Introduction to the Proceedings

The International Conference on Educational Data Mining (EDM) is the annual flagship conference of the International Educational Data Mining Society. The 19th iteration of the conference, EDM 2026, is participating in the Festival of Learning 2026 in Seoul, Republic of Korea, from June 27 to July 3, 2026. This year’s theme is Educational Data Mining Across Borders: Bridging Disciplines, Contexts, and Perspectives. Educational Data Mining has reached a stage where methods and findings are relevant across multiple areas of research and practice. Many of the challenges in learning and teaching cannot be addressed from a single perspective, and solutions are strengthened by crossing disciplinary, contextual, and cultural boundaries. This year’s theme highlights work that integrates perspectives, links research and practice, and broadens the impact of educational data mining to support diverse and inclusive learning opportunities.

The scientific programming for EDM 2026 includes:

• Keynote talks by outstanding researchers in the field as part of the Festival of Learning 2026, including EDM’s keynote speaker Bongshin Lee (Yonsei University, Republic of Korea).

• A plenary Prof. Ram Kumar Educational Data Mining Test of Time Award talk by Radek Pelánek (Masaryk University, Czech Republic), along with a discussion panel.

• A Data Set Award talk by Dorottya Demszky (Stanford University, USA) and Heather Hill (Harvard University, USA).

• Tutorials and workshops as part of the Festival of Learning 2026.

• Nineteen paper presentation sessions with 1 or 2 parallel tracks.

• A poster presentation session.

• An industry track session.

• A doctoral student consortium.

This year, EDM received 138 submissions to the Full Papers track (10 pages), 114 to the Short Papers track (6 pages), and 102 to the Posters and Demos track (4 pages). The program committee accepted 32 full papers (23.2% acceptance rate), 24 short papers (21.1% acceptance rate), and 46 poster and demo papers (45.1% acceptance rate). The EDM 2026 Industry track fostered exchange between industry application research and basic research, and 3 papers were included in the industry track. Furthermore, EDM 2026 also continued its tradition of providing opportunities for young researchers to present their work and receive feedback from their peers and senior researchers. The Doctoral Consortium this year featured 19 participants.

We thank all the authors who submitted their work and the program committee members for their expert inputs. We thank the members of the organization committee for their leadership that made this conference possible. We thank the sponsors of EDM 2026 and the Festival of Learning 2026 for their generous support. And, a big thank you to the local organizing committee of the Festival of Learning 2026 who made this event memorable.

General Chairs

• Hiroaki Ogata (Kyoto University, Japan)

• Hyojeong So (Ewha Womans University, Republic of Korea)

• Young Hoan Cho (Seoul National University, Republic of Korea)

Program Chairs

• Anthony Botelho (University of Florida, USA)

• Maria Mercedes T. Rodrigo (Ateneo de Manila University, Philippines)

• Adish Singla (MPI-SWS, Germany)

Equity, Diversity, Inclusion, and Accessibility Chairs

• Anna Rafferty (Carleton College, USA)

• Jade Maï Cock (LIP6 Sorbonne Université, France)

Industry Track Chairs

• Jeeheon Ryu (Chonnam National University, Republic of Korea)

• Hae-Deok Song (Chung Ang University, Republic of Korea)

• Jeongyun Han (University of Seoul, Republic of Korea)

Poster & Demo Track Chairs

• Shashi Kant Shankar (Ahmedabad University, India)

• Irena Koprinska (The University of Sydney, Australia)

Doctoral Consortium Chairs

• Jaclyn Ocumpaugh (University of Houston, USA)

• Nathaniel Blanchard (Colorado State University, USA)

JEDM Track Chairs

• Guanliang Chen (Monash University, Australia)

• Carrie Demmans Epp (University of Alberta, Canada)

Workshop & Tutorials Chairs

• Jina Kang (University of Illinois Urbana-Champaign, USA)

• Benjamin Paaßen (Bielefeld University, Germany)

Awards Chairs

• Tanja Käser (EPFL, Switzerland)

• Nigel Bosch (University of Illinois Urbana-Champaign, USA)

• Mingyu Feng (WestEd, USA)

Web Chairs

• Paul Salvador Inventado (California State University Fullerton, USA)

• Hongming (Chip) Li, Student Volunteer (University of Florida, USA)

Proceedings Chairs

• Miguel Portaz (UNED, Spain)

• Ana Serrano Mamolar (University of Burgos, Spain)

IEDMS Officers

IEDMS Board of Directors

Program Committee

Senior Program Committee Members

Regular Program Committee Members

Empowering People to Learn: Inclusive Human-Data Interaction in an AI-Driven World

Data and AI increasingly shape how we make decisions, manage our health, understand the world around us, and learn. From wearable sensors and health trackers to personalized dashboards and AI-powered assistants, data has become deeply embedded in everyday life. Data visualization has long served as a powerful means of understanding data and a compelling medium for data-driven storytelling. In this talk, I introduce Human-Data Interaction (HDI) as an essential bridge between data, meaningful insight, and informed action, reflecting on the evolution of my research from data visualization toward a broader exploration of HDI. I will discuss how HDI can empower people to learn from data through reflection, sensemaking, and action. I will also highlight opportunities to help make these data experiences more accessible and inclusive for people with diverse abilities and needs.

