Research On The Method And Implementation Technologies Of Multi-agent Cooperative Dialectical Reasoning

The inconsistency, impreciseness and incompleteness of information are inevitable inenvironments with mass of informantion. People are usually lost within this complexsituation, since they do not know which one they can catch up. How to acquire the mostreasonable decision knowledge from the environment with inconsistent information isbecoming an urgent topic in the research of information science. In our human history,humans use argumentation to express disagreement, to reach consensus and to bothformulate and convey reasoning. Argumentation plays significant role in conflictresolution in our daily life, in which palyers put forward arguments for or against thecontroversial standpoint and tries to convince each other. In the research of computerscience, especially Artificial Intelligence (AI), the theory of argumentation also provides anovel approach for non-monotonic reasoning. As a form of non-monotonic reasoning,argumentation could be viewed as a dispute resolution, in which the agents present theirarguments to establish, defend or attack certain proposition. Over the past years,argumentation has come to be increasingly central as a hot topic in AI and been gainingincreasing interest in AI research community.In order to know whether a controversial point (i.e. argument) is reasonable or notwithin inconsistent environment (i.e. argumentation framework), this dissertation studiesthe keystone and method for dealing with inconsistent information through argumentation.The dissertation proposes a model of argument games for preferred semantics of abstractargumentation framework and a framework for multi-agent cooperative dialecticalreasoning for the resolution of controversial issue and conflicting standpoint in thedistributed environment with inconsistent information, providing a novel method fordealing with inconsistent information in distributed setting. Meanwhile, theimplementation technologies of multi-agent cooperative dialectical reasoning, such asprotocol of multi-party dialogue games, instantiation of argumentation framework andarguments generation are discussed in the dissertation.In particular, the key research problems and novelty of the dissertation include:Firstly, the dissertation proposes a novel Model of Argument Games for PreferredSemantics (MAGPS) of abstract argumentation framework. Dialectical proof is the mainapproach for computing the acceptability of an argument in argumentation framework, andmostly scholars make their agreement on modeling the dialectical proof by argumentgames. The existing models of argument games usually represent the argument exchangeas dialogue between Pro and Opp by turns. These argument games suffer from thelimitations of randomness and blindness in argument exchange. In order to improve therelevance and validness of proposed argument in a dialogue, the dissertation studies anovel model of argument games for preferred semantics based on the notions of critical factor, legal move function and critical countermeasure et al., which make agents (Pro andOpp) to propose attacking argument aiming to the key-elements of a dialogue. In theory, itis sound and complete for using MAGPS for computing acceptability of an argumentationframework. And from the experimental results, MAGPS shares following merits: i)MAGPS answers the acceptability in shorter dialogues than most of existing model, and ii)MAGPS puts down the complexity of computing the acceptability of argumentationframework.Secondly, the dissertation represents a method of multi-agent cooperative dialecticalreasoning based on MAGPS. Based on our argument games, the dissertation then proposesa method and the framework of multi-agent cooperative dialectical reasoning forcomputing acceptability within distributed argumentation framework. The basic idea of ourmethod is that participants, who aimed at looking after more reasonable and well-foundedknowledge for decision-making, present their arguments for or against the controversialstandpoint (i.e. argument in test) in a dialogue way. The framework consists of two typesof agents, i.e. moderator and participating agents(participate in short), where participatesare joining in exchanging argument under the management and coordination of moderator.This procedure of critical thinking about problems and evaluating conflicting viewpoints iselaborated as a multi-agent cooperative dialectical reasoning procedure in the dissertationthat provides a novel method for deciding the acceptability of the argument in test underdistributed argumentation settings.Thirdly, the dissertation elaborates a protocol of multi-party dialectical reasoning(ProMDR) for multi-agent cooperative dialectical reasoning. One impedient of multi-agentcooperative dialectical reasoning is the design and implementation of protocol ofmulti-party dialogue games. Most of existing researches on dialogue games focuse ontwo-party setting. However, multi-party dialogue games are much more complicated thantwo-party dialogue games. With this consideration, the dissertation analyses thecharacteristics and difficulties of design and implementation of multi-party dialogue games,and summarises some principles on multi-party dialogue games protocol. Finally,according to these principles, a protocol of multi-party dialogue games for multi-agentcooperative dialectical reasoning is elaborated. ProMDR consists of a set of locutions and aset of dialogue rules, in which locutions include dispute locutions and control locutions,and dialogue rules consist of control rules, combination rules, and commitment rules. Thedissertation also describes ProMDR from axiom and operational semantics angles of view.Finally, the dissertation provides a method to instantiating abstract argumentationframework with monotonic logic, and an approach to generating attacking arguments usingmonotonic reasoning. Most of scholars working on argumentation put their focus onabstract argumentation framework, but ignore the issues such as what is an argument,where does it from and how to construct it et al. However, in practice, the generation of arguments is a requisite ability for an argumentative agent in argumentation dialogue,which bridges the gap between the theory and application of argumentation. Thedissertation firstly provides a method for instantiating abstract argumentation frameworkusing monotonic logic (i.e., we provides an instantiation of distributed argumentationframework represented as multi-agent system with argumentative knowledge base in Hornlogic), and then represents an approach, consisting of following three steps: parsingargument, mining attack relationship, and constructing argument, for generating attackingarguments dynamically aiming at a given argument. This approach enables participants indialogue games to generate attacking arguments aiming at critical elements in a dialogue.To sum up, the dissertation presents an approach of multi-agent cooperativedialectical reasoning based on our novel model of argument games and researches on thekey implementation technologies involved. Our research is a beneficial exploration ofinconsistent information resolution under distributed setting. We argue that our researchmakes a nice groundword for future research and has found wide application in bothmilitary and civil area.