在线学习因其自主化、多样化、个性化及灵活化等众多优势，已成为技术发展境域下不可或缺的一种学习方式。但由于在线学习的时空不受限，教师无法实时监管学生并及时调整教学计划，师生间情感交流的缺乏也极易导致学生出现学习不专注的现象，严重影响了学习效果。因此，通过学习干预的方式解决学习者在线学习中的不专注问题具有十分重要的意义。目前针对在线学习专注度的干预存在难以实时监控、未关注问题成因、实施尚不系统化等弊端，无法实现对学习者专注度的个性化、精准化干预。为能帮助教师及时监控并提升学习者专注度，本研究试图依据“问题识别”“策略设计”“模型构建”“系统呈现”的逻辑顺序，从在线学习场景中学习者的专注状态识别、学习专注度的干预策略设计、有效实现专注度干预的模型构建及在线学习专注度干预系统呈现等四个方面开展研究，助力精准专注度干预的全面实现。

首先，基于学习者的多模态数据进行专注度识别，通过E4智能手环采集学习者血容量脉冲、心率、心跳间期、皮肤电活动、皮肤温度等五种生理信号数据，遵循特征工程的方法进行专注状态识别：在特征构建阶段，基于以往研究中提及的各类生理信号与人体神经系统活动的关系，构建了来自五种信号的28个特征；在特征提取阶段，将每个特征与数据指标进行关联，完成各特征的量化工作；在特征选择阶段，以基于脑电数据的二分类专注度等级为识别对象，基于相关分析法对特征进行筛选，剔除了与专注等级相关性较低的9个特征，保留19个特征；最后，运用较为常用的七种分类算法进行特征评估，分析发现识别准确率均达到可接受的水平，其中支持向量机算法的准确率最高，可达75.86%，证实了所选特征的有效性。

其次，进行在线学习场景中专注度的干预策略设计。为能针对问题成因有的放矢地设计策略，探究在线学习场景中学习专注度的影响因素是必备前提：本研究根据扎根理论的基本原则对有关在线学习专注度下降成因的被试访谈文本进行了三级编码，梳理得出学习资源内容、身体原因、物理环境、元认知能力、学习动机、学习情绪等六类主要影响因素，并通过问卷数据进行了验证；而后依据各因素的可干预性进行筛选与维度划分，通过相关分析、线性回归分析、路径分析等方法，得出正性情绪、负性情绪与元认知策略能力等因素会对专注度产生直接影响，而追求成功的动机、避免失败的动机及元认知监控能力则会通过影响以上因素间接影响专注度。结合各类影响因素及其对专注度的影响机制，本研究设计了三类干预策略：学习内容支持干预，包含线索引导、剪裁策略、详细指导等策略；元认知能力提升干预，包含弹窗提示、策略贴士、笔记大纲等策略；学习动机增强干预，包含提供课程目标、电子徽章、弹出题目等策略。

而后，根据实现精准学习干预的基本流程，构建了由学习者多模态数据收集、问题诊断、策略匹配、策略实施、效果验证等主要环节构成的学习干预系统框架，并基于此设计了在线学习场景中专注度的干预模型：在学习者多模态数据收集环节，采集学习者的基础数据与生理数据，以支持模型的各环节运行；在问题诊断环节，划定专注问题的分析单元，并基于先前分析所得的19个生理信号特征判断各单元的专注状态；在策略匹配环节，基于已构建的专注度干预策略库，依照“干预-反应模型”的思路设计了逐层递进式的策略匹配方式，形成干预策略匹配机制；策略实施环节中，则结合各干预策略的特性及专注等级确定各策略的实施方式，形成面向每一位学习者的干预策略链；在效果验证环节，根据有限状态机的思路渐进式地优化干预措施，力求精准解决专注问题。各环节各司其职，共同构成指向精准专注度干预的闭合回路。

最后，基于前述研究成果，依托eCloud智慧教育云平台设计了专注度干预系统。采用德尔菲法了解了中小学一线教师对该系统的功能需求后，分别针对系统教师端的干预设计、学情查看等功能及学生端的干预接受、学情查看等功能进行了呈现设计，并通过Axure RP9产出系统原型，基于认知走查法、启发式评估等方式征询了相关专家及一线教师的建议，对系统进行了迭代修正。此后通过问卷调查法与用户测试法从系统设计与系统使用两个方面对该系统进行了评估，结果显示用户对该系统的架构设计、策略设计、系统设计有效性及系统使用意象等方面的主观评价均较高，且用户使用系统的有效性、效率与满意度等均达到较好的水平，证明了该系统的应用价值。

本研究对提升在线学习场景中学习者的专注度具有一定的理论与现实意义，未来可基于对专注度的多模态表征及特征规律的深入探究、专注问题的动态归因与用户的实际应用反馈等进一步完善模型与系统，以提升研究成果在实际学习场景中的推广应用价值。