Metrics for Evaluation of Student Models

This talk is based on the paper "Metrics for Evaluation of Student Models", initially published in Journal of Educational Data Mining (Vol. 7, No 2., 2015). Researchers use many different metrics for evaluation of performance of student models. The aim of this paper is to provide an overview of commonly used metrics, to discuss properties, advantages, and disadvantages of different metrics, to summarize current practice in educational data mining, and to provide guidance for evaluation of student models. In the discussion we mention the relation of metrics to parameter fitting, the impact of student models on student practice (over-practice, under-practice), and point out connections to related work on evaluation of probability forecasters in other domains. We also provide an empirical comparison of metrics. One of the conclusion of the paper is that some commonly used metrics should not be used (MAE) or should be used more critically (AUC).

The NCTE Transcript Data focuses on observations of teachers and students in 4th and 5th grade elementary mathematics classrooms. The anonymized transcripts include significant metadata, such as annotation of discourse moves for each turn, information about teacher background and classroom practices, and demographics. It has been used in a variety of applications, including exploring how features of teacher and teacher-student language promote inquiry and growth mindset as well as evaluating and building generative AI systems.

Many middle-school math problems are image-dependent: the diagram or graph carries essential information. This matters for intelligent tutoring and accessibility, where systems must reason over figures and also decline responsibly when figures are missing. We evaluate six contemporary multimodal large language models (MLLMs)—three reasoning models and three non-reasoning models—on 376 Illustrative Mathematics (IM) items labeled as image-role Required (the figure contains task-critical information not recoverable from text alone without added assumptions). Each model attempts every item three times with and without the figure under a shared prompt and scoring protocol. To reduce image-role label subjectivity, we classify items as not Required when they are solvable from text alone without additional assumptions. With images, the top-performing reasoning models achieve accuracy in the mid-50%, while non-reasoning models fall in the mid-30s to low-40s. Without images, models overwhelmingly refuse rather than guess, with only rare correct-by-chance answers. Models show moderate agreement on which items are solvable, and we release two benchmark subsets of items solved consistently across models. A qualitative audit of 83 items shows that visual misreading is the dominant failure mode for non-reasoning models, while reasoning models more often produce correct answers accompanied by adequate explanations. These results suggest tutoring systems should gate automated scoring and learner-model updates on visual-evidence availability and use scaffolds that require explicit visual-evidence binding before algebra. For accessibility, systems should treat no-image refusals as missing-context signals and elicit the figure or a structured description, enabling description-substitution experiments. We release code, prompts, and summary artifacts for replication. Code and data: https://osf.io/ct7bg/.

The present study examines the assistance dilemma of providing combinations of perceptual supports in math education by fitting a Bayesian knowledge tracing model (BKT) to an open-source, online order-of-operations intervention dataset. This online intervention dataset contained 688 U.S. middle schoolers’ accuracy data who solved order-of-operations problems in pre-test, intervention, post-test, and delayed post-test sessions. Combinations of spacing and color cues were embedded in intervention materials, alongside immediate correctness feedback and worked examples, creating nine between-subject cue conditions (congruent, incongruent, or neutral spacing and/or color). To examine the effect of perceptual cues on immediate performance and long-term learning, we trained a BKT variant, conditioning guess, slip, learn, and forget parameters on perceptual cue conditions. The estimated cue-specific BKT parameters suggest two main findings. First, the presence of congruent perceptual cues (color and/or spacing, e.g., 2 - 8 + 3 x 11) increased guessing rate and reduced slipping rate, suggesting that visually highlighting high-order suboperations facilitated students’ immediate problem solving. Second, students who underwent intervention with incongruent color cues (e.g., 2 - 8 + 3 x 11) showed reduced guessing rate but exhibited the highest learning rate and the lowest forgetting rate. These results imply that providing misleading visual highlights may serve as desirable difficulties, which hinder immediate performance but benefit long-term learning. Furthermore, the impact of perceptual cues on immediate performance and long-term learning indicated by BKT parameters approximately aligned with the patterns revealed by behavioral measures (i.e., intervention, post-test, and delayed post-test accuracy). Such consistencies demonstrate the feasibility of using the BKT framework to understand the assistance dilemma of perceptual supports. The code and data are available at this link.

Personalization has become a cornerstone of online learning platforms, offering tailored experiences that enhance learner engagement, satisfaction, and performance. Moving beyond one-size-fits-all approaches, personalized systems can provide flexible access, improved efficiency, and support both cognitive and non-cognitive development. In language learning, personalization affords increased motivation, self-efficacy, confidence, and technology acceptance, while reducing instructor workload. Despite these benefits, many language learning platforms remain non-personalized. This study explores the integration of personalization in an online language learning platform taking into account three learner variables: preference, proficiency, and engagement. Preference captures reported thematic and content choices, proficiency reflects performance-based knowledge levels, and engagement measures sustained interaction or dropout risk. By personalizing based on these learner variables, we aim to tailor learning materials that align with learners’ interests, match their abilities, and foster sustained participation. To our knowledge, this is the first study to investigate these three variables collectively in an online language learning platform, contributing to the advancement of personalized technology-enhanced language learning and personalized language learning. Across the three personalization variables, the experiments reveal that lightweight recommender systems outperform deep models for preference prediction, DeepIRT offers the strongest while interpretable proficiency estimates, and random survival forest, especially when informed with proficiency estimates, most effectively models engagement. In the context of NedBox as a use case, and in line with teachers’ views on meaningful personalization, these models enable homepage learning material recommendations driven by learner preference and engagement, as well as proficiency‑aligned exercise entry points. The source code is provided as a supplementary file for this submission.