a transaction, especially where negotiation is required. Also, some market mechanisms (e.g. auctions) can be of long duration, and pursuing the best deals can require participants to engage in multiple negotiations simultaneously. In this context, the issue is how to build models that support agents to engage in open and dynamic environments concurrently.
We develop a model that allows software agents to negotiate concurrently with other agents. The main goal is to deploy agents that achieve the best outcome in agreements that allocate resources such as goods and services. We build upon previous work to develop a complete concurrent negotiation architecture, describing all the necessary components to allow an agent to take decisions in a concurrent negotiation. We also revise the known alternating protocol to support concurrent negotiations in open e-markets. This allows participants to request-to-reserve, cancel or exit a negotiation.
We also develop a novel strategy, Conan, which relaxes assumptions previously made regarding deadlines and knowledge about the market and the opponents. Consequently, our strategy is more realistic for open e-markets in deciding what action to take and, if the action is to offer, what offer to make next. We represent Conan using a logic-based knowledge representation. We then build a negotiation simulator Recon to simulate and evaluate our strategy. Recon supports the development of software agents negotiating concurrently with other agents; previous work only supported single bilateral negotiation. We then create a set of realistic experimental negotiation scenarios using opponents from the existing literature. We show empirically that Conan outperforms the state-of-the-art and other agents in terms of average utility gained from negotiations.
We develop a model that allows software agents to negotiate concurrently with other agents. The main goal is to deploy agents that achieve the best outcome in agreements that allocate resources such as goods and services. We build upon previous work to develop a complete concurrent negotiation architecture, describing all the necessary components to allow an agent to take decisions in a concurrent negotiation. We also revise the known alternating protocol to support concurrent negotiations in open e-markets. This allows participants to request-to-reserve, cancel or exit a negotiation.
We also develop a novel strategy, Conan, which relaxes assumptions previously made regarding deadlines and knowledge about the market and the opponents. Consequently, our strategy is more realistic for open e-markets in deciding what action to take and, if the action is to offer, what offer to make next. We represent Conan using a logic-based knowledge representation. We then build a negotiation simulator Recon to simulate and evaluate our strategy. Recon supports the development of software agents negotiating concurrently with other agents; previous work only supported single bilateral negotiation. We then create a set of realistic experimental negotiation scenarios using opponents from the existing literature. We show empirically that Conan outperforms the state-of-the-art and other agents in terms of average utility gained from negotiations.
Original language | English |
---|---|
Qualification | Ph.D. |
Awarding Institution |
|
Supervisors/Advisors |
|
Award date | 27 Apr 2016 |
Publication status | Unpublished - 2016 |