A CHAT-Anchored Learning Analytics Pipeline for AI Literacy Education

Synthesis: Moon (2026) develops a Cultural-Historical Activity Theory (CHAT)-anchored learning analytics pipeline that couples three facets — temporal participation, discourse quality, and concept sophistication — each mapped to CHAT activity-system elements, to support early detection and social-epistemic integration in small discussion-based AI literacy classes. Deployed in a five-week AI literacy course (n = 25; 438 posts), DTW-based k-medoids clustering identified two stable participation archetypes (validated via bootstrap silhouette analysis); discourse quality showed modest lexical gains while reasoning depth stagnated and overall concept sophistication declined (Δz = –1.24). An Isolation-Forest detector flagged Week 2 volatility two weeks before participation decline, and automated discourse-quality scores converged with human-coded ICAP levels. The study demonstrates how a theory-aware, effect-size-centered workflow delivers interpretable insights and early warnings in a single small class without network-reconstruction overhead.

Key Findings

• A CHAT-mapped analytics design. The pipeline maps three analytic facets to CHAT elements: temporal participation (Subject/Community), discourse quality (Division of Labor/Rules), and concept sophistication (Object/Tools). This grounds learning-analytics features in the activity system rather than treating them as generic engagement metrics.
• Two stable participation archetypes. DTW-based k-medoids clustering (with bootstrap silhouette validation) identified two archetypes in the 25-student, 438-post course. "Reserved Observers" showed persistent behavior–episteme mismatches, signaling a Subject–Community tension.
• Shallow gains, deep stagnation. Discourse quality showed modest lexical-polish improvements, but reasoning depth stagnated and overall concept sophistication declined (Δz = –1.24) — evidence that conversational activity did not translate into conceptual growth.
• Early-warning detection. An Isolation-Forest detector flagged Week 2 volatility two weeks before a participation decline, robust across contamination settings 0.05–0.25 — a practical early-warning signal for instructors.
• Validity via ICAP alignment. Automated discourse-quality indices converged with human-coded ICAP levels, supporting substantive alignment between the CHAT-anchored analytics and an established engagement taxonomy.

Implications for AI in Education

The paper is a methodological contribution showing that Learning Analytics can be theory-anchored — mapping measurement facets to an activity system — and still be feasible in a single small class (no large-sample statistics or full network reconstruction required). For practice, it offers instructors an interpretable, effect-size-centered workflow for tracking social-epistemic integration in discussion-based AI literacy courses and flagging at-risk participation early. It connects CHAT to the knowledge base's AI Literacy and ICAP concepts, positioning activity theory as an analytic lens for the learning-analytics side of AI in education.

Connected Concepts

• Activity Theory AIED
• Learning Analytics
• AI Literacy
• Icap Framework
• Student AI Interaction
• Higher Ed

Connected Articles

• Jiang GenAI Activity Theory Disciplines 2026 — Disciplinary differences in GenAI use and disclosure through an activity theory lens

Citation

Moon, J. (2026). A cultural-historical activity theory-anchored learning analytics pipeline for early detection and social-epistemic integration in AI literacy education. Interactive Learning Environments (in press).