Socially Motivated Approach to Simulate Negotiation Process

Abstract

Automated negotiation is treated as multi-disciplinary area of rigorous research consisting of Multi Agent Systems (MAS) such as optimization, decision support system, game theory and e-commerce. Thus, automated negotiations can be treated as a search space problem where different autonomous agents try to occupy their own utilities via finding the best possible option obtained from the search space. In this work, a comprehensive study was conducted on the negotiations' elements in order to model Agent Z. Cultural algorithms were used in the modeling as they offer a dual inheritance system between the belief space and population space. Agent Z was modeled to be compatible with General Environment for Negotiation with Intelligent multi-purpose Usage Simulator (GENIUS). Moreover, the simulator allows agent Z to negotiate with a human negotiator through a user interface window. Agent Z has three major improvements through its versions (V1.0, V2.0, and V3.0). In Agent Z-V1.0, the agent was modeled to deal with cooperating opponents. In this version, the population size was fixed and independent on the domain size. Moreover, the termination operator was time and utility dependent. In Agent Z-V2.0 the number of children was changed to a percentage of the domain size instead of a fixed number in version 1. Additionally, a memory was added to the agent to store best offers sent from the opponents. An important modification in V2.0 was the algorithm of producing the child generation from several parents unlike V1.0, which was produced from two parents only. An internal clock was added to V2.0 that divided the time into two phases. Lastly Agent Z-V3.0 was a hybrid agent that integrated V1.0 and V2.0 and a C factor was introduced. Agent Z is much more cooperative than other agents and tends to find win-win agreement rather than win a negotiation. It has been observed that when negotiators cooperate the time to reach an agreement is faster than competing agents. Finally Agent Z was found to perform well with time and utility dependent agents unlike utility only dependent agents who were forced to accept the best offer sent when time is about to finish.