Description

This PhD project sits at the crossroads of artificial intelligence, cognitive science, and educational science. It aims to develop multimodal, personalized, robust, and explainable machine learning models to estimate learners' real cognitive effort in hybrid digital learning environments, particularly educational digital third places.
Cognitive effort plays a central role in learner engagement: it conditions the depth of information processing, the consolidation of knowledge, and the ability to transfer learning to new situations. However, cognitive effort is difficult to observe directly and is most often approximated through subjective measures (self-report questionnaires, mental workload scales), whose reliability is limited by perception biases, social comparison, or contextual factors. The literature distinguishes real effort i.e. the cognitive resources actually mobilised by the learner, from perceived effort, and the subjective perception reported by the individual. Those two dimensions that can significantly diverge.
Digital learning environments generate large amounts of interaction data: navigation logs, questionnaire responses, physiological signals (heart rate, electrodermal activity), and performance indicators. These heterogeneous data, still rarely exploited in an integrated manner, constitute rich material for estimating real cognitive effort in a more objective and personalised way.
Three main scientific challenges structure this project. The first consists in formalising a robust conceptual framework for real cognitive effort, clearly distinguishing real effort from perceived effort, integrating individual biases (profile, motivation, emotional state) and contextual factors (task, technical device, environment), and defining a coherent set of observable indicators suitable for use in AI models. The second challenge concerns the design of multimodal and personalised machine learning models capable of fusing subjective, behavioural, physiological, and performance data, capturing the temporal dynamics of effort, and maintaining a balance between accuracy, robustness, and computational complexity compatible with real educational deployment constraints. The third challenge involves integrating responsible AI principles from the very beginning of model design: providing explainability mechanisms interpretable by teachers and educational researchers, detecting and reducing potential biases (gender, academic level, socioeconomic context), and assessing the pedagogical and ethical impact of the produced indicators.

The work will be organised over three years: the first year dedicated to an interdisciplinary literature review, conceptual framework development, and data collection protocol design; the second year focused on developing and comparing multimodal model architectures (sequence models, deep multimodal architectures, latent-factor models) and integrating personalisation; the third year centred on explainability methods, bias analysis, mitigation strategies, and evaluation of indicator acceptability with educational stakeholders.

This project is aligned with the ambitions of the ENACT cluster and aims to produce concrete decision-support tools for educators that are more adaptive, fair, and transparent.

Skills required

The PhD candidate must hold (or be in the process of obtaining) a Master’s degree or an engineering degree in computer science, cognitive science, data science or mathematics. The candidate’s design and analytical skills, as well as their ability to work independently, are important for this thesis, as are development skills and a good ability to interact and listen (given the context of the thesis, which involves interactions with teachers and pupils)

Bibliography

Scariot, A. P., Andrade, F. G., Silva, J. M. C. da, & Imran, H. (2016). Students Effort vs. Outcome: Analysis Through Moodle Logs. In proc. ICALT.
Steele, J. (2020), ‘What is (perception of) effort? Objective and subjective effort during task performance’, PsyArXiv.
Moissa, B., Bonnin, G., & Boyer, A. (2021). Measuring and Predicting Students’ Effort: A Study on the Feasibility of Cognitive Load Measures to Real-Life Scenarios. In proc. EC-TEL.
Eeva S.H. Haataja, Asko Tolvanen, Henna Vilppu, Manne Kallio, Jouni Peltonen, Riitta-Leena Metsäpelto, Measuring higher-order cognitive skills with multiple choice questions –potentials and pitfalls of Finnish teacher education entrance, Teaching and Teacher Education, Volume 122, 2023.
Yuling Yang, Mingzhi Zhou, Jiangping Zhou, Re-understanding accessibility through a cognitive process: a conceptual framework and quantification, Applied Geography, Volume 186, 2026, 103835, ISSN 0143-6228.
Kazuhisa Takemura, A computer simulation of cognitive effort and the accuracy of two-stage decision strategies in a multiattribute decision-making process, Editor(s): Kazuhisa Takemura,In Perspectivs in Behavioral Economics and the Economics of Beh, Escaping from Bad Decisions, Academic Press, 2021, Pages 113-139.
Xingle Ji, Lu Sun, Kun Huang, The construction and implementation direction of personalized learning model based on multimodal data fusion in the context of intelligent education, Cognitive Systems Research, Volume 92, 2025.
Shuzhen Yu, Alexey Androsov, Hanbing Yan, Exploring the prospects of multimodal large language models for Automated Emotion Recognition in education: Insights from Gemini, Computers & Education, Volume 232, 2025.
Yan Huang, Wei Xu, Paisan Sukjairungwattana, Zhonggen Yu, Learners’ continuance intention in multimodal language learning education: An innovative multiple linear regression model, Heliyon, Volume 10, Issue 6, 2024.
Maira Klyshbekova, Gisela Reyes Cruz, Caitlin Bentley, Stef Garasto, Amy Aisha Brown, Christine Aicardi, Brian Ball, Mohammad Naiseh, Oana Andrei, A UK perspective on responsible education for responsible AI: a multidisciplinary review and evaluation framework, Journal of Responsible Technology, Volume 25, 2026.
Ming Ma, Davy Tsz Kit Ng, Zhichun Liu, Gary K.W. Wong, Fostering responsible AI literacy: A systematic review of K-12 AI ethics education, Computers and Education: Artificial Intelligence, Volume 8, 2025.

Keywords

Machine learning, Multimodal data, Real cognitive effort, Explainability