VHCor: Vietnamese Healthcare Corpus-A Comprehensive Dataset for Vietnamese Medical Department Recognition

Considering the extended response times experienced on hospital online forums, there is an urgent requirement for a dedicated system that aids patients in efficiently identifying the appropriate medical departments through categorization support. Several studies have explored this area within the Vietnamese context, focusing on the challenge of comprehending patients' symptoms and providing responses to the relevant medical departments. In this study, we achieved two primary contributions. Firstly, due to the lack of benchmark datasets, we constructed the Vietnamese Healthcare Corpus (VHCor), a novel dataset designed for medical department recognition tasks in low-resource languages like Vietnamese. This corpus comprises 26, 390 questions provided by patients across 28 diverse medical specialties sourced from public Vietnamese hospital forums. Secondly, we conducted experiments on baseline methods to preliminarily evaluate our VHCor corpus. As a result, utilizing PhoBERT with K-Fold achieved an impressive accuracy rate of 89% which surpasses logistic regression algorithms. These findings highlight the potential for advancing specialty recognition in the Vietnamese medical domain and accentuating the effectiveness of PhoBERT in handling medical tasks.