Currently, there are many types of conversational agents whose goal is to emulate human behavior. These agents offer more believable conversations when their responses come from a deliberative process that mimics individuals’ character. Conversational agents are mainly used for response selection linguistics and context situated strategies. These approaches usually build rules to find answers in dialogues; however, this is not the best alternative when the communicative intentions are not literal and context dependent. Deliberative Agents can solve these issues and improve their selection process through the integration of preferences and personality in their cognitive process. Hence, this work investigates how to drive the expression of dialogues of a Conversational Deliberative Agent (CDA) through personality and fuzzy outranking relations; for this, it proposes the characterization of context, and corpus through speech acts theory, and also a selection process based on fuzzy outranking relations to compare corpus phrases and context to choose the best response. The main contributions of this work are (1) the agent architecture that integrates preferences and personality of an individual in the response selection cognitive process; (2) a characterization model of speech acting through criteria based on belief, desires, and intentions to define a more human behavior expression; and (3) the use of fuzzy outranking relations to select phrases from a corpus to match dialogue intentions. An experimental design demonstrated the aptitude of the developed CDA to offer quality responses on a tutor application. Also, the results showed the capacity of speech acts to handle contexts in dialogues.

当前，有许多类型的会话代理，其目标是模仿人类行为。当这些代理人的反应来自模仿个人品格的深思熟虑的过程时，他们会提供更可信的对话。会话代理主要用于响应选择语言学和情境定位策略。这些方法通常会建立规则以在对话中找到答案。但是，当交流意图与文字和上下文无关时，这不是最佳选择。协商代理可以通过在他们的认知过程中整合偏好和个性来解决这些问题并改善他们的选择过程。因此，这项工作研究了如何通过个性和模糊的排位关系来驱动对话协商代理（CDA）的对话表达。为此，本文提出了通过言语行为理论对语境和语料库进行刻画的方法，并提出了一种基于模糊排序关系的语料选择过程，比较语料库短语和语境以选择最佳反应。这项工作的主要贡献是：（1）在响应选择认知过程中整合个人偏好和个性的主体架构；（2）通过基于信念，欲望和意图的标准行事的言语表征模型，以定义更人类的行为表达；（3）使用模糊排名关系从语料库中选择短语以匹配对话意图。实验设计表明，已开发的CDA能够在导师的应用程序上提供质量响应。也，