Back to search

University of York

Reinforcement Learning for Argumentation

Abstract

dc:description.abstract

Argumentation as a logical reasoning approach plays an important role in improving communication, increasing agree-ability, and resolving conflicts in multi-agent-systems (MAS). The present research aims to explore the effectiveness of argumentation in reinforcement learning of intelligent agents in terms of, outperforming baseline agents, learning transfer between argument graphs, and improving relevance and coherence of dialogue quality. This research developed `ARGUMENTO+' to encourage a reinforcement learning agent (RL agent) playing abstract argument game for improving performance against different baseline agents by using a newly proposed state representation in order to make each state unique. When attempting to generalise this approach to other argumentation graphs, the RL agent was not able to effectively identify the argument patterns that are transferable to other domains. In order to improve the effectiveness of the RL agent to recognise argument patterns, this research adopted a logic-based dialogue game approach with richer argument representations. In the DE dialogue game, the RL agent played against hard-coded heuristic agents and showed improved performance compared to the baseline agents by using a reward function that encourages the RL agent to win the game with minimum number of moves. This also allowed the RL agent to adopt its own strategy, make moves, and learn to argue. This thesis also presents a new reward function that makes the RL agent's dialogue more coherent and relevant than its opponents. The RL agent was designed to recognise argument patterns, i.e. argumentation schemes and evidence support sources, which can be related to different domains. The RL agent used a transfer learning method to generalise and transfer experiences and speed up learning.

Degree

thesis:*
Name dc:type.qualificationname
Ph.D
Level dc:type.qualificationlevel
doctoral
Grantor dc:publisher.institution
University of York
Year dc:date.issued
2019

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • ALAHMARI, SULTAN
Advisors dc:contributor.advisor
  • Yuan, Tangming
  • Pears, Nick

Rights

Language dc:language.dcterms_iso639_2
eng

Identifiers

dc:identifier.*
Identifier
uk.bl.ethos.811423

Chain of custody

source
Harvested from
White Rose University Consortium
Base URL
etheses.whiterose.ac.uk/cgi/oai2
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
related terms
citation

ALAHMARI, SULTAN. Reinforcement Learning for Argumentation. doctoral thesis, University of York, 2019.