Research Themes

INtERACTION

Human–AI interaction in teaching and learning contexts.

ModeLLING

Computational and statistical models of collaboration and self-regulated learning, built from multimodal learning analytics.

View projects →

INTervention

Analytics feedback that supports reflection and self-regulation — and its impact in practice.

View projects →

Adoption

The psychological and social challenges of adopting AI and analytics in educational practice.

View projects →

Interaction

Human–AI interaction in teaching and learning contexts.

Enhancing Student Engagement and Knowledge Construction in Online Discussion Forums Through an AI-Generated Summary-Driven Learning Design

Contact: Xinyuan Hao

Research summary

Online discussion forums are widely used in higher education, yet sustaining meaningful participation and deep knowledge construction remains challenging. This research investigates the impact of an AI-generated summary-driven learning design on students’ engagement, knowledge construction, and social capital in asynchronous online discussions. Adopting a design-based research methodology across four iterative cycles, the study combines learning design interventions with interviews and focus groups to examine both learning processes and student experiences. Findings indicate that the AI-generated summary-driven learning design can increase posting, replying, and viewing activity, support more students in reaching higher levels of knowledge construction, and strengthen bridging social capital within the learning community. The work contributes empirical evidence on the educational value of AI-generated learning support and offers practical design principles for integrating generative AI into online learning environments.

Keywords

online discussion forums
generative AI
learning design
student engagement
knowledge construction
social capital

Selected publications

Hao, X., & Cukurova, M. (2025). Enhancing engagement and social capital through an AI-generated summary-driven learning design in online discussion forums. In K. Tammets, S. Sosnovsky, R. F. Mello, G. Pishtari, & T. Nazaretsky (Eds.), Two Decades of TEL: From Lessons Learnt to Challenges Ahead (EC-TEL 2025), Lecture Notes in Computer Science, vol. 16063. Springer. doi:10.1007/978-3-032-03870-8_16
Hao, X., & Cukurova, M. (2023). Exploring the effects of “AI-generated” discussion summaries on learners’ engagement in online discussions. In N. Wang, G. Rebolledo-Mendez, V. Dimitrova, N. Matsuda, & O. C. Santos (Eds.), Artificial Intelligence in Education. Posters and Late Breaking Results, Workshops and Tutorials, Industry and Innovation Tracks, Practitioners, Doctoral Consortium and Blue Sky (AIED 2023), Communications in Computer and Information Science, vol. 1831. Springer. doi:10.1007/978-3-031-36336-8_24

Modelling

Computational and statistical models of collaboration and self-regulated learning, built from multimodal learning analytics.

Enhancing Teacher Interventions by Designing an Adaptive Support for Online Collaborative Problem Solving with Large Language Models

Contact: Kester Wong

Research summary

Collaborative Problem Solving (CPS) is a multilevel process involving the perspectives of both individuals and groups, in which activated cognitive domain knowledge and the social aspects of interaction are dynamically engaged towards shared understanding (Luckin et al., 2017). Existing studies have explored the use of coaching (D’Mello et al., 2024) and social robots (Nasir et al., 2023) to support and improve aspects of CPS competency; however, just-in-time guidance on CPS processes and the alignment of such support to teaching practice remain underexplored. This research proposes an adaptive support for CPS involving scaffolding of problem-solving and collaborative processes, taking a data-driven approach to the design of an adaptive scaffold intervention. Explorations investigate the use of transformer-based models for CPS diagnosis, with these diagnoses and interventions subsequently compared to those produced by teachers as a preliminary step towards designing an adaptive scaffold intervention system that helps teachers better support students in CPS.

