uznlp-uz/absa-edu

Uzbek Education Aspect-Based Sentiment Analysis

Uzbek Education Aspect-Based Sentiment Analysis (absa-edu) is an aspect-level sentiment dataset containing Uzbek-language opinions about education. Each row provides a text sample, an aspect term and category, an opinion expression, sentiment and polarity labels, intensity, context, negation, irony, and script information.

Dataset Summary

• Dataset ID: uznlp-uz/absa-edu
• Language: Uzbek (uz)
• Domain: Education
• Rows: 7,499
• Columns: 14
• Aspect categories: 30
• Sentiment classes: 4 (NEG, POS, MIX, NEU)
• Splits: train, validation, test
• Format: UTF-8 TSV
• License: CC BY 4.0

Data Files

File	Hugging Face split	Rows	Share	Format
train.tsv	train	5,999	80.00%	TSV
val.tsv	validation	750	10.00%	TSV
test.tsv	test	750	10.00%	TSV
full.tsv	Full dataset	7,499	100.00%	TSV

The three TSV splits have no overlapping IDs. Their union is identical to the set of IDs in full.tsv.

Data Fields

Field	Type	Description
№	integer	Sequential row number in the full dataset.
ID	string	Unique sample identifier, for example EDU-0001.
Matn_namunasi	string	Full Uzbek text sample containing the opinion.
Aspekt_termini	string	Explicit aspect term mentioned in the text.
Aspekt_kategoriyasi	string	Human-readable aspect category in Uzbek.
Aspekt_label	string	Machine-readable aspect category label.
Opinion_ifodasi	string	Text span expressing the opinion about the aspect.
Sentiment	string	Sentiment class: negative, positive, mixed, or neutral.
Polarity_score	integer	Ordinal polarity score from -2 to 2.
Intensivlik	string	Sentiment intensity: past, o‘rta, yuqori, or kuchli.
Kontekst_turi	string	Discourse/context type, such as praise, complaint, mixed opinion, or neutral report.
Inkor	integer	Negation flag: 1 if negation is present, otherwise 0.
Kinoya	integer	Irony flag: 1 if irony is present, otherwise 0.
Yozuv	string	Script label: Latin, Cyrillic, or Mixed.

Label Values

Sentiment

Label	Meaning
NEG	Negative sentiment
POS	Positive sentiment
MIX	Mixed sentiment
NEU	Neutral sentiment

Polarity Score

Score	Interpretation
-2	Strong negative polarity
-1	Negative polarity
0	Neutral or mixed polarity
1	Positive polarity
2	Strong positive polarity

Statistics

Overview

Metric	Value
Total rows	7,499
Columns	14
Unique IDs	7,499
Unique text samples	5,729
Unique aspect terms	696
Unique opinion expressions	4,661
Aspect categories	30
Duplicate complete rows	0
Empty field values	0
Minimum text length	5 words
Maximum text length	52 words
Mean text length	19.07 words
Median text length	17 words

Sentiment Distribution

Sentiment	Full	Train	Validation	Test	Full share
NEG	2,577	2,056	259	262	34.36%
POS	2,516	2,012	252	252	33.55%
MIX	1,235	990	123	122	16.47%
NEU	1,171	941	116	114	15.62%
Total	7,499	5,999	750	750	100.00%

