LLM-Based Response Generation for Korean Adolescents: A Study Using the NAVER Knowledge iN Q&A Dataset with RAG

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Adolescent Psychology

Background:

• Korean adolescents face unique mental health challenges.
• Existing support systems may lack cultural relevance and personalization.
• Large language models (LLMs) offer potential for scalable interventions.

Purpose of the Study:

• To develop and validate a retrieval-augmented generation (RAG) based LLM system for Korean adolescents.
• To create a culturally relevant dataset of adolescent concerns.
• To compare the effectiveness of RAG-based versus non-RAG LLM responses.

Main Methods:

• Collected 3,874 adolescent posts and expert responses (2014-2024) from NAVER Knowledge iN.
• Processed and categorized data by negative emotions and sources of worry.
• Implemented a RAG system using FAISS for retrieval and GPT-4o mini for generation, evaluating response quality.

Main Results:

• RAG-based LLM responses significantly outperformed non-RAG responses across all metrics.
• RAG responses provided more specific, empathetic, and actionable guidance.
• Family, academics, and peer relationships were key stressors, often co-occurring with depression and anxiety.

Conclusions:

• RAG-based LLMs show promise for culturally tailored adolescent mental health support.
• The system offers a scalable approach to personalized interventions.
• Future work should expand data and enhance conversational abilities for comprehensive care.