EITM: Recognizing, Modeling, and Simulating Coupled Processes in Dyadic Interaction

有大量文献表明，家庭内部关系的质量与家庭成员的身心疾病发病率之间存在对应关系。这项研究解决了两个基本问题：“家庭成员之间，无论是成年人还是成年人与孩子之间，什么样的行为模式与关系质量有关？”以及“这些模式能否被量化到足以在计算机模拟中重现？”我们回答这些问题的研究工作分为四个步骤。首先，我们使用从已婚夫妇的行为互动中收集的现有观察数据来确定哪些行为集群-它们在互动中的时间和流量与关系的质量最相关。其次，我们使用这些提取的行为集来构建模式识别软件（例如，隐马尔可夫模型）来区分痛苦和非痛苦的关系。第三，鉴别模式成为构建二元相互作用计算机模拟的蓝图；我们将开发计算机模拟，重现在痛苦和非痛苦的已婚夫妇中看到的模式序列。最后，利用现有数据，我们将应用模式提取和计算机模拟的方法来观察亲子互动中的行为。本研究的广泛影响是双重的：首先，通过利用模式识别技术，我们推进了人类互动观测数据量化的最新技术；其次，通过计算机模拟进行模式分类和测试是开发基于计算机的识别系统以识别高危人群的第一步。

There is a substantial literature showing the correspondence between the quality of relationships within a family and the incidence of physical and psychological maladies among its members. This study addresses two fundamental questions: "What patterned behaviors between family members, either adults or adults with their children, are associated with relationship quality" and "Can these patterns be quantified sufficient to be reproduced in computer simulations?" Our research efforts to answer these questions are organized in four steps. First, we use existing observational data collected from the behavioral interactions of married couples to determine which behavioral clusters-their timing and flow-in the interaction are most associated with the quality of the relationship. Second, we then use these extracted behavior sets to construct pattern recognition software (e.g., Hidden Markov Models) to discriminate distressed from non-distressed relationships. Third, the discriminating patterns then become the blueprint for constructing computer simulations of dyadic interaction; we will develop computer simulations that reproduce the sequencing of patterns seen in distressed and non-distressed married couples. Finally, using existing data, we will apply the developed methodology of pattern extraction and computer simulation to behaviors observed in parent-child interactions. The broader impacts of this research are twofold: first, by utilizing pattern recognition technology, we advance the state of the art for quantifying observational data of human interaction, and second, the classification of patterns and testing via computer simulation is the first step in developing a computer-based recognition system for identifying at-risk populations.