Conversational AI for Child Abuse Detection Through Multistage Counseling: Model Development and Validation Study
-
By
-
Hyun-Young Moon
-
Youn-Gyu Jin
-
YoonJu Kim
-
Gwang-Cheol Lee
-
Hyeontaek Oh
-
Hyun A Kim
-
Dinara Aliyeva
-
Hyunjoo Na
-
Kang-Min Kim
-
July 10, 2026
-
Clinical Scorecard: Development and Validation of a Multistage Counseling Model Utilizing Conversational AI for Detecting Child Abuse
At a Glance
| Category | Detail |
|---|
| Condition | Child Abuse |
| Key Mechanisms | Conversational AI for counseling and multilabel classification for abuse detection. |
| Target Population | Children and adolescents experiencing or at risk of abuse. |
| Care Setting | Child and adolescent counseling environments. |
Key Highlights
- Proposes a novel framework, CACAD, to automate child abuse counseling.
- Integrates two auxiliary modules for flexible and safe question generation.
- Employs multilabel classification with uncertainty quantification for reliable abuse detection.
- Demonstrates superior counseling quality and detection performance compared to existing methods.
- Addresses the burden on human counselors in high-stakes situations.
Guideline-Based Recommendations
Diagnosis
- Utilize CACAD for initial counseling and abuse detection.
Management
- Flag cases exceeding uncertainty thresholds for human review.
Monitoring & Follow-up
- Evaluate the effectiveness of CACAD through human evaluation and quantitative experiments.
Risks
- Be aware of potential inaccuracies and inappropriate questions generated by LLMs.
Patient & Prescribing Data
Children and adolescents in need of abuse detection and counseling.
Automated counseling can alleviate counselor workload and improve detection rates.
Clinical Best Practices
- Incorporate conversational AI tools to enhance counseling efficiency.
- Ensure human oversight for cases flagged by the AI system.
Related Resources & Content
