在线协作学习是一种有效的教学方法，已广泛应用于教育领域。在线协作学习也称为远程协作学习，它被定义为一种教育活动，通过网络和资源让不同地点的学习者共同参与学习活动，了解掌握在线协作学习的交互情况能够帮助教师及时发现问题并提供帮助。目前，在线协作学习过程中学习者产生的交互文本数据量逐渐增多，单一、滞后的人工分析方法难以在短时间内发现学习者学习过程中的问题和困难，教师无法实时掌握并提供个性化的学习支持和反馈。因此，基于机器学习的学习分析技术逐渐被应用于在线协作学习交互分析领域，旨在对交互文本进行自动化分析和呈现，帮助教师和学习者实时掌握学习情况，改善在线协作学习效果。但是，传统的机器学习算法在应用上存在一定的缺陷，比如忽略数据之间的相关性造成误差从而影响结果的准确性。随着技术的发展，深度学习算法逐渐被应用于自动化学习分析技术研究领域，致力于更加全面、准确的反映学习者的学习行为过程和结果。目前已有研究大多从交互文本特征体现在线协作学习的交互情况，但文本特征类型较为单一，并且特征与任务内容关联较低，学习者仍然无法识别活动中主题内容的不足和问题。本研究旨在从在线协作学习任务主题角度，基于交互文本自动分析交互主题类别、提取交互主题特征、构建交互主题路径，并通过实证研究探究交互主题模型的应用效果。研究分为两个部分：（1）交互主题模型构建技术研究；（2）交互主题模型应用的实证研究。

（1）交互主题模型构建技术研究。本研究基于导师团队前期收集的“能力”任务和“问题解决”任务的交互数据作为训练集，采用BERT模型从主题类别、主题&认知、主题&元认知、主题&情感、主题&行为构建主题类别模型和多维度主题特征模型。另外，采用潜在狄利克雷LDA主题模型算法对在线协作学习交互文本进行主题抽取，构建交互主题路径图，呈现小组交互过程中的主题演变。

（2）交互主题模型应用的实证研究。实证研究共开展两个任务活动：“能力”和“问题解决”。研究邀请249人参与活动，实验历时5个月，共开展63组在线协作学习实验。实验设计分为三组实证研究，分别检验基于自动分析交互主题类别、交互主题特征、交互主题路径的在线协作学习方法的有效性。在“能力”任务中探究了基于自动分析交互主题类别的在线协作学习方法的有效性，在“问题解决”任务中分别探究了基于自动分析交互主题特征和基于自动分析交互主题路径的在线协作学习方法的有效性。通过分析学习者的协同知识建构、小组绩效、集体调节行为以及认知负荷验证交互主题模型应用效果的有效性。
      本研究的创新和特色体现在基于任务主题的分析角度关注在线协作学习中交互主题讨论结果和交互主题发展路径，从交互主题类别、主题特征和交互主题路径三个维度对交互文本特征进行自动化分析，避免单纯从交互频率给予学习者评价和反馈，为协作学习交互分析提供了全新的思路。另外，本研究通过对在线协作学习中交互主题的可视化分析和呈现，帮助教师实时、准确捕捉学习者讨论的主题内容，了解学习者任务主题的讨论进展，能够及时给学习者提供有针对性的学习支持，从而促进在线协作学习效果的提高。通过本研究结论发现基于自动分析交互主题类别、交互主题特征、交互主题路径的在线协作学习方法对学习者的协同知识建构、小组绩效、集体调节行为有积极影响，并且，不会给学习者造成额外的认知负荷。

        Online Collaborative Learning is an effective teaching method which has been widely used in the field of education. Online collaborative learning, also known as distance collaborative learning, is defined as an educational activity that involves learners in different locations through networks and resources. Understanding and mastering the interactions of collaborative learning can help teachers identify problems and aid in a timely manner. At present, the amount of interactive text data generated by learners in the process of online collaborative learning is gradually increasing. It is difficult for a single and lagging manual analysis method to identify problems and difficulties encountered during the learning process of learners in a short time. And teachers are unable to provide real-time and accurate learning support and feedback. Therefore, automated learning analysis technology based on machine learning has been gradually applied in the field of online collaborative learning interaction analysis to realize automatic analysis and visual presentation of interactive texts, helping teachers and learners to grasp the learning situation in real time, and improving the effect of online collaborative learning. However, traditional machine learning algorithms have some defects in application, such as resulting in errors and affecting the accuracy of the results due to ignoring the correlation between data. With the development of technology, deep learning algorithm has been gradually applied to the research field of automatic learning analysis technology, which is committed to reflecting learners' learning behavior process and results more comprehensively and accurately. At present, most of the existing studies reflect the interaction of online collaborative learning from the characteristics of interactive text, but the type of text features is relatively single, and the correlation between features and task content is low. So, learners still can’t identify the deficiencies and problems of the subject content in activities. From the perspective of online collaborative learning task topic, this research aims to automatically analyze interactive topic categories, extract interactive topic features and construct interactive topic paths based on interactive text, and explore the application effect of interactive topic model through empirical research. The research is divided into two parts: (1) The research on the construction technology of Interactive Topic Modelling; (2) The empirical research of application of Interactive Topic Modelling.

(1) Research on the construction technology of interactive topic model. Based on the interaction data of the "ability" task and "problem solving" task collected by the mentor team in the early stage as the training set. The natural language processing Bert algorithm is used to construct the topic category model and multi-dimensional topic feature model from topic category, topic & cognition, topic & metacognition, topic & emotion, topic & behavior. Meanwhile, the Latent Dirichlet LDA topic model algorithm is used to extract topics from the online collaborative learning interactive texts, showing the evolution process of group topic discussions.

(2) Empirical research on the application of Interactive Topic Modelling. The empirical research carried out two activities: "ability" and "problem solving". 249 people were invited to participate in activities. The experiment lasted for 5 months, and 63 groups of online collaborative learning experiments were carried out. The experimental design is divided into three groups of empirical studies, respectively verifying the effectiveness of the online collaborative learning method based on interactive topic categories, interactive topic characteristics, and interactive topic paths. In the "ability" task, the effectiveness of the online collaborative learning method based on the interactive topic category is explored, and in the "problem solving" task, the effectiveness of the online collaborative learning method based on the interactive topic feature and based on the interactive topic path is explored. The validity of the interactive topic modelling is verified by analyzing learners' collaborative knowledge building, group performance, socially shared regulated behavior and cognitive load.

The innovation and characteristics of this research are reflected in focusing on the interactive topic results and development paths in online collaborative learning from the perspective of task topic analysis. And this research realizes automatic analysis of interactive texts from the three dimensions of interactive topic categories, topic characteristics and interactive topic paths, which avoids giving learners evaluation and feedback based solely on the interaction frequency and provides a new idea for collaborative learning interaction analysis. In addition, through the visual analysis and presentation of interactive topics in online collaborative learning, this research helps teachers to accurately capture the topic content of learners’ discussions in real time, understand the progress of learners’ task topic discussions, and provide learners with timely and targeted information. Provided learning support, thereby promoting the improvement of online collaborative learning effect. Through the conclusion of this research, it can be found that the online collaborative learning method based on automatic analysis of interactive topic categories, interactive topic characteristics and interactive topic paths has a positive impact on learners' collaborative knowledge building, group performance and socially shared regulated learning, and it will not cause additional cognitive load to learners.