Asosiy kontentga oʻtish
Boshqa

UzMED-ABSA: An Uzbek Medical Domain Dataset for Aspect-Based Sentiment Analysis

Elov Botir BoltayevichAlisher Navo'i Tashkent State University of Uzbek Language and LiteratureAlaev RuhilloNational University of Uzbekistan
2026
ABI

Annotatsiya

UzMED-ABSA is an Uzbek medical-domain dataset for aspect-based sentiment analysis. The dataset contains 7,500 annotated aspect-level rows in TSV format and is designed for research on Uzbek healthcare text analysis, aspect category classification, sentiment classification, polarity scoring, negation detection, and sarcasm or irony detection. Each record includes an Uzbek medical or healthcare-related text sample, aspect term, human-readable aspect category, normalized aspect label, opinion expression, sentiment label, polarity score, intensity label, context type, negation flag, sarcasm flag, script label, and source row number. The dataset covers 35 healthcare aspect categories, including doctor competence, doctor communication, doctor attitude, nurse service, reception process, waiting time, diagnosis accuracy, treatment effectiveness, medication recommendation, laboratory tests, equipment quality, clinic environment, hygiene, emergency care, pain management, surgery quality, inpatient care, billing transparency, insurance benefits, privacy and ethics, patient safety, telemedicine, access, and overall recommendation. The dataset includes 5,725 unique text samples, 5,786 unique sample IDs, and 1,288 unique aspect terms. Sentiment labels are distributed as follows: POS 3,052, NEG 2,940, NEU 1,003, and MIX 505. Polarity scores range from 1 to 5, where 1 indicates strong negative sentiment, 3 indicates neutral or mixed sentiment, and 5 indicates strong positive sentiment. The writing-system distribution is 6,040 Latin-script rows, 801 Cyrillic-script rows, and 659 mixed-script rows. The dataset also includes 824 rows with negation and 202 rows with sarcasm or irony. The TSV file is encoded in UTF-8. During preparation, column names were converted to English snake_case, text fields were Unicode-normalized, whitespace was stripped and collapsed, Uzbek apostrophe variants were normalized to U+02BB, categorical labels were canonicalized, and numeric fields were validated as integers.

Hali tarjima qilinmagan

Identifikatorlar

Iqtiboslar va manbalar

0 ta iqtibos0 ta foydalanilgan manba