To propose and evaluate a novel framework, Conversational Artificial Intelligence for Child Abuse Detection (CACAD), for automated counseling and detection of child abuse indicators.
Approach:
Framework Overview: CACAD consists of two stages: the first stage involves an LLM generating counseling questions with support from auxiliary modules, while the second stage analyzes the conversation for multilabel classification of abuse types.
Evaluation Method: The effectiveness of CACAD was evaluated through counseling sessions and abuse detection experiments using a child and adolescent counseling dataset from AI-Hub.
Key Findings:
CACAD demonstrated superior counseling quality and child abuse detection performance compared to all baseline models.
The framework integrates uncertainty quantification to flag cases for human review, enhancing reliability in abuse detection.
Interpretation:
Limitations:
LLMs may generate inaccurate predictions and inappropriate questions, posing risks in high-stakes counseling.
The reliance on automated systems may not fully replace the nuanced understanding of human counselors.
Recent CDC advisories, testing updates, and immunization recommendations highlight developments in infectious diseases, immunization, and diagnostic testing relevant to physicians across multiple specialties.
