---
dataset_name: Nepali Bias and Information Dataset
labels: hate_speech, Misinformation, Bias, Normal
license: apache-2.0
task_categories:
- text-classification
language:
- ne
size_categories:
- 1K<n<10K
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
dataset_info:
  features:
  - name: text
    dtype: string
  - name: label
    dtype: string
  splits:
  - name: train
    num_bytes: 9715866.0
    num_examples: 9408
  download_size: 3741711
  dataset_size: 9715866.0
tags:
- hate_speech
- misinformation
- bias
---

# 🧠 NaBI: Nepali Bias & Information Dataset

The **NaBI (Nepali Bias & Information)** Dataset is a curated and annotated dataset developed for the classification of Nepali language content into four critical categories related to information integrity and harmful content. It is designed to aid in the development of NLP models for moderation, content filtering, and sociopolitical analysis.

## 💡 Dataset Description

The dataset consists of Nepali text samples sourced from public domains like **user comments, Facebook posts, and editorial pieces**, annotated into four distinct categories:

- **Bias**: Text exhibiting partial, one-sided, or prejudiced viewpoints. This includes:
  - **Editorial Bias**: Political slant, caste or gender favoritism.
  - **User Comment Bias**: Implicit bias or favoritism in social interactions or commentary.
  
- **Normal**: Neutral, factual, and civil content. No signs of misinformation, bias, or harmful intent.

- **Misinformation**: False or misleading content. Includes manipulated facts, deceptive headlines, and rumors.

- **Hate Speech**: Language that promotes discrimination, hostility, or violence towards individuals or groups based on caste, ethnicity, gender, political affiliation, etc.

## 📂 Use Cases

This dataset is ideal for:
- Building classifiers to detect and filter hate speech or misinformation in Nepali content.
- Researching media bias and its influence on public opinion.
- Studying sociolinguistic patterns and harmful discourse in Nepali digital spaces.

## 📊 Dataset Structure

Each entry in the dataset includes:
- **Text**: A Nepali language sentence or paragraph.
- **