Aspect Category Distribution

Aspect label	Uzbek category	Count
ACADEMIC_ENVIRONMENT	Akademik muhit	240
ADMIN_SERVICE	Ma’muriy xizmatlar	255
ASSESSMENT_FAIRNESS	Baholash adolatliligi	255
ASSIGNMENT_WORKLOAD	Topshiriqlar va yuklama	255
ATTENDANCE_DISCIPLINE	Davomat va intizom	255
CLASSROOM_INFRASTRUCTURE	Auditoriya va jihozlar	255
COURSE_CONTENT	Fan/kurs mazmuni	255
CURRICULUM	O‘quv reja va dastur	254
DIGITAL_CONTENT	Raqamli kontent sifati	240
DORMITORY_CAMPUS	Yotoqxona va kampus xizmatlari	240
EXAM_PROCESS	Imtihon va nazorat jarayoni	255
FEEDBACK	Fikr-mulohaza va qayta aloqa	255
INCLUSION_EQUALITY	Inkluzivlik va adolat	240
INTERNSHIP_CAREER	Amaliyot va ishga joylashish	240
LANGUAGE_COMMUNICATION	Til va kommunikatsiya sifati	240
LEARNING_OUTCOME	Ta’lim natijasi	240
LEARNING_RESOURCES	Kutubxona va o‘quv resurslari	255
LMS_PLATFORM	Onlayn ta’lim platformasi	255
OVERALL_SATISFACTION	Umumiy qoniqish	240
PAYMENT_CONTRACT	Kontrakt va to‘lovlar	255
PRACTICAL_TRAINING	Amaliy mashg‘ulotlar	255
SAFETY_SECURITY	Xavfsizlik va tartib	240
SCHEDULE	Dars jadvali	255
SCHOLARSHIP_SUPPORT	Stipendiya va moliyaviy qo‘llab-quvvatlash	255
STUDENT_SUPPORT	Talabalarni qo‘llab-quvvatlash	240
TEACHER_ATTITUDE	O‘qituvchining muomalasi	255
TEACHER_COMPETENCE	O‘qituvchi malakasi	255
TEACHING_CLARITY	Darsni tushuntirish sifati	255
TEACHING_METHOD	O‘qitish metodikasi	255
TECHNICAL_SUPPORT	Texnik xizmat va internet	255

Polarity Score Distribution

Score	Count	Share
-2	923	12.31%
-1	1,654	22.06%
0	2,406	32.08%
1	1,915	25.54%
2	601	8.01%

Intensity Distribution

Intensity	Count	Share
past	1,637	21.83%
o‘rta	4,299	57.33%
yuqori	881	11.75%
kuchli	682	9.09%

Negation and Irony

Feature	Absent (0)	Present (1)	Present share
Negation (Inkor)	6,354	1,145	15.27%
Irony (Kinoya)	7,235	264	3.52%

Script Distribution

Script label	Count	Share
Latin	6,455	86.08%
Cyrillic	697	9.29%
Mixed	347	4.63%

Loading

Load the predefined splits from the Hugging Face Hub:

from datasets import load_dataset

dataset = load_dataset("uznlp-uz/absa-edu")
print(dataset)
print(dataset["train"][0])

Load an individual split:

from datasets import load_dataset

train = load_dataset("uznlp-uz/absa-edu", split="train")
validation = load_dataset("uznlp-uz/absa-edu", split="validation")
test = load_dataset("uznlp-uz/absa-edu", split="test")

Load the full TSV file directly:

from datasets import load_dataset

full_dataset = load_dataset(
    "csv",
    data_files="full.tsv",
    delimiter="\t",
    split="train",
)

With pandas:

import pandas as pd

train = pd.read_csv("train.tsv", sep="\t")
validation = pd.read_csv("val.tsv", sep="\t")
test = pd.read_csv("test.tsv", sep="\t")
full_dataset = pd.read_csv("full.tsv", sep="\t")

Intended Use

The dataset can be used for Uzbek aspect-based sentiment analysis research and development, including:

• sentiment classification;
• aspect category classification;
• aspect term and opinion expression analysis;
• polarity and intensity prediction;
• negation and irony analysis;
• multilingual and multi-script NLP experiments;
• education-domain NLP benchmarking.

Notes and Limitations

• The dataset is specific to the education domain; results may not generalize to other domains.
• A text may occur more than once when it is associated with different aspect-level annotations.
• The Polarity_score value 0 is used by both neutral and mixed examples; use Sentiment when these classes must be distinguished.
• Script labels are normalized to Latin, Cyrillic, and Mixed.
• Users should independently evaluate model behavior for dialectal, orthographic, and code-mixed Uzbek text before deployment.

License

This dataset is released under the Creative Commons Attribution 4.0 International license.

Citation

If you use this dataset, cite the dataset repository:

@misc{uzbek_absa_edu_2026,
  title        = {Uzbek Education Aspect-Based Sentiment Analysis},
  author       = {{UzNLP}},
  year         = {2026},
  howpublished = {\url{https://huggingface.co/datasets/uznlp-uz/absa-edu}},
  license      = {CC BY 4.0}
}