Online learning has become an indispensable learning style in the context of technological development because of its independence, diversification, individuality, flexibility and many other advantages. However, due to the unlimited time and space of online learning, teachers are unable to supervise students in real time and adjust teaching plans in a timely manner. The lack of emotional communication between teachers and students is also easy to lead to the phenomenon of students not focusing on learning, which seriously affects the learning effect. Therefore, it is of great significance to solve the problem of learners' inattention in online learning through learning intervention. At present, the intervention of online learning concentration has some disadvantages, such as it is difficult to monitor in real time, the causes of problems not paid attention to, and the implementation is not systematic. Therefore, personalized and precise intervention of online learning concentration cannot be realized. In order to help teachers timely monitor and improve learners' concentration, this research attempts to follow the logical order of "problem identification", "strategy design", "model construction" and "system presentation", carrying out from the following four major aspects: the identification of learners' concentration state in the online learning scenario, the design of intervention strategies for learning concentration, the construction of effective intervention model and the presentation of online learning concentration intervention system, so as to facilitate the comprehensive realization of accurate concentration intervention.

Firstly, this study conducted concentration recognition based on learners' multi-modal data.  Five physiological signal data including blood volume pressure, heart rate, inter-beat intervals, electrodermal activity and skin temperature were collected by E4 smart bracelet, and concentration state recognition was carried out in accordance with feature engineering method. In the feature construction stage, 28 features from five kinds of signals were constructed based on the relationship between physiological signals and human nervous system activities mentioned in previous studies. In the feature extraction stage, each feature was associated with the data index to complete the quantitative work of each feature. In the feature selection stage, the dichotomous concentration level based on EEG data was used as the recognition object, and the features were selected based on correlation analysis method. 9 features with low correlation with concentration level were removed, and 19 features were retained. Finally, seven commonly used classification algorithms were used for feature evaluation, and it was found that the recognition accuracy reached an acceptable level, among which the support vector machine algorithm had the highest accuracy, which can reach 75.86%, confirming the effectiveness of the selected features.

Secondly, this study designed intervention strategies for concentration in online learning scenario. In order to design strategies aiming at the causes of problems, there is a necessary prerequisite to explore the influencing factors of learning concentration in online learning scenario. According to the basic principles of grounded theory, the research carried out three-level coding of the interview texts related to the causes of decreased online learning concentration, summarized six main influencing factors including physical environment, physical causes, resource content, metacognitive ability, learning motivation and learning emotion, and verified them through questionnaire data. Then, according to the interferability of each factor, screening and dimension classification were carried out. Through correlation analysis, linear regression analysis and path analysis, it was concluded that positive emotion, negative emotion and metacognitive strategy use ability had a direct impact on concentration. However, the motivation to pursue success, motivation to avoid failure and metacognitive monitoring ability indirectly affect concentration through influencing the above factors. Three types of intervention strategies were designed based on various influencing factors and their influencing mechanisms on concentration: learning content support intervention, including cue guidance, clipping strategy and detailed guidance; Metacognitive ability improvement interventions, including pop-up hints, strategic tips and notes outline; Learning motivation enhancement interventions, including providing course objectives, electronic badges and pop-up questions.

Then, based on the basic process of realizing precision learning intervention, this study constructed a learning intervention system framework consisting of learners' multi-modal data collection, problem diagnosis, strategy matching, strategy implementation, and effect verification, and designed an intervention model of concentration in online learning scenario based on the framework. In the process of multi-modal data collection of learners, basic data and physiological data of learners were collected to support the operation of each link of the model. In the process of problem diagnosis, the analysis unit of concentration was defined, and the state of concentration of each unit was judged based on the 19 physiological signal characteristics obtained from the previous analysis. In the strategy matching process, based on the established concentration intervention strategy library, the strategy matching method was designed step by step according to the idea of " response to intervention model ", forming the intervention strategy matching mechanism. In the strategy implementation process, the implementation mode of each strategy was determined based on the characteristics and concentration level of each intervention strategy, and the chain of intervention strategies for each learner was formed. Finally, in the effect verification process, the intervention measures were gradually optimized according to the finite-state machine idea, striving to accurately solve the concentration problem. Each link plays its own role, which together constitute a closed loop pointing to precise concentration intervention.

Finally, based on the aforementioned research results, this study designed an intervention system for concentration based on eCloud smart education cloud platform. After the Delphi method is adopted to clarify the system functional requirements of primary and secondary school front-line teachers, the presentation design was carried out for the functions of the system respectively, including the intervention design and learning condition view of the teacher side, and the intervention acceptance and learning condition view of the student side. After the Axure RP9 prototype was produced, cognitive walkthrough and heuristic evaluations were used to consult the relevant experts and teachers’ advice, and the system was modified iteratively. And then through the questionnaire survey and user testing method evaluated the system from two aspects of system design and use. The results showed that the users have high subjective evaluation on the system's architecture design, strategy design, system design effectiveness and system use image, and the effectiveness, efficiency and satisfaction of system using achieved a good level, which proved the application value of the system.

This study has a certain theoretical and practical significance to improve learners' concentration in the online learning scenario. In the future, the model and the system can be further improved based on the multimodal representations of concentration and features in-depth exploration, dynamic attribution of concentration problems and user's actual application feedback, so as to promote the popularization and application value for research results in the actual learning scenario.