Keywords

collaborative problem solving
adaptive scaffolding
teacher support
multimodal audio data
process mining
agentic AI workflow

Selected publications

Wong, K., Bulathwela, S., & Cukurova, M. (2026). Towards an adaptive AI scaffold for developing student collaborative problem solving. Learning and Instruction. Accepted; pending DOI.
Wong, K., Shihui, F., Bulathwela, S., & Cukurova, M. (2026). Scaffolding reshapes dialogic engagement in collaborative problem solving: Comparative analysis of two approaches. Proceedings of the 16th International Learning Analytics and Knowledge Conference (LAK ’26), 227–238. doi:10.1145/3785022.3785053
Wong, K., Wu, B., Bulathwela, S., & Cukurova, M. (2025). Rethinking the potential of multimodality in collaborative problem solving diagnosis with large language models. In A. I. Cristea, E. Walker, Y. Lu, O. C. Santos, & S. Isotani (Eds.), Artificial Intelligence in Education (AIED 2025), Lecture Notes in Computer Science, vol. 15878 (pp. 18–32). Springer. doi:10.1007/978-3-031-98417-4_2
Wong, K., Bulathwela, S., & Cukurova, M. (2025). Explainable collaborative problem solving diagnosis with BERT using SHAP and its implications for teacher adoption. arXiv preprint. doi:10.48550/arXiv.2507.14584
Wong, K., Bulathwela, S., & Cukurova, M. (2025). Exploring human–AI complementarity in CPS diagnosis using unimodal and multimodal BERT models. arXiv preprint. doi:10.48550/arXiv.2507.14579

Intervention

Analytics feedback that supports reflection and self-regulation — and its impact in practice.

Supporting Students' Regulation of Learning in Small Group Collaboration in Higher Education: A Multimodal Learning Analytics Approach

Contact: Wannapon Suraworachet

Research summary

Effective collaboration requires awareness of group challenges and timely regulation. This research investigates how learning analytics can detect challenge moments in small-group discourse and support students' regulation through an NLP-based predictive model and a feedback dashboard. Findings from a real-world deployment show that analytics-driven feedback can raise students' awareness, particularly at the individual level, though group regulation remains limited with practical challenges. This work offers a theoretically grounded, empirically validated framework for detecting and supporting socially shared regulation of learning in authentic higher education settings.

Keywords

regulation of learning
collaboration
large language models
feedback
collaboration analytics
discourse
natural language processing

Visuals

DUTE class
Collaboration analytics feedback
Collaboration analytics feedback v2

Selected publications

Suraworachet, W., Seon, J., & Cukurova, M. (2024). Predicting challenge moments from students’ discourse: A comparison of GPT-4 to two traditional natural language processing approaches. Proceedings of the 14th Learning Analytics and Knowledge Conference, 473–485. doi:10.1145/3636555.3636905
Villa-Torrano, C., Suraworachet, W., Gomez-Sanchez, E., Asensio-Perez, J. I., Bote-Lorenzo, M. L., Martinez-Mones, A., Zhou, Q., Cukurova, M., & Dimitriadis, Y. (2025). Using learning design and learning analytics to promote, detect and support Socially-Shared Regulation of Learning: A systematic literature review. Computers & Education, 232, 105261. doi:10.1016/j.compedu.2025.105261
Suraworachet, W., Zhou, Q., & Cukurova, M. (2025). University students’ perceptions of a multimodal AI system for real-world collaboration analytics: Lessons learned from a case study. Journal of Computer Assisted Learning, 41(5), e70103. doi:10.1111/jcal.70103

Supporting SRL in Online Learning through LA-Driven LLM Feedback: Examining Learner Perceptions

Contact: Mutluhan Yilmaz

Research summary

This doctoral research investigates the role of LLM-generated, trace-based feedback in supporting self-regulated learning (SRL) skills in online higher education. Drawing on learning analytics data, it compares LLM- and tutor-generated feedback across multiple feedback instances over a term, moving beyond single-session designs to examine longitudinal SRL development. Both the quality of the feedback produced and learners’ perceptions of it are examined, contextualising its potential instructional value. Findings are expected to contribute to the design of scalable, high-quality feedback interventions in online learning contexts.

