Dual LLM Cross-Validation for Classifying and Analyzing Long COVID Symptoms on Korean Social Media

View at Publisher

Ask AI

Bookmark

Share

Key Points

This research aims to classify long COVID symptoms found in South Korean social media posts using dual LLM cross-validation.
Collected 5,555 posts using keywords related to long COVID from Naver Blogs and News.
Applied two independent large language models, Llama 3 8B and Qwen 2.57B, for classification into direct, indirect, and irrelevant categories.
Measured agreement rate and Cohen's Kappa for binary classification between the two models.
The agreement rate between models for binary classification was 80.7%, with a Cohen's Kappa of 0.614, indicating substantial agreement.
Among the 3,184 posts classified as directly relevant, 66.0% were confirmed symptoms, showcasing a significant information noise issue in social media health information.
The symptom extraction revealed 190 categories with fatigue (27.0%) and brain fog (10.1%) being notably prevalent, highlighting public concern over cognitive symptoms.

Study classifies long COVID symptoms on Korean social media, suggesting language models enhance analysis accuracy.

No abstract yet. The journal hasn't published one we can show.

No takes yet. Share an insight, caveat, or question.

JongHwi Song (2026) studied this question.

View at Publisher

Ask AI

Bookmark

Share