学习云空间中基于大数据的多模态学习者情感分析与归因研究

There are major challenges to solve the lack of sentiment in online learning, including grasping the relevance connotation of learners’sentiment comprehensively, obtaining the sentiment of learners accurately and diagnosing the causes of triggering sentiment. Therefore, this project aims to provide personalized sentiment guidance and evaluation for teaching and learning in learning cloud space, and studies multi-modal sentiment analysis and causality technologies of learners based on big data in education as well as its application. Integrating learning theory and cognitive psychology theory with big data technology deeply, this project mins the implicit relevance to construct a multi-dimensional and fine-grain sentiment model of learners, which is conductive to further study the geometric deep learning technologies oriented to multi-modal sentiment data sources so as to achieve sentiment perception. Subsequently, the causality expression of learners’ sentiment model is established based on structural causal model and deep learning technologies, and a high efficient reasoning mechanism of causality is also designed, so the fast diagnosis of learners’ sentiment causality is to be realized based on domain knowledge. On the above basis, combined with the technical accumulation and platform support by our research team, learning assistance of intelligent space and evaluation of teaching objective in terms of sentiment are established for testing the feasibility of research achievements, and perfecting them constantly, simultaneously providing a new application paradigm for network learning space. The implementation of this project will optimize learning ecology of “with sentiment to promote knowledge construction in order to mutually improve sentiment education and knowledge construction” in learning cloud space, which has important theoretical significance and practical application value to the development of “Three Links & Two Platforms”.

全面把握学习者情感关联内涵、精准获取学习者情感并诊断情感触发原因是解决在线学习情感缺失的主要挑战。本项目旨在为学习云空间中的教与学提供个性化的学习情感指导与评测，研究基于大数据的多模态学习者情感分析技术、归因技术与应用。项目深度融合学习理论、认知心理理论与大数据理论技术，挖掘隐含关联以构建多维细粒度学习者情感模型，研究面向多模态情感数据源的几何深度学习技术以及时准确实现学习者情感感知，基于结构化因果模型建立学习者情感因果网络及其量化模型，结合因果推断与深度学习设计高效归因推理机制，实现学习者情感原因的快速诊断。以此为基础，结合研究团队已取得的技术积累和平台支持，建立基于情感的智能空间学习辅助与教学目标评测，以检验项目研究成果的可行性并予以完善，同时为网络学习空间提供新的应用范式。项目的实施将优化云空间“以情促知，知情共育”的学习生态，对“三通两平台”的发展具有重要理论意义和实际应用价值。

本项目深度融合学习理论、认知心理理论与大数据理论技术，研究面向多模态情感数据源的学习者情感及时准确感知，设计基于因果推断与深度学习的高效归因推理机制，实现学习者情感原因的快速诊断，并在云空间环境下进行实践应用验证。取得的主要研究成果如下：.1. 充分利用大数据及其适应性的分析技术，全面挖掘学习云空间中情感大数据隐含或间接的情感影响因素，并融入在线学习环境中各种交互过程，对模型概要结构与关联关系进行分析，最终形成包括学业情感与多维影响因素的四阶段情绪模型。.2. 针对学习云空间情感复杂上下文信息，提出多模态情感数据的有效分析方法，并设计基于模态异质性弥合与多模态动态图融合方法，实现学习者当前时域下的情感准确表达。.3. 依据情感大数据特点与学习云空间大规模数据处理需求，利用结构化因果方程、领域知识等手段，构建学习者情感因果网络图模型，并基于有限状态自动机、因果图神经网设计了因果演化分析方法，实现学习者情感的归因诊断。.4. 通过将情感分析方法扩展至学习者认知状态感知与识别中，项目组针对知识追踪、学习性能预测与以多模态为基础的核心技术突破等方面开展了研究，在保障学习者情感体验的前提下，实现了学习效果的提质增效。.5. 依托已有学习云平台，以情感教学、认知心理学理论为依托，在理论演绎和实践归纳基础上，设计基于情感补偿或指导的个性化学习辅助服务，实现情感理论技术成果的验证与应用。