Personalized insights into liver disease management: a text mining analysis of online consultation data.

BACKGROUND

Liver diseases pose a significant global health burden with complex management challenges. Online health consultation platforms provide a valuable resource of unstructured patient-physician interactions. This study applies an integrated text mining framework to extract insights from this data, aiming to inform liver disease research and care strategies.

METHODS

We analyzed 8,149 liver disease-related online consultation records from a leading Chinese health platform. The analytical framework integrated KeyBERT-enhanced keyword extraction with traditional approaches (TF-IDF, TextRank), BERT-CRF medical entity recognition, topic modeling (LDA), and association rule mining. Expert validation by hepatology specialists provided clinical verification of extracted patterns. Stratified analyses across demographic factors and disease types identified subgroup-specific patterns.

RESULTS

Text mining analyses demonstrated robust performance in medical terminology extraction (KeyBERT F1-score: 0.87), identified key topic patterns in liver disease consultations through enhanced entity recognition (F1-scores: 0.89-0.91), and revealed significant clinical associations through comprehensive rule mining (lift: 2.2-4.5). Stratified analyses further highlighted notable demographic variations in disease patterns and progression pathways.

CONCLUSION

This study validates the effectiveness of integrated text mining approaches in uncovering clinically relevant patterns from online consultation data, with particular strength in medical entity recognition and association detection. The robust methodological framework provides empirical support for differentiated approaches in liver disease management, while demographic variations in disease patterns underscore the necessity for personalized clinical strategies. However, translation of these findings into clinical practice requires longitudinal validation studies integrating multiple data sources.