Silveira Inês, Varandas Rui, Gamboa Hugo
LIBPhys, NOVA School of Science and Technology, Largo da Torre, 2829-516, Caparica, Portugal.
PLUX Wireless Biosignals S.A, 1050-059, Lisboa, Portugal.
Comput Methods Programs Biomed. 2025 Sep;269:108863. doi: 10.1016/j.cmpb.2025.108863. Epub 2025 Jun 4.
Attention, cognitive workload/fatigue, and emotional states significantly influence learning outcomes, cognitive performance, and human-machine interactions. However, existing assessment methodologies fail to fully capture the multimodal nature of these cognitive processes, limiting their application in adaptive learning environments. This study presents the Cognitive Lab, a comprehensive multimodal dataset designed to investigate these cognitive processes across real-time learning scenarios. Specifically, it aims to capture and enable the classification of (1) attention and cognitive workload states using standard cognitive tasks, (2) cognitive fatigue arising from prolonged digital activities, and (3) emotional and learning states during interactive lessons.
The Cognitive Lab dataset consists of three distinct subsets, each developed through specific experimental scenarios targeting different aspects of learning. Dataset 1 comprises recordings from eight participants performing N-Back and mental subtraction tasks, aimed at assessing attention and cognitive workload. Dataset 2 includes data from 10 participants engaged in a digital lesson, complemented by Corsi block-tapping and concentration tasks, to evaluate cognitive fatigue. Lastly, Dataset 3 captures data from 18 participants during an interactive Jupyter Notebook lesson, focusing on emotional states and learning processes. Each scenario combined biosignals (accelerometry, ECG, EDA, EEG, fNIRS, respiration) with Human-Computer Interaction (HCI) features (mouse-tracking, keyboard activity, screenshots). Machine learning models were applied to classify cognitive states, with cross-validation ensuring robust results.
The dataset enabled accurate classification of learning states, achieving up to 87% accuracy in differentiating learning states using mouse-tracking data. Furthermore, it successfully differentiated attention, cognitive workload, and cognitive fatigue states using biosignal and HCI data, with fNIRS, EEG, and ECG emerging as key contributors to classification performance. Variability across participants highlighted the potential for subject-specific calibration to enhance model accuracy.
The Cognitive Lab dataset represents a resource for investigating cognitive phenomena in real-world learning scenarios. Its integration of biosignals and HCI features enables the classification of cognitive states and supports advancements in adaptive learning systems, cognitive neuroscience, and brain-computer interface technologies.
注意力、认知工作量/疲劳以及情绪状态会显著影响学习成果、认知表现和人机交互。然而，现有的评估方法未能充分捕捉这些认知过程的多模态性质，限制了它们在自适应学习环境中的应用。本研究展示了认知实验室，这是一个全面的多模态数据集，旨在研究实时学习场景中的这些认知过程。具体而言，其目的是捕捉并实现对以下方面的分类：(1) 使用标准认知任务的注意力和认知工作量状态；(2) 长时间数字活动引起的认知疲劳；(3) 互动课程中的情绪和学习状态。
认知实验室数据集由三个不同的子集组成，每个子集通过针对学习不同方面的特定实验场景开发而成。数据集1包含八名参与者执行N-回溯和心算任务的记录，旨在评估注意力和认知工作量。数据集2包括10名参与数字课程的参与者的数据，并辅以科西方块敲击和注意力集中任务，以评估认知疲劳。最后，数据集3在交互式Jupyter Notebook课程期间捕捉18名参与者的数据，重点关注情绪状态和学习过程。每个场景都将生物信号（加速度计、心电图、皮肤电活动、脑电图、功能近红外光谱、呼吸）与人机交互（HCI）特征（鼠标跟踪、键盘活动、屏幕截图）相结合。应用机器学习模型对认知状态进行分类，交叉验证确保了可靠的结果。
该数据集能够准确分类学习状态，使用鼠标跟踪数据区分学习状态时准确率高达87%。此外，它使用生物信号和HCI数据成功区分了注意力、认知工作量和认知疲劳状态，功能近红外光谱、脑电图和心电图成为分类性能的关键贡献因素。参与者之间的差异突出了进行个体特异性校准以提高模型准确性的潜力。
认知实验室数据集是研究现实世界学习场景中认知现象的一种资源。其生物信号和HCI特征的整合能够对认知状态进行分类，并支持自适应学习系统、认知神经科学和脑机接口技术的进步。
