Hugging Face's logo

Datasets:
lducc
/
fpt-comments-sentiment

Tasks:
Text Classification
Modalities:
Tabular
Text
Formats:
csv
Sub-tasks:
sentiment-classification
topic-classification
Languages:
Vietnamese
Size:
1K - 10K
ArXiv:
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
fpt-comments-sentiment / README.md
lducc's picture
lducc
update dataset
00a9491 verified
preview code
|
Raw
History Blame Contribute Delete
3.7 kB
metadata 
language:
  - vi
task_categories:
  - text-classification
task_ids:
  - sentiment-classification
  - topic-classification
pretty_name: fpt-comments-sentiment
license: other
size_categories:
  - 1K<n<10K
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train.csv
      - split: test
        path: data/test.csv

fpt-comments-sentiment

This is a hand-curated Vietnamese feedback dataset built from FPT-related discussions across Facebook communities, FuOverflow, Reddit/VOZ-style forums, FPT web pages, and other student-community sources. (Inspired by the NEU-ESC dataset)

The data is curated by manually collected examples with Facebook group comments scraped using Playwright, then cleaned, redacted, hand-reviewed, and labeled with support from LLM-assisted review.

Each row contains a redacted comment with two labels: one for sentiment and one for topic. The dataset is intended for Vietnamese sentiment analysis, toxicity detection, and community-topic classification in FPT University contexts.

This dataset is created solely for use within the FPT coursework.

Sources

The data was manually curated from FPT-related public/community sources, including Facebook pages/groups, Reddit, VOZ, FuOverflow, and other FPT-related websites.

Source links include:

Raw per-comment URLs, author/profile URLs, post URLs, permalinks, and scrape IDs are not included in the released CSV files.

Data Files

Default splits:

  • data/train.csv
  • data/test.csv

Convenience full export:

  • data/all.csv

Each CSV has:

  • Text
  • Sentiment
  • Classification

Labels

Sentiment:

  • 0: Neutral
  • 1: Positive
  • 2: Negative
  • 3: Toxic

Classification:

  • 0: Spam
  • 1: News
  • 2: Academic
  • 3: Other
  • 4: Service
  • 5: Jobs & Recruitment
  • 6: Personal Affairs
  • 7: Social Affairs
  • 8: Help & Share
  • 9: Club & Events

Splits

Split Rows
Train 4,377
Test 1,092
Total 5,469

Label Counts

Sentiment:

Label Rows
Neutral 2,974
Positive 1,380
Negative 1,067
Toxic 48

Classification:

Label Rows
Spam 70
News 16
Academic 1,912
Other 2,057
Service 491
Jobs & Recruitment 165
Personal Affairs 121
Social Affairs 107
Help & Share 442
Club & Events 88

Notes

The text was redacted before export. The CSV files do not include raw author names, profile URLs, post URLs, permalinks, response URLs, or scrape IDs.

Angle-bracket placeholders are used for redaction; starred masks were normalized to context placeholders such as <NAME>.

Toxic labels are intended for targeted abuse, insults, slurs, hostile commands, or similar direct attacks. Profanity alone was not treated as toxic.

Citation

If you use the label schema/protocol inspired by NEU-ESC, please cite:

@misc{mai2025neuesc,
  title = {NEU-ESC: A Comprehensive Vietnamese dataset for Educational Sentiment analysis and topic Classification toward multitask learning},
  author = {Mai, Phan Quoc Hung and Nguyen, Quang Hung and Duong, Phuong Giang and Nguyen, Hong Hanh and Long, Nguyen Tuan},
  year = {2025},
  eprint = {2506.23524},
  archivePrefix = {arXiv},
  primaryClass = {cs.CL},
  doi = {10.48550/arXiv.2506.23524},
  url = {https://arxiv.org/abs/2506.23524}
}