Human-AI Co-Design Study for Stroke Rehabilitation Agents

A research paper details a Design-Based Research methodology for developing Embodied Conversational Agents (ECAs) in post-stroke rehabilitation. The study compares LLM response metrics like lexical diversity (MTLD ≈ 134 vs. 93.9) and repetition (Yule’s K ≈ 66.8 vs. 115.4) between synthetic and real user interactions. It reports usability scores from patient relatives (M = 86.67) and professionals (M = 72.50), and Intrinsic Motivation Inventory scores for both groups.

Use Cases

• Analyze lexical diversity (MTLD) and repetition (Yule’s K) metrics to compare LLM response quality between synthetic and real user questions.
• Compare usability scores between participant groups like patient relatives and rehabilitation professionals to assess interface acceptance.
• Evaluate Intrinsic Motivation Inventory scores to gauge user engagement with the proposed ECA system.
• Examine reported hapax rates and gender invariance in LLM responses for consistency and factual accuracy assessment.

Strengths

• Provides specific quantitative results, including lexical diversity (MTLD ≈ 134) and usability scores (M = 86.67).
• Compares data from two distinct participant groups: patient relatives and rehabilitation professionals.
• Methodology validated through a pilot deployment using the Furhat robot as an ECA.

Limitations

• The dataset is a 711.2 KB DOCX file containing a research paper, not a structured data table for direct analysis.
• Sample size for the pilot deployment with real participants is not specified, limiting statistical generalizability.
• Data is confined to a specific research context of post-stroke rehabilitation, not a broad application.

Provenance

Source

figshare, author Anna Lekova.

Collection Method

Design-Based Research methodology combining synthetic patient generation with pilot deployment using the Furhat robot.

Time Range

null

Freshness

Last updated March 26, 2026.

Geography

null