Epistemic network analysis shows metacognitive behaviors co-occur with invitation actions in DR and DE but with human-initiated reasoning and elaboration in CI¶

Claim · All claims
Evidence · 1 study · associational r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? In DR, metacognitive behaviors co-occurred with invitation-related actions, whereas in CI they were linked to human-initiated reasoning and elaboration; a similar invitation-linked pattern appeared in DE. → Zhanxin Hao 2026

Evidence¶

Zhanxin Hao 2026¶

Zhanxin Hao, Xiaobo Liu, Jiaxin Fan, Yun Long, Jifan Yu, Wenli Chen, Yu Zhang. (2026). Unpacking Interaction Profiles and Strategies in Human-AI Collaborative Problem Solving: A Cognitive Distribution and Regulation Perspective. https://arxiv.org/abs/2603.21288 doi:10.48550/arxiv.2603.21288

q2 · i? · associational · r2

Epistemic network analysis of co-occurrence among coded behaviors (model fit Pearson correlation above 0.96), with pairwise Welch t-tests on the MR1 dimension showing significant separation among all three clusters (e.g., C1 vs C2 d = 1.91). Network subtractions showed "metacognitive behaviors in DR were more likely to co-occur with invitation-related actions."

"metacognitive behaviors in DR were more likely to co-occur with invitation-related actions (e.g., reasoning invitation, elaboration invitation), whereas in CI, they were more frequently linked to human-initiated reasoning and elaboration."

Discussion¶

Related Claims¶

• Semantic similarity between human and AI discourse is highest in the Delegated Reasoning profile and significantly lower in Concerted Interpretation — related
• The Delegated Reasoning group significantly outperformed the Concerted Interpretation group on collaborative problem-solving task performance — related
• Cluster analysis of student-AI dialogue identifies three distinct collaboration profiles: Delegated Reasoning, Concerted Interpretation, and Delegated Elaboration — related
• Human and AI discourse differ functionally: students lead invitations, connection, and regulation moves, while AI leads reasoning, coordination, and agreement — related
• Intrinsically motivated students form reflective-iterative metacognitive co-occurrence networks, but overall ENA centroid positions do not differ significantly between motivational groups — related
• Proactive AI agent dialogues showed stronger conceptual–adequate reasoning links while reactive dialogues showed stronger factual/off-task associations — related
• A secondary ENA adding ICAP engagement codes showed proactive interactions linked constructive engagement with adequate reasoning, while reactive dialogues emphasised procedural pairings — related
• The Most Improved profile showed an integrated AI-support pattern with strong ENA connections among coaching, modeling, and reflection and frequent model uptake in group discussion — related
• Epistemic network analysis provides a better account of temporal patterns between socio-cognitive events than coding-and-counting — related
• The Concerted Interpretation group reported significantly higher self-regulation than Delegated Reasoning, with no profile differences in co-regulation or socially-shared regulation — related