在人机协同的智能时代，培养具有高阶素养的创新人才是教育领域的核心任务。高阶素养的培育需要实施过程性评价，学习投入分析就是理解学习过程的关键议题。目前的学习投入研究过于关注学习者个体的行为投入，个体行为投入作为外显化投入难以深入揭示影响学习的机制，也难以回应学习社会化的本质，因此本研究聚焦于协作认知投入展开探索。协作认知投入具有内隐属性，教学人员无法直观地了解学习者协作认知投入方面的问题并开展教学改进。因此，如何对协作认知投入进行客观监测并实施干预显得尤为重要。同时，协作认知投入的内涵具有多维度多层次特性，需要学习者在心理、身体、社交等方面进行全方位地认知参与。因此，伴随式采集学习者在协作学习中的多模态数据，为全面、客观地量化协作认知投入带来了新思路。

基于此，本研究聚焦于“如何通过收集学习者在协作学习过程中产生的多模态数据，实现协作认知投入水平的量化表征，并对此进行有效干预”这一核心问题进行了研究设计。本研究按照“是什么”、“如何测量”、“为什么”、“如何改进”的逻辑顺序，从协作认知投入的概念模型与测量框架构建、多模态表征、影响路径分析、干预设计与实施四个方面开展研究。四部分研究是一个层层递进、逐步深入的关系。概念模型与测量框架确定了协作认知投入的组成要素，多模态表征则实现了各要素的量化测评，测评结果可以支持协作认知投入的影响路径识别，以影响路径为作用点设计干预能够促进协作认知投入水平的提升。

第一个研究是协作认知投入的概念模型与测量框架构建。在厘清协作认知投入的概念内涵之后，本研究以认知临场感、集体认知、协同知识建构作为理论基础，通过理论推演形成协作认知投入的APR概念模型（Activation—Processing—Response）。该模型由激活系统（Activation）、加工系统（Processing）与反应系统（Response）组成，三个子系统通过相互协调维持着协作认知投入的发生。基于APR概念模型，本研究分别厘清每个子系统下设的测量维度，并通过两轮德尔菲法，由12位相关领域专家对框架进行评价和修订，形成最终的协作认知投入测量框架。其中，激活系统包括唤醒度与适应度两个维度，加工系统包括探究度、整合度、反思度与支持度四个维度，反应系统包括生产力、影响力与创新性三个维度。

第二个研究是协作认知投入的多模态表征。本研究构建了主题为“海绵校园设计”的协作问题解决活动，以北京市某高校的184位学习者为研究对象，采集协作过程中的生理数据与会话文本数据。基于学习者产生的多模态数据，本研究依次使用信号分析技术、深度学习词向量技术、语义分析技术提取多模态数据特征，分别针对激活系统、加工系统与反应系统构建主成分回归模型和文本自动分类模型，实现对协作认知投入各子系统与各测量维度的量化测评。基于多模态表征结果，本研究使用K-means算法挖掘典型的协作认知投入模式，在184位参与者中识别出“规律建构型学习者”、“深度投入型学习者”、“积极感知型学习者”三类典型模式。

第三个研究是协作认知投入的影响路径分析。本研究将自我决定理论作为影响路径分析框架，从基本心理需要的视角，确定了自主需要、能力需要、关系需要、先验知识作为协作认知投入的四项影响因素。本研究使用基本心理需要问卷与先验知识测试题对影响因素进行测量，使用协作认知投入的多模态表征模型计算每个子系统的水平，以上述184位学习者作为研究对象，通过结构方程模型与定性比较分析相结合的方法进行影响路径挖掘。研究发现，激活系统存在两条影响路径、加工系统存在四条影响路径、反应系统存在四条影响路径，各子系统的影响路径之间具有等效替代的特性。

第四个研究是协作认知投入的干预设计与实施。本研究针对协作认知投入的核心影响因素（自主需要、能力需要与关系需要），设计了对应的自主支持型任务单、问题提示脚本与引导式互动脚本三类干预策略。同时，本研究设计了差异化干预机制，可以根据学习者的协作认知投入模式为其匹配适切的干预方案。为了验证差异化干预机制的效果，本研究以北京市某高校的166位大学生学习者为研究对象开展了教育准实验。对照组1不施加任何干预，对照组2施加全部三种干预，实验组则实施本研究设计的差异化干预。多元方差分析结果显示，接受差异化干预的学习者在激活水平与反应水平上显著高于无干预组，在加工水平上则显著高于无干预组和全干预组，表明差异化干预机制更能契合学习者的实际需求，从而有效提升学习者的协作认知投入水平。

本研究最核心的创新点在于构建了能够有效反映协作认知投入多维度多层次内涵的概念模型与测量框架。在理论层面，概念模型可以深度剖析协作认知投入内涵，解释协作认知投入的发生机制，并对学习投入理论进行深化和拓宽；在方法层面，测量框架可以支持协作认知投入的多模态表征，从学习者心理、身体、社交等方面进行协作认知投入全方位、自动化地测评；在实践层面，概念模型能够有效解释不同类型的协作学习问题，表征结果则可以帮助教师精准了解学情，为教学设计和教学改进提供依据。

In the era of human-machine collaboration, the core task in the field of education is to train innovative talents with high-level literacy. The training of high-level literacy requires the implementation of process evaluation, and the analysis of learning engagement is the key issue to understand the learning process. Current studies on learning engagement focus too much on the individual behavioral engagement of learners. As an explicit engagement, it is difficult to reveal the mechanism that affects learning and to respond to the nature of learning socialization. Therefore, this study focused on exploring collaborative cognitive engagement. Collaborative cognitive engagement has implicit attributes. Teachers cannot intuitively understand the problems of learners’ collaborative cognitive engagement and make teaching improvements. Therefore, how to objectively monitor and intervene the collaborative cognitive engagement is particularly important. At the same time, the connotation of collaborative cognitive engagement is multi-dimensional and multi-level, which requires learners to participate in all aspects of psychology, body and social interaction. Therefore, the concomitant collection of multimodal data in collaborative learning brings new ideas to quantify collaborative cognitive engagement comprehensively and objectively.

