在线协作学习改变了学习者以往受限于时空的交互方式，有助于提升学习绩效，因而在教育领域受到了广泛关注。与此同时，在线协作学习过程中产生的交互文本数据也随之大量增加，而现有的交互分析方法存在明显的不足，如方法单一、分析滞后以及无法挖掘有效信息等，导致无法及时了解小组的知识掌握情况。另外，当前的在线学习评价方式更侧重于个人评价，且主要关注学习者个体的分数结果，没有考虑小组的认知状态，教师难以掌握小组整体的真实学习情况，无法提供有效的学习支持。针对上述问题，本研究采用自然语言处理技术、知识图谱技术和认知诊断技术对在线协作学习情境的小组认知状态进行自动分析和诊断，实现对小组认知水平的精准诊断，并基于此构建小组认知图谱，为每个小组提供认知层面的精准反馈与支持；同时，本研究通过开展准实验来检验基于小组认知图谱的在线协作学习方法的应用效果。

在小组认知图谱的构建技术方面，本研究包括两个步骤：教育知识图谱的构建和小组认知状态的诊断。对于教育知识图谱的构建，首先进行实体识别，采用自然语言处理技术和知识图谱技术来对在线交互文本进行数据处理，通过对比三种深度学习模型（BERT、BERT-LSTM-CRF、BERT-BiLSTM-CRF）在“教学设计”和“能力培养”两个任务的检测评估指标，选择效果最佳的BERT-BiLSTM-CRF模型来实现实体识别，两个任务的准确率分别为0.90、0.91；其次，进行关系识别，通过查询目标知识图谱中已存储的实体和关系，以实现关系识别；最后，进行知识图谱的存储和可视化，将实体识别和关系识别的结果存储并可视化于在线协作学习平台，形成教育知识图谱。而对于小组认知状态的诊断，首先结合教育知识图谱、教材及其他相关资料确定两个任务的认知属性；其次，根据认知属性及专家建议设计认知诊断测试项目，并经过多轮修订完善，再采用NCDM（Neural Network Cognitive Diagnosis Model）来诊断小组知识掌握情况，两个任务的准确率分别为0.92、0.91，并获取小组认知状态诊断结果，最后将诊断结果可视化于教育知识图谱，形成小组认知图谱，为每个小组提供反馈与建议。

为检验基于小组认知图谱的在线协作学习方法的应用效果，本研究共邀请了225名学生参与“教学设计”和“能力培养”两个任务的在线协作学习实验，共计75组，每组3人。本研究通过采用准实验法、问卷调查法和访谈法来检验采用传统的在线协作学习方法（控制组）、基于教育知识图谱的在线协作学习方法（实验组A）和基于小组认知图谱的在线协作学习方法（实验组B）对小组协同知识建构水平、小组绩效以及认知负荷等方面的影响，整个研究持续10个月。研究结果表明：（1）“教学设计”任务和“能力培养”任务中，基于教育知识图谱的在线协作学习方法在协同知识建构水平和小组绩效等方面均优于传统的在线协作学习方法，并且两种方法之间没有明显的认知负荷差异。（2）“教学设计”任务和“能力培养”任务中，基于小组认知图谱的在线协作学习方法在协同知识建构水平和小组绩效等方面均优于基于教育知识图谱的在线协作学习方法，而两种方法在认知负荷方面可能存在一定的差异。

本研究通过分析在线协作学习交互文本和小组认知诊断测试项目数据来实现小组认知图谱的构建，在技术层面上探索了认知图谱构建技术；在应用层面上拓宽了小组认知诊断技术在教育领域的应用，帮助学生及时掌握认知状态，为教师开展精准教学提供依据。同时，本研究发现基于小组认知图谱的方法能有效提升小组的协同知识建构水平和小组绩效，改善协作学习效果。本研究对改进在线协作学习、精准诊断小组认知状态、提升学生在线协作学习效果具有一定的参考意义。

