情感教学agent：建模与反馈策略研究

发布时间：2018-02-27 15:06

本文关键词： 情感计算智能教学系统教学agent 情感建模情感反馈策略　出处：《西南大学》2008年博士论文　论文类型：学位论文

【摘要】： 智能agent是人工智能非常重要的一个研究领域,因其自治性、主动性、社会性的特点得到广泛应用。越来越多的智能教学系统开始采用智能agent作为构建体系结构的支撑技术,并由此形成了教学agent这一新的研究方向。现代认知科学和神经生理科学的研究显示情感对人类的认知具有重要的影响。而教育科学研究领域也从未停止过对学生学习过程中情感体验的关注。因此,赋予教学agent情感处理的能力就成为该领域一个非常明确的研究目标。受心理学关于情感模型研究的启发,智能agent研究领域也开始关注情感要素在智能系统建模中的作用,并陆续出现了一些智能agent的情感模型。然而,当前的大部分模型仍集中于agent的外部反应行为而很少考察其内部状态。因此,当前的情感模型多是基于一些静态规则和预先定义的领域知识生成情感,缺乏自适应性。同时,现有智能agent的情感模型对情感信息的处理还是基于二值逻辑,没有体现出情感本身的动态性和模糊性的本质。另一方面,本文针对教学agent的比较研究发现,尽管现有一些设计已尝试利用学生的情感状态去选择更适合学生的教学方法,却鲜有激发和调动学生的积极情感从而对学生形成情感支持的研究。此外,现有的教学agent即使已尝试情感建模,但仅停留在增强可信性的层次上,尚未真正赋予教学agent情感识别的能力和情感合成机制。本文提出了基于模糊逻辑的教学agent情感建模方法,体现了情感动态型和模糊性的本质;使用机器学习算法实现了对教学agent情感经验信息的学习,增强了模型的自适应性。在此基础上,本研究设计了一个能够识别学生情感,具有情感表现能力,能为学生在学习过程中提供情感支持,并最终促进学生学习效果的教学agent——情感教学agent(Emotional and Pedagogical Agent,EPA)。论文的主要工作和创新点包括: (1)提出了基于模糊逻辑和机器学习的情感建模方法。使用模糊集合表征事件对目的影响程度,目的重要程度和期望程度,并探索了基于模糊逻辑规则的情感评估方法,以相对少的模糊规则实现在可观测事件和情感状态上的平滑转换,体现了人类情感动态性和模糊性的特点;使用机器学习算法实现了教学agent对事件预期值等经验信息的学习,增强了教学agent情感模型的自适应性。 (2)提出了基于学生动机类型的情感识别方法。有关学生学习动机的研究显示,学生的学习动机类型(表现取向,掌握取向)对其学习中可能的情感状态产生重要影响。本研究在情感识别方法中引入学生的动机类型,体现了学生情感识别过程中的个体差异。 (3)设计了旨在对学生形成情感支持的情感教学agent反馈策略。心理学和教育学领域的研究显示,学生的消极情感会对其学习形成障碍,而积极情感则会促进学生的学习。本研究在识别学生情感状态,获知其动机类型的基础上,为情感教学agent设计了一系列情感反馈策略及其决策机制,通过情感行为和话语信息触发学生的积极情感,消除消极情感的影响,实现对学生的情感支持。
[Abstract]:Agent intelligent artificial intelligence is a very important research field, because of its autonomy, initiative, social characteristics have been widely used. More and more intelligent tutoring systems begin to adopt intelligent agent as the key technology of the architectures, and thus the formation of the agent teaching as a new research direction of modern cognitive science and. Neurophysiology study shows have important influence on human cognition and emotion. And the field of education and science research has never stopped learning emotion experience in the process of attention to students. Therefore, given the ability of teaching agent emotional processing becomes a very clear research objective in this field.
Psychology about inspired emotional model research, research in the field of smart agent also began to pay attention to the emotional elements in the intelligent system modeling function, and there are some intelligent agent emotion model. However, for most of the current external reaction model is still focused on agent and rarely investigated its internal state. Therefore, the current model of emotion many are some static rules and predefined domain knowledge generation based on emotion, lack of self adaptability. At the same time, the existing processing emotion model of intelligent agent of emotional information is based on two valued logic, no emotion itself reflects the dynamic and fuzzy nature.
On the other hand, compared to research on Teaching of agent this paper, although some existing design has attempted to use the students' emotional state to choose more suitable teaching methods for students, positive emotion and arouse students rarely stimulate to form of emotional support for the students. In addition, the existing teaching agent even have attempted modeling of emotions, but only stay in the enhanced credibility level, has not really given teaching agent emotion recognition ability and emotion synthesis mechanism.
This paper puts forward the teaching agent Emotion Modeling Method Based on fuzzy logic, essence of the emotional dynamic type and fuzziness; using machine learning algorithm of agent teaching experience information learning, enhance the adaptability of the model. On this basis, this study designed a can recognize students' emotions, with emotion performance, provide emotional support for students in the learning process, and ultimately promote the learning effect of students teaching agent (Emotional and Pedagogical agent teaching Agent, EPA). The main work and innovation points include:
(1) an affective modeling method based on fuzzy logic and machine learning is proposed.
The degree of using fuzzy set representations of events to influence, important degree and degree of expectation, and to explore the emotional evaluation method based on fuzzy logic rules with fuzzy rules can be implemented in relatively few smooth observations and emotional change in state, embodies the human emotion dynamic and fuzzy characteristics; machine learning algorithm to achieve the value of information on agent learning experience events expected, adaptability of teaching agent emotion model.
(2) a method of emotion recognition based on the type of student motivation is proposed.
Study on students' learning motivation, learning motivation types (performance orientation, mastery oriented) have an important impact on the possible emotional state during their learning. This study introduces students in recognition of the types of motivation, reflects the students' individual differences in emotion recognition process.
(3) a agent feedback strategy is designed to create emotional support for students.
The researches in psychology and pedagogy, students' negative emotions are hindrance to their learning, and positive emotion can improve the students' learning. The study of emotion in recognition of student status, based on the known types of motivation, emotion teaching agent designed a series of emotional feedback strategies and decision-making mechanism, actively trigger the feelings of the students through the emotional behavior and discourse information, to eliminate the influence of negative emotions, the students realize the emotional support.

【学位授予单位】：西南大学
【学位级别】：博士
【学位授予年份】：2008
【分类号】：G420

【引证文献】