Related Experiment Video
Updated: Aug 5, 2026
08:43
A Study on an Intelligent Diagnosis and Treatment Assistant System for Acupuncture in Diminished Ovarian Reserve Based on a Knowledge Graph
Published on: May 29, 2026
Dual-Source Retrieval-Augmented Generation Chatbot for Women's Health (HerCare): Design and Multimethod Evaluation
Kimia Tuz Zaman1, Wordh Ul Hasan2, Nova Ahmed3
1Computer Science Department, North Dakota State University, 258 Quentin Burdick Building NDSU, 1320 Albrecht Boulevard, Fargo, ND, 58105, United States, 1 701-231-9662.
JMIR Formative Research
|July 31, 2026
Summary
HerCare, a novel AI for women's health, successfully integrates medical facts with personal stories, enhancing user trust and emotional support. This dual-source approach offers a safer, more resonant health AI experience.
Area of Science:
- Artificial Intelligence in Healthcare
- Human-Computer Interaction
- Women's Health Technology
Background:
- Existing conversational agents for women's health struggle to balance clinical accuracy with emotional resonance.
- This gap leads to either sterile advice or unreliable anecdotal information, failing to meet diverse user needs.
- A tension exists between ensuring factual safety and providing empathetic support in sensitive health contexts.
Purpose of the Study:
- To develop and evaluate HerCare, a conversational agent designed to address the limitations of current women's health AI.
- To implement a novel dual-source retrieval-augmented generation architecture integrating medical knowledge and peer narratives.
- To enhance user trust through transparent source attribution, making the provenance of responses visible.
Main Methods:
- A remote, web-based field study was conducted with 243 participants (women aged 18-45).
- Recruitment utilized social media and university mailing lists; IRB approval was obtained.
- A quantitative multimethod evaluation combined usability questionnaires (Chatbot Usability Questionnaire, NPS) with computational linguistic analyses (VADER, NRC Emotion Lexicon).
Main Results:
- High reported usability (median 78.1) and strong advocacy (NPS 60.0) were observed among participants.
- Post-interaction ratings indicated high satisfaction with helpfulness, ease of use, and clarity (median 4-5).
- Computational analysis showed a shift from neutral/negative user queries to strongly positive agent responses, featuring an empathy pattern (validate-then-redirect).
Conclusions:
- The dual-source architecture of HerCare is associated with high perceived empathy and trust, effectively combining clinical accuracy with emotional support.
- Formative findings suggest the feasibility of integrating clinical sources with lived experiences for safer and more resonant health AI.
- The study identifies a promising design pattern for future empathy-attuned health AI systems, warranting further controlled evaluation.