uznlp-uz/uzMED-ABSA

uzMED-ABSA

Dataset Summary

uzMED-ABSA is an Uzbek medical-domain dataset for aspect-based sentiment analysis. The current release contains 7,500 annotated aspect-level rows in a single TSV file.

Each row includes:

• —an Uzbek medical or healthcare-related text sample
• —an aspect term and aspect category
• —a normalized aspect label
• —an opinion expression linked to the aspect
• —a sentiment label and numeric polarity score
• —sentiment intensity and context type
• —negation, sarcasm, and writing-system flags

The dataset covers 35 healthcare aspect categories, including doctor competence, doctor communication, nurse service, reception process, waiting time, diagnosis, treatment, medication, laboratory tests, equipment, hygiene, emergency care, surgery, inpatient care, privacy, safety, telemedicine, access, and overall recommendation.

Supported Tasks

• —aspect category classification
• —aspect term extraction
• —opinion expression extraction
• —sentiment classification
• —polarity scoring
• —negation detection
• —sarcasm / irony detection
• —writing-system analysis for Uzbek text

Files

• —uzmedabsa.tsv: main file in tab-separated format

Dataset Structure

Columns

Tagsets

Sentiment Labels

Polarity Score

Intensity Labels

Script Labels

Aspect Categories

Statistics

Overview

• —Rows: 7,500
• —Columns: 14
• —Split: train only
• —Aspect categories: 35
• —Aspect labels: 35
• —Unique aspect terms: 1,288
• —Unique text samples: 5,725
• —Unique IDs: 5,786
• —IDs with multiple rows: 1,477 (multi-aspect annotations)
• —Extra rows from repeated IDs: 1,714
• —Exact duplicate rows: 0
• —Repeated (text, aspect_term, aspect_label) annotations: 5
• —Text length: 39-307 characters, mean 136.4
• —Text length: 5-39 words, mean 17.1

Sentiment Distribution

Polarity Distribution

Sentiment by Polarity Score

Intensity Distribution

Script Distribution

Negation

Sarcasm

Context Type Distribution

Normalization

• —Column names were converted to English snake_case.
• —Text fields are UTF-8 encoded and Unicode-normalized.
• —Whitespace was stripped and collapsed to single spaces.
• —Uzbek apostrophe letters were normalized to ʻ (U+02BB).
• —Sentiment and aspect labels were canonicalized.
• —Numeric flags and polarity scores are stored as integers.
• —Embedded tabs and newlines were removed from cell values before TSV export.

Usage

from datasets import load_dataset

dataset = load_dataset(
    "csv",
    data_files={"train": "uzmedabsa.tsv"},
    delimiter="\t",
    encoding="utf-8",
)

print(dataset["train"][0])

Or directly from the Hub:

from datasets import load_dataset

dataset = load_dataset("uznlp-uz/uzMED-ABSA")
print(dataset["train"][0])

Citation

If you use uzMED-ABSA in your research, please cite this dataset.

License

This dataset is released under the CC-BY-4.0 license.