Dataset Card for indonesian-twitter-hate-speech-cleaned

Dataset Summary

Cleaned Indonesian Twitter Hate Speech is a curated dataset consisting of Indonesian-language tweets labeled as either hate or neutral. The dataset was collected through a combination of direct scraping from Twitter and aggregation from multiple publicly available GitHub repositories.

The data has been cleaned to remove duplicates, irrelevant content, and non-textual noise, making it suitable for training and evaluating hate speech detection models in Indonesian.

Dataset Details

Dataset Description

This dataset consists of short text entries (tweets) in the Indonesian language, each labeled as either hate or neutral. It is designed for binary text classification tasks, especially for hate speech detection in social media contexts.

Out-of-Scope Use

• Should not be used for actual moderation without human review
• May not generalize to formal or long-text Indonesian corpora
• Not suitable for multi-label or fine-grained hate category classification

Dataset Structure

The dataset consists of two columns:

• text: The tweet text in Indonesian.
• label: Classification label (neutral or hate).

Example Instance

{
  "text": "Pada akhirnya yang asing akan kembali asing.",
  "label": "neutral"
}

Dataset Creation

Curation Rationale

To provide a cleaned and unified Indonesian hate speech dataset for training and benchmarking hate speech classification models in Bahasa Indonesia.

Source Data

Data Collection and Processing

• Data was scraped directly from Twitter using keyword filters and collected from several GitHub repositories related to Indonesian hate speech datasets.
• Cleaning steps included deduplication, removal of links, emojis, and non-textual noise.
• Label harmonization to ensure binary structure: only hate and neutral.

Who are the source data producers?

• Twitter users (public tweets)

Dataset Sources

• Repository: Combination of Twitter scraping and public datasets
• Public Source Included: id-multi-label-hate-speech-and-abusive-language-detection by Okky Ibrohim & A. Budi

Annotations

Annotation process

Some labels were inherited from existing public datasets, and others were manually verified for binary classification consistency.

Who are the annotators?

Primarily the dataset curator and original dataset authors from GitHub sources.

Personal and Sensitive Information

Some tweets may contain names, opinions, or references to groups or individuals. No personally identifiable information (PII) was intentionally collected. Users must comply with Twitter's policy when reusing or publishing based on this data.

Bias, Risks, and Limitations

This dataset may contain:

• Cultural or linguistic biases
• Skewed class distribution
• Offensive content

It reflects the biases of the Twitter community and the original GitHub datasets used.

Recommendations

Users should take caution when deploying models trained on this dataset in production, particularly for moderation, due to risks of false positives/negatives and ethical concerns.

Citation

BibTeX:

@misc{cleanedIndoHate2025,
  title = {Cleaned Indonesian Twitter Hate Speech Dataset},
  author = {Abdillah, Pradana Yahya},
  year = {2025},
  url = {https://huggingface.co/datasets/haipradana/indonesian-twitter-hate-speech-cleaned}
}

APA:

Abdillah, P. Y. (2025). Cleaned Indonesian Twitter Hate Speech Dataset. Hugging Face. https://huggingface.co/datasets/haipradana/indonesian-twitter-hate-speech-cleaned

Glossary

• Hate speech: Offensive or abusive language that attacks a person or group.
• Neutral: Tweets with no offensive or hateful content.

More Information

• If you use this dataset, please consider citing it.
• Feedback and contributions are welcome.

Dataset Card Authors

Pradana Yahya Abdillah

Dataset Card Contact

• Email: yahyaabdillah98@gmail.com