劝说对话中的信念修正方法研究

发布时间：2018-08-30 13:34

【摘要】：在研讨系统中,专家之间围绕一个主题进行研究讨论,以一种相互发送争议的劝说对话形式解决冲突、达成共识。劝说对话可以分为两种类型,一种是争议之间逻辑性较强、确定性的对话;一种是争议之间推测较多,极具不确定性的对话。在一次研讨中,由于专家知识的局限性,会出现研讨不充分导致达成的共识不够理性的情况。提取相关主题研讨历史中一致性的知识并构建为知识库可以给专家提供辅助支持作用。因此一次研讨中不仅需要得到研讨的结果,构建一个有助于研讨的辅助知识库是十分必要的,这就需要对研讨的劝说对话建模以及对如何消除矛盾形成一致性的知识集进行研究。在人工智能领域中,信念是智能体对周围环境的认知。由于信念的动态性以及不确定性,信念之间会出现因为矛盾而导致信念集不相容的情况。信念修正是消除这种矛盾使信念集相容的一种有效手段。将信念修正理论引入到研讨系统中,能够有效的筛选出对话中可接受的信念(知识)并作为后续研讨的参考,以弥补专家知识的局限性使最后达成理性的共识。本文的研究目标是将信念修正方法引入到研讨的劝说对话中,最后得到一个可接受信念集。首先将知识形式化的表示,将研讨中的知识信念形式化的表示为辩论理论中的事实和规则。然后对劝说对话过程建模,根据争议之间的攻击、支援、反驳关系以及时间序列构建为一个三元组的辩论对话框架。最后基于该辩论对话框架提出了应用于确定性对话的可辩护性(Defensibly)求解算法以及应用于不确定对话的可信度(Certainty factor)求解算法。可辩护性是用0和1来表示一个信念的可接受性,凭借争议之间的攻击关系自下而上的传递可辩护性。可信度是对信念的一种概率性的体现,用0~1之间的概率值来量化信念的置信程度,与可辩护性求解算法不同的是需要专家来对不确定性的规则做出相关评价。为验证本文提出的信念修正方法在研讨的劝说对话下具备有效性,本文利用Jade框架开发了原型系统,并就专家之间的确定性的和不确定性的劝说对话实例问题进行相关合理性分析。结果表明,本文提出的劝说对话的信念修正方法能够在劝说对话中有效的修正信念。
[Abstract]:In the research system, experts study and discuss a topic to resolve the conflict and reach a consensus by means of a form of persuasive dialogue that sends disputes to each other. Persuasive dialogue can be divided into two types, one is the logical and deterministic dialogue between disputes, the other is the dialogue with more speculation and uncertainty. In one study, due to the limitation of expert knowledge, the consensus reached by the study is not rational enough. Extracting the knowledge of historical consistency and constructing the knowledge base can provide support to experts. Therefore, it is necessary not only to obtain the results of a seminar, but also to build a knowledge base that is helpful to the study. Therefore, it is necessary to study the modeling of persuasive dialogue and the consistent knowledge set on how to eliminate contradictions. In the field of artificial intelligence, belief is the agent's understanding of the surrounding environment. Because of the dynamic and uncertainty of belief, the incompatibility of belief set will occur due to the contradiction between beliefs. Belief correction is an effective means to eliminate this contradiction and make belief set compatible. By introducing the belief correction theory into the research system, we can effectively screen out the acceptable beliefs (knowledge) in the dialogue and make up for the limitation of the expert knowledge so as to reach a rational consensus in the end. The aim of this paper is to introduce the belief correction method into the persuasive dialogue, and finally to obtain an acceptable belief set. Firstly, the formal representation of knowledge and the representation of knowledge belief in the discussion are formalized as the facts and rules in the debate theory. Then, the process of persuasion dialogue is modeled and constructed into a triple debate dialogue framework according to the attack, support, refutation relation and time series. Finally, based on the framework of the debate dialogue, the defensible (Defensibly) solution algorithm for deterministic dialogue and the credibility (Certainty factor) solution algorithm for uncertain dialogs are proposed. Defensibility is to express the acceptability of a belief by 0 and 1. Credibility is the embodiment of probability of belief. Using the probability value between 0 and 1 to quantify the confidence degree of belief, different from the defensible algorithm, it needs experts to evaluate the uncertainty of the rules. In order to verify the validity of the proposed belief correction method under the persuasive dialogue discussed in this paper, a prototype system is developed using the Jade framework. The rationality of the deterministic and uncertain persuasive dialogue between experts is analyzed. The results show that the belief correction method proposed in this paper can effectively correct beliefs in persuasive dialogue.
【学位授予单位】：湖北工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18

【参考文献】