Online collaborative learning has been widely concerned in the field of education because it changes the interaction mode of learners who are limited by time and space in the past and helps to improve learning performance. At the same time, the interactive text data generated in the process of online collaborative learning has also increased greatly, and the existing interactive analysis methods have obvious shortcomings, such as single method, lagging analysis and unable to mine effective information, resulting in the inability to timely understand the knowledge grasp of the group. In addition, the current online learning evaluation method focuses more on individual evaluation, and mainly focuses on the score results of individual learners, without considering the cognitive state of the group, which makes it difficult for teachers to grasp the real learning situation of the whole group and fail to provide effective learning support. To solve the above problems, natural language processing technology, knowledge graph technology and cognitive diagnosis technology were used in this study to automatically analyze and diagnose the group cognitive state in online collaborative learning situations, so as to achieve accurate diagnosis of the group cognitive level. Based on this, the group cognitive graph was constructed to provide accurate feedback and support at the cognitive level for each group. At the same time, this study conducted a quasi-experiment to test the application effect of online collaborative learning method based on group cognitive graph.

In terms of the construction technology of group cognitive graph, this study includes two steps: the construction of educational knowledge graph and the diagnosis of group cognitive state. For the construction of educational knowledge graph, entity recognition is carried out first, and natural language processing technology and knowledge graph technology are used to process the data of online interactive text. By comparing the detection and evaluation indexes of three deep learning models (BERT, Bert-LSTM-CRF, Bert-Bilstm-CRF) in the two tasks of "instructional design" and "ability cultivation", the Bert-BilstM-CRF model with the best effect is selected to realize entity recognition. The accuracy of the two tasks were 0.90 and 0.91 respectively. Secondly, relationship recognition is carried out by querying the stored entities and relationships in the target knowledge graph. Finally, the knowledge graph is stored and visualized, and the results of entity recognition and relationship recognition are stored and visualized on the online collaborative learning platform to form the educational knowledge graph. For the diagnosis of group cognitive state, the cognitive attributes of the two tasks were determined first by combining the educational knowledge graph, teaching materials and other relevant materials. Secondly, Cognitive Diagnosis test items are designed according to cognitive attributes and expert suggestions, and after several rounds of revision and improvement, Neural Network Cognitive Diagnosis Model (NCDM) is adopted to diagnose the knowledge grasp of the group. The accuracy of the two tasks is 0.92 and 0.91 respectively. The group cognitive status diagnosis results were obtained, and the diagnosis results were visualized on the educational knowledge graph to form the group cognitive graph and provide feedback and suggestions for each group.

In order to test the application effect of the online collaborative learning method based on group cognitive graphs, this study invited 225 students to participate in the online collaborative learning experiment of the two tasks of "instructional design" and "ability development", with a total of 75 groups of 3 students in each group. In this study, quasi-experimental, questionnaire and interview methods were used to examine the effects of traditional online collaborative learning methods (control group), online collaborative learning methods based on educational knowledge graph (experimental group A) and online collaborative learning methods based on group cognitive graph (experimental group B) on group collaborative knowledge construction level, group performance and cognitive load. The study lasted for 10 months. The results show that: (1) in both "instructional design" and "ability building" tasks, the online collaborative learning method based on educational knowledge graph is superior to the traditional online collaborative learning method in terms of the level of collaborative knowledge construction and group performance, and there is no significant difference in cognitive load between the two methods. (2) In the tasks of "instructional design" and "ability cultivation", the online collaborative learning method based on the group cognitive graph is superior to the online collaborative learning method based on the educational knowledge graph in terms of the level of collaborative knowledge construction and group performance, and there may be some differences between the two methods in terms of cognitive load.

By analyzing the interactive text of online collaborative learning and the data of group cognitive diagnostic test project, this study realized the construction of group cognitive graph, and explored the construction technology of cognitive graph at the technical level. At the application level, the application of group cognitive diagnosis technology in the field of education is expanded, which helps students to grasp the cognitive state in time, and provides a basis for teachers to carry out precision teaching. At the same time, this study found that the method based on group cognitive graph can effectively improve the level of group collaborative knowledge construction and group performance, and improve the effect of collaborative learning. This study has certain reference significance for improving online collaborative learning, accurately diagnosing group cognitive state, and enhancing students' online collaborative learning effect.