Designing negotiation agents for automated negotiating agents competition (ANAC)

• Main Author: Lim, Bryan Yi Yong
• Other Authors: Bo An
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/137917
• Institution: Nanyang Technological University

Negotiation plays an important role when multiple agents with different interests wish to reach an agreement to cooperate towards the same goal. In order to create novel research in the autonomous agent design field in AI, competitions such as Automated Negotiating Agents Competition(ANAC) were introduced, which brought researchers from the negotiation community together and provided unique benchmarks for evaluating practical negotiation strategies. This paper discusses the strategies adopted by Agent29 which is developed in accordance with the regulation of the ANAC 2020. The strategy of Agent29 is designed based on the nature of the Werewolf Game. The ability of Agent29 can be categorized into two parts, role estimation part which estimates the role of the agents, and decision making part which decides what to do. For role estimation, a combination of two Naive Bayes classifiers is used to estimate the roles of the agents based on the frequency of action. On the other hand, Agent29 used a multiple strategy chooser to choose a strategy based on the win rate of each strategy profile while making a decision. The performance of Agent29 was evaluated by competing with other autonomous agents under different competition settings by using the AIWolf platform, and the results were recorded and analysed. Although the results showed that Agent29 does not perform as well as the finalist agents, conclusions based on the experiments have been made which can be used to improve the agent in the future. For future enhancement, more information can be taken into consideration while making decisions or estimating the role of the other agents. Further testing is recommended.