Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Cog-TiPRO: Iterative Prompt Refinement with LLMs to Detect Cognitive Decline via Longitudinal Voice Assistant Commands

작성자
  • Haebom

Author

Kristin Qi, Youxiang Zhu, Caroline Summerour, John A. Batsis, Xiaohui Liang

Outline

This pilot study utilizes a voice assistant system (VAS) as a noninvasive method for the early diagnosis of cognitive decline. We collected voice command data from 35 older adults (15 of whom interacted with the VAS daily) over 18 months. To address the challenges of analyzing short, irregular, and noisy voice commands, we propose the Cog-TiPRO framework. Cog-TiPRO combines linguistic feature extraction through LLM-based repetition prompt refinement, HuBERT-based acoustic feature extraction, and Transformer-based temporal modeling. Using iTransformer, we achieved 73.80% accuracy and 72.67% F1-score for Mild Cognitive Impairment (MCI) detection, a 27.13% improvement over existing methods. The LLM approach identified linguistic features that characterize the daily command usage patterns of individuals experiencing cognitive decline.

Takeaways, Limitations

•
Takeaways:
◦
Suggesting the possibility of early diagnosis of cognitive decline using a voice assistant system.
◦
Contributes to overcoming existing difficulties with the possibility of non-invasive and continuous monitoring.
◦
Achieving high accuracy through the LLM-based Cog-TiPRO framework.
◦
New linguistic features characterizing cognitive decline discovered.
•
Limitations:
◦
The number of participants was limited as this was a pilot study.
◦
Generalizability to everyday voice command data needs to be verified.
◦
Further research is needed on different types and severities of cognitive decline.
◦
Scalability verification is required for various voice assistant systems and languages.
👍
Slashpage로 제작됨