Keywords

self-regulated learning
feedback
online learning
large language models
learning analytics

Selected publications

Yilmaz, M., Gauthier, A., & Cukurova, M. (2026). Supporting online learners’ regulation skills with the help of learning analytics and generative artificial intelligence. Educational Technology Research and Development, 1–30.
Yilmaz, M., Temur, H. B., Emmungil, L., Çelik, E., Gauthier, A., & Cukurova, M. (2026). Supporting self-regulated learning through generative AI feedback in online higher education: The importance of student perceptions of the source of feedback. International Journal of Educational Technology in Higher Education, 23(1), 16.

Scaffolding Students’ Engagement with Learning Analytics Feedback in Higher Education

Contact: Yildiz Uzun

Research summary

Learning analytics feedback, frequently delivered through dashboards and visualisations, holds promise for supporting self-regulated learning (SRL) in higher education, yet students often struggle to interpret and act upon such feedback. This research investigates how scaffolding through a generative AI assistant can mediate students’ engagement with analytics feedback. A first strand of empirical work establishes the relationship between engagement with analytics feedback, SRL competence, and course performance, providing the foundation for design. A second strand analyses the dialogues that emerge between students and a GenAI assistant when interpreting analytics visualisations, characterising the questions students raise and the response patterns that productively support sense-making. Together, the work contributes design principles for AI-scaffolded analytics feedback that bridges visualisation and learner interpretation, with implications for integrating generative AI into feedback infrastructures in higher education.

Keywords

scaffolding
engagement
GenAI assistant
self-regulated learning
learning analytics feedback
visualisations

Selected publications

Uzun, Y., Gauthier, A., & Cukurova, M. (2026). What students ask, how a generative AI assistant responds: Exploring higher education students’ dialogues on learning analytics feedback. Proceedings of the 16th International Learning Analytics and Knowledge Conference (LAK ’26), 42–52. doi:10.1145/3785022.3785028
Uzun, Y., Suraworachet, W., Zhou, Q., Gauthier, A., & Cukurova, M. (2025). Engagement with analytics feedback and its relationship to self-regulated learning competence and course performance. International Journal of Educational Technology in Higher Education, 22(1), 17. doi:10.1186/s41239-025-00515-3

Adoption

The psychological and social challenges of adopting AI and analytics in educational practice.

Benchmarking and Mitigating Gender Bias Through Scaffolding Interaction in Educational Applications of LLMs

Contact: Yishan Du

Research summary

Large language models are rapidly entering educational workflows, including teacher-facing tasks such as formative feedback and assessment, where implicit and context-dependent gender bias may surface through subtle shifts in tone, emphasis, and pedagogical framing with cumulative downstream consequences for students’ motivation, academic identity, and long-term trajectories. Two gaps remain in the field: a lack of robust, replicable, and interpretable benchmarking methods that detect subtle bias beyond template-based audits; and limited empirical evidence on how teacher–LLM interaction co-constructs, amplifies, or corrects model tendencies. This work therefore advances a two-strand programme connecting technical bias benchmarking with interactional bias co-construction. The first strand develops an education-grounded auditing framework using counterfactual educational inputs that differ only in gender cues, benchmarking representative LLMs on authentic student-writing feedback tasks via embedding-space mapping while controlling for stochasticity, with complementary qualitative and text-analytic interpretation. The second strand brings teachers into the loop through a workshop-style behavioural study under no-AI versus AI-assisted conditions, capturing decision changes, confidence shifts, adoption/editing behaviour, and bias indices, towards a taxonomy of human–AI bias interaction regimes.

Keywords

gender bias
large language models
educational fairness
counterfactual auditing
teacher–AI interaction
formative feedback
bias benchmarking

Selected publications

Du, Y., Borchers, C., & Cukurova, M. (2025). Benchmarking educational LLMs with analytics: A case study on gender bias in feedback. arXiv preprint arXiv:2511.08225. arXiv:2511.08225

Work with us

Interested in our research themes?

We welcome PhD students and collaborators who want to push these questions forward.

m.cukurova@ucl.ac.uk See recent publications →