Xie, D., Anderson, K., Eze, T., Lin, C., Shin, B., & Worsley, M. (2026) — Northwestern University. AIED 2026.
Key Finding
Agentic analytics using AI-produced concept-map artifacts as shared human-AI representations improves collaboration quality analysis and AI response grounding over transcript-only baselines.Synthesis
CLARA introduces a novel architecture for learning-analytics where AI-produced artifacts (concept maps, seven-dimension collaboration assessments) serve as shared representations between human dashboard users and AI reasoning agents. By indexing these artifacts into vector databases, CLARA establishes a human-AI common ground that simultaneously scaffolds human interpretation of collaboration data and grounds AI reasoning — improving both retrieval performance and response quality. This dual-use architecture has implications beyond collaboration analytics: the principle of AI-produced artifacts as shared infrastructure could apply to intelligent-tutoring-systems where student models, knowledge-tracing-irt estimates, and concept maps could serve as common ground between tutor agents and teacher dashboards. The artifact-as-knowledge-infrastructure approach also resonates with retrieval-augmented-generation patterns used in retrieval-augmented-tutoring-algorithm-kite and cyberscholar-genai-writing-feedback, where structured content representations improve AI response quality. CLARA's focus on semantic dimensions beyond behavioral signals parallels the shift advocated by ai-tutor-behavioral-evaluation toward evaluating what students actually do, not just what the AI says.Related Pages
- expert-cognition-dashboard — ECD extends dashboard design from visualization to embedded expert cognition reasoning