Ordered Network Analysis of Epistemic Emotions during Collaborative Problem Solving

Synthesis: Anindho, Venkatesha, Ocumpaugh and Blanchard apply Ordered Network Analysis to trace how epistemic emotions such as confusion and frustration persist and transition during co-situated collaborative problem solving. The work advances affect-aware learning analytics by modeling the temporal ordering of emotional states rather than static frequencies, informing when interventions should trigger in Affective Tutoring systems and Multimodal AI detectors like An Interpretable Closed-Loop Intelligent Tutoring System for Multimodal Affective Feedback in Asynchronous Presentation Training. It grounds affect dynamics in Collaborative Learning contexts, complements sensor-based approaches like A Formative Study of Brief Affective Text as a Complement to Wearable Sensing for Longitudinal Student Health Monitoring and video-based EduGage: Methods and Dataset for Sensor-Based Momentary Assessment of Engagement in Self-Guided Video Learning.

Key Findings

• Ordered Network Analysis traces how epistemic emotions such as confusion and frustration persist and transition during co-situated collaborative problem solving.
• Modeling the temporal ordering of emotional states (rather than static frequencies) reveals how affect unfolds over time.
• The analysis informs when interventions should be triggered in Affective Tutoring systems.
• The approach complements sensor-based (A Formative Study of Brief Affective Text as a Complement to Wearable Sensing for Longitudinal Student Health Monitoring) and video-based (EduGage: Methods and Dataset for Sensor-Based Momentary Assessment of Engagement in Self-Guided Video Learning) affect detection by grounding emotions in collaborative-learning contexts.

What this means for practice

• Designers. Trigger affect-aware responses on ordered trajectories of confusion and frustration rather than isolated signals or summary frequencies, because the models show short-range persistence and directional transitions instead of instantaneous states.
• Distinguish productive struggle from disengagement in collaborative tasks: slower groups showed sustained confused–conflicted connections, while faster groups showed frequent but fleeting disengagement that may indicate one or two members carrying the work.
• Treat probe-caught and self-caught reports as different measurements — probe-caught networks showed stronger self-loops around curiosity, optimism, and confusion, whereas self-caught reports clustered around frustration, surprise, and conflict — and log which mechanism produced each report.
• Timestamp every affect report and preserve the raw event sequence, since the analysis depends on ordering within a moving window rather than on totals.
• Researchers. Triangulate retrospective cued-recall reports with physiological or behavioral signals, because recall bias and temporal imprecision feed directly into the modeled ordering of affective states.

Limitations

• The study has 27 participants organized into 9 groups of three, recruited from within the authors' own department, with a disproportionately male sample (18 of 27), which the authors flag as limiting generalizability.
• Affective states were drawn from a fixed set of seven predefined labels (confused, curious, frustrated, disengaged, optimistic, surprised, conflicted) reported during retrospective cued-recall, so the analysis inherits recall bias and temporal imprecision.
• Reporting mechanisms were imbalanced (64% probe-caught versus 36% self-caught), which the authors note may influence the relative density of connections between states.
• All nine groups solved the Weights Task correctly, so the study carries no measure of learning or solution quality and cannot relate affective structure to performance.

Citation

Sifatul Anindho, Videep Venkatesha, Jaclyn Ocumpaugh, Nathaniel Blanchard (2026). Ordered Network Analysis of Epistemic Emotions during Collaborative Problem Solving.