近年来，人机交互技术日渐成熟，人们对交互质量的要求越来越高，自然、和谐的人机交互体验成为研究者们重要的追求目标。为了达到这个目标，这就要求机器具备与人相近的情感交互的能力，希望机器能够识别、推测、理解人的情绪状态和意图，并且做出适当的反馈。而对于智能体、虚拟人、机器人来说，希望它们在感知和理解人的情感状态之外，自身还能对外界环境变化产生情感反应，并通过表情、动作等恰当地表达出来。因此，研究者们在情感识别、情感建模、情感表示等情感相关研究领域投入了大量精力。 情感属于心理因素，因此情感建模的研究离不开心理学的理论基础。许多情感模型和系统都是基于心理学的情感理论构建的。这些心理学理论对人类情感产生和变化过程的重要方面进行了深入剖析，总结出了人类情感的主要特点、规律、反应机理等。基于这些心理学理论进行情感建模，更能构建接近人类真实情感过程的情感模型。最广为人知的情感基本理论是认知评价理论。认知评价理论认为，情绪是通过主体在经历外部刺激时，对其主观上认为重要的某个或某些方面进行评价而产生的，因此评价过程具有主观性，评价结果也因人而异。目前许多情感模型就是在该理论基础上构建的。此外，还有一些情感模型是基于统计概率模型的方法构建的。因为情感十分复杂，情感的产生具有不确定性，并且情感表现出概率性分布的特点，因此许多研究将情感划分为离散状态从而构成情感空间，构造情感状态在情感空间中的概率转移矩阵，以概率的方式来表示情感的产生和转移过程。 本文基于现有的理论基础和方法，完成了如下工作： （1）基于OCC认知评价模型和Ekman基本情绪理论，构建了一个面向虚拟主体的基于内外情绪关联的情感过程模型来模拟情绪产生与演变的过程。在该情感模型中，内在的情绪产生过程采用OCC认知评价方法来模拟，外在情绪采用与Ekman基本情绪模型相对应的表情来表征。本文对OCC模型及其评价过程进行了修改，并且针对修改的OCC模型和Ekman模型提出了一种新的映射方式，由此在内隐的情绪和外显的情绪表情之间建立了关联。该模型的情感演变过程即情绪强度的演变过程，本文采用一个非线性函数，即Sigmoid函数来计算情绪强度，同时考虑了外部刺激（Stimuli）、个性（Personality）、心情（Mood）以及同一时刻的其他情绪（Emotion）对当前情绪状态的影响。最后，通过仿真实验来讨论和分析该模型所模拟的情绪及其变化的众多性质，验证了模型的合理性和可用性。 （2）基于贝叶斯网络构建了一个多因素的情感产生与转移概率模型，并利用训练后的模型进行情感推理。该模型的贝叶斯网络中的节点由整个情感过程涉及的众多情感因素构成，整个网络结构体现了各情感因素之间的相互影响关系。接着，本文针对已训练好的模型提出一种推理机制对情感产生与转移过程进行连续状态推理，推理过程采用和积算法。最后，对该模型进行实例验证，构建了课堂教学场景中的学生学习情感产生与转移模型，采集课堂教学视频数据进行模型训练，采用交叉验证的方法验证使用该模型进行情感推理的准确率；根据推理机制，使用训练好的模型进行连续情感推理分析。

In recent years, human-computer interaction technology has matured, people's requirements for interaction quality have become higher and higher, and natural and harmonious human-computer interaction experience has become an important pursuit goal for researchers. In order to achieve this goal, the machine is required to have the ability of emotional interaction close to people. It is hoped that the machine can recognize, infer and understand the emotional state and intentions of people, and make appropriate feedback. For agents, virtual humans, and robots, it is hoped that in addition to perceiving and understanding human emotions, they can also generate emotional reactions to changes in the external environment and express them appropriately through expressions and actions. Therefore, researchers have invested a lot of energy in emotion-related research such as emotion recognition, affective modeling, and emotional expression. Emotion is a psychological factor, so the study of affective modeling cannot be separated from the theoretical basis of psychology. Many affective models and systems are built on psychology-based emotional theory. These psychological theories have conducted an in-depth analysis of the important aspects of human emotion generation and change process, and summed up the main characteristics, laws, and reaction mechanisms of human emotions. Based on these psychological theories for affective modeling, it is possible to construct an affective model that approximates the true emotional process of human beings. The most widely known basic theory of emotion is cognitive evaluation theory. Cognitive evaluation theories believe that emotions are generated by the subject’s evaluation of one or more aspects that are subjectively important when experiencing external stimuli. Therefore, the evaluation process is subjective and the evaluation results vary from person to person. At present, many affective models are based on these theories. In addition, some affective models are built on statistical probabilistic models. As we know, the emotion is very complex, the generation of emotion has uncertainty, and the emotion displays the characteristics of probability distribution. Therefore, many studies divide the emotions into discrete states to form the emotional space, and construct the probability transfer matrix of the emotional state in the emotional space so that the process of emotion generation and transfer can be described in a probabilistic way. Based on the existing theoretical basis and methods, this article completes the following work: (1) Based on the OCC cognitive evaluation model and Ekman's basic emotion theory, an emotional process model based on internal and external emotional associations was constructed to simulate the process of emotional generation and evolution for virtual agent. In this emotional model, the internal emotion generation process is modeled by the OCC cognitive evaluation method, and the external emotion is represented by the facial expression corresponding to the Ekman basic emotion model. This paper modifies the OCC model and its evaluation process, and proposes a new mapping method for the modified OCC model and Ekman model. This establishes a correlation between implicit emotion and explicit emotional expression. The emotion evolution process of the model is the evolution of emotional intensity. This paper uses a nonlinear function, sigmoid function, to calculate the emotional intensity. At the same time, the external stimulus, personality, mood, and the effect of other emotions on the current emotional state are taken into account. Finally, simulation experiments are conducted to discuss and analyze the many properties of the model's simulated emotions and changes, and verify the rationality and availability of the model. (2) A multi-factor probabilistic model of emotion generation and transfer is constructed based on Bayesian network, and the trained model is used to carry out new emotional reasoning. The nodes in this model's Bayesian network consist of many emotional factors involved in the completely emotional process. The entire network structure reflects the mutual influence of each emotional factor. Then, this dissertation proposes a reasoning mechanism for the trained model to conduct continuous state reasoning on the process of emotion generation and transfer. The inference process uses the sum-product algorithm. Finally, the model is validated by an example. The students' learning emotion generation and transfer model in the classroom-teaching scene is constructed. The classroom teaching video data is collected for model training, and the cross-validation method is used to validate the accuracy of the model's inference. According to the inference mechanism, this paper uses a trained model for continuous emotional inference analysis.