Based on this, this study focused on the core issue of how to quantify the level of collaborative cognitive engagement by collecting multimodal data generated by learners in collaborative learning and to effectively intervene in it. Based on the logical order of “what”, “how to measure”, “why”, “how to improve”, this study conducted research from four aspects: conceptual model and measurement framework construction, multimodal representation, impact path analysis, intervention design and implementation. The four-part study is a progressive and progressive relationship. The conceptual model and measurement framework determine the components of collaborative cognitive engagement, and the multimodal representation achieves the quantitative evaluation of each element. The evaluation results can support the impact path identification of collaborative cognitive engagement. Interventions designed with impact path as the focus can promote the improvement of collaborative cognitive engagement.

The first study is the conceptual model and measurement framework of collaborative cognitive engagement. After clarifying the conceptual connotation of collaborative cognitive engagement, this study took cognitive presence, collective cognition and collaborative knowledge construction as the theoretical basis, and formed the APR(Activation—Processing—Response) conceptual model of collaborative cognitive engagement through theoretical deduction. The model consists of an activation system, a processing system and a response system. The three subsystems coordinate to maintain the occurrence of collaborative cognitive engagement. Based on the ARP conceptual model, this study clarified the measurement dimensions under each subsystem, and then evaluated and revised the framework through two rounds of Delphi method by 12 relevant domain experts to form the final measurement framework of collaborative cognitive engagement. The activation system includes two dimensions: arousal and adaptability. The processing system includes four dimensions: exploration, integration, reflection and support. The response system includes three dimensions: productivity, impact and innovation.

The second study is the multimodal representation of collaborative cognitive engagement. This study designed a collaborative problem-solving activity with the theme of “Sponge Campus Design”. 184 students from a university in Beijing were selected as the subjects to collect physiological data and conversation text data during the collaboration process. Based on the multimodal data generated by the learners, this study used signal analysis technology, deep learning word vector technology, and semantic analysis technology to extract multimodal data features, constructed principal component regression model and text automatic classification model respectively for activation system, processing system and response system, and quantified the subsystems and dimensions of collaborative cognitive engagement. Based on the results of multimodal representation, this study used K-means algorithm to mine typical collaborative cognitive engagement patterns, and identified three typical patterns among 184 participants: Regular Constructive Learner, Deep Engagement Learner and Active Perceptive Learner.

The third study is the impact path analysis of collaborative cognitive engagement. This study used self-determination theory as the framework of path analysis. From the perspective of basic psychological needs, this study identified four factors that impact collaborative cognitive engagement, which were autonomy needs, competence needs, relationship needs and prior knowledge. This study used the basic psychological needs questionnaire and a prior knowledge test to measure the impact factors, used the multimodal representation model of collaborative cognitive engagement to calculate the level of each subsystem, and took 184 above-mentioned learners as subjects, and used the method of combining structural equation model with qualitative comparative analysis to conduct impact path mining. It was found that there were two influence paths in the activation system, four influence paths in the processing system, and four influence paths in the reaction system. The influence paths of each subsystem have the characteristics of equivalent substitution.

The fourth study is the design and implementation of interventions for collaborative cognitive engagement. In this study, three types of intervention strategies were designed, namely, autonomy-supporting task list, question prompt script and guided interaction script, for the core influencing factors of collaborative cognitive engagement (autonomy needs, competence needs and relationship needs). At the same time, this study designed a differentiated intervention mechanism to match the appropriate intervention scheme according to the learner’s collaborative cognitive engagement pattern. In order to verify the effectiveness of the differential intervention mechanism, this study conducted a quasi-experiment among 166 college students in a university in Beijing. The control group 1 did not apply any intervention, the control group 2 did all three interventions, and the experimental group did the differentiated intervention designed by this study. The results of multivariate analysis of variance showed that the level of activation and response was significantly higher in the differentiated intervention group than in the non-intervention group, and significantly higher in the processing level than in the non-intervention group and the all- intervention group, indicating that the differentiated intervention mechanism can better meet the actual needs of the learners, thus effectively improving the level of collaborative cognitive engagement of the learners.

The core innovation of this study is to construct a conceptual model and measurement framework which can effectively reflect the multi-dimensional and multi-level connotation of collaborative cognitive engagement. On the theoretical level, the conceptual model can deeply analyze the connotation of collaborative cognitive engagement, explain the mechanism of collaborative cognitive engagement, and deepen and broaden the learning engagement theory. At the methodological level, the measurement framework can support the multimodal representation of collaborative cognitive engagement, which can be measured comprehensively and automatically from the aspects of learners’ psychological, physical and social aspects. In practice, conceptual models can effectively explain different types of collaborative learning problems, and characterizing results can help teachers accurately understand the learning situation, and provide a basis for teaching design and teaching improvement.