user state modeling

Instructional designers often struggle to select optimal pedagogical interventions due to the lack of predictive models for learning outcomes. Method: We introduce the Human Learner Model (HLM)—the first unified computational learning model integrating cognitive modeling, knowledge tracing, and intervention-effect simulation—to predict learning gains from interventions (e.g., problem sequencing, item design) without human experimentation, and to generate theory-grounded, interpretable learning curves. Contribution/Results: HLM achieves high-accuracy prediction of real-world human A/B instructional experiment outcomes for the first time. It enables zero-shot learning curve generation and attribution analysis of intervention efficacy, revealing underlying psychological mechanisms. By bridging cognitive theory and educational practice, HLM provides an interpretable, scalable computational framework for instructional design—particularly valuable in data-scarce settings.

Existing personalized learning approaches treat student modeling, item selection, and feedback generation as disjoint components, resulting in coarse-grained student representations, adaptive assessments that neglect diagnostic posterior inference, and non-actionable feedback. To address these limitations, we propose EduLoop-Agent, an end-to-end personalized learning agent that establishes a closed-loop “diagnosis–recommendation–feedback” paradigm. It employs a neural cognitive diagnosis model (NCD) for fine-grained, interpretable knowledge-tracing; introduces bounded-effort cognitive adaptive testing (BECAT) to dynamically generate diagnostic-oriented items; and leverages large language models (LLMs) to produce actionable, weakness-targeted feedback. Evaluated on the ASSISTments dataset, EduLoop-Agent significantly improves question response prediction accuracy, item recommendation relevance, and feedback utility. To our knowledge, it is the first framework to achieve joint optimization across all three stages, realizing a fully integrated, closed-loop personalized learning system.

This study investigates the cognitive mechanisms underlying teachers’ design of multi-agent instructional workflows, reconceptualizing AI-TPACK (Artificial Intelligence–Technological Pedagogical Content Knowledge) as a dynamic cognitive practice rather than a static knowledge construct. Drawing on a mixed-methods approach that integrates cluster analysis and Markov chain modeling, the research analyzes behavioral logs from 61 teachers, 15 instructional artifacts, and 12 interviews to identify three distinct design archetypes: Systematic Optimizers, Prolific Creators, and Passive Observers. The findings elucidate how systems thinking, pedagogical beliefs, and self-efficacy dynamically shape the integration of AI-TPACK, offering empirical grounding for differentiated support strategies in professional development for intelligent educational technologies.

This work addresses the problem that recommender algorithms implicitly shape users’ digital identities and erode their autonomy. To mitigate this, we propose an interactive reflective tool designed to enhance algorithmic literacy. Methodologically, we introduce the novel “hypothetical inference” paradigm: leveraging large language models, the tool reconstructs platform algorithms’ semantic inferences from users’ fragmented behavioral data, generating interpretable, personalized explanations; we further pioneer the integration of temporal evolution visualization (i.e., time-series graphs) into algorithmic literacy tool design. Evaluated through a qualitative 14-participant human–computer interaction study, the tool significantly improves users’ critical awareness of algorithmic systems and their capacity for self-explanation. Our contribution lies in advancing explainable AI toward end-user operability—demonstrating a novel, black-box-agnostic approach that empowers users to reflect upon and autonomously regulate algorithmic influences without requiring access to proprietary algorithmic internals.

This work addresses a critical limitation in existing user simulators, which reproduce observable behaviors but fail to capture users’ underlying cognitive states—such as confusion or satisfaction—during search interactions. To bridge this gap, the study introduces, for the first time, a cognitively grounded approach to modeling user behavior from interaction logs. Leveraging information foraging theory and human expert judgments, the authors develop a multi-agent system capable of scalably inferring users’ cognitive trajectories from large-scale behavioral data. The proposed method significantly improves performance on downstream tasks, including conversational outcome prediction and user difficulty recovery. Furthermore, the authors release cognitive annotations and accompanying tools for widely used datasets such as AOL and Stack Overflow, establishing a new paradigm for more human-like user simulation and evaluation of retrieval systems.

This study addresses the challenges of assessing students’ comprehensive competencies in algorithm courses and the disconnect between academic instruction and industry needs. Grounded in the CC2020 competency model, it proposes a multidimensional assessment framework that integrates knowledge, skills, and professional dispositions. Behavioral data from programming assignments and written coursework of 169 students were collected using the xAPI specification. Learning behavior sequences were modeled via Markov processes, and cluster analysis was employed to identify distinct competency profiles. Additionally, a timeliness metric for submissions was introduced to quantify task difficulty. The framework not only enables computable representations of student competencies but also provides empirical support for personalized instructional interventions and curriculum refinement, thereby effectively bridging the gap between academic training and industry requirements.

This work addresses a critical limitation in existing knowledge tracing approaches, which model student interactions as a monolithic process and thereby overlook the distinct phases of competence acquisition and proficiency refinement, leading to biased representations of knowledge states. To remedy this, we propose Phase-Aware Knowledge Tracing (PAKT), the first framework to explicitly incorporate a dual-phase paradigm—separately modeling competence and proficiency. PAKT employs a tailored sequence decomposition mechanism to delineate learning phases and integrates a multi-branch Transformer architecture with a type-aware readout module to jointly capture phase-specific dynamics and holistic knowledge states. Through causal analysis, we identify and mitigate confounding bias arising from behavioral confounders in conventional models. Extensive experiments demonstrate that PAKT consistently outperforms state-of-the-art methods across six benchmark datasets, achieving average AUC gains of 0.82% and up to 1.33% in peak improvement.

This study addresses the challenge that existing learning support systems struggle to effectively model and intervene in learners’ knowledge monitoring—the metacognitive ability to accurately assess one’s own understanding. To bridge this gap, the authors propose the Capture-Calibrate-Coach (3C) framework, which systematically models knowledge monitoring for the first time. The approach constructs a heterogeneous learner–concept graph and employs graph neural networks to infer learners’ implicit knowledge awareness states, even when unreported. Personalized feedback is then generated based on five distinct metacognitive patterns. Evaluated on a cohort of 684 students, the method achieves an AUC of 85.21% in predicting implicit awareness states, significantly outperforming baseline models. Furthermore, a user study with 47 participants confirms that the generated feedback is highly regarded for its effectiveness in identifying knowledge gaps and offering actionable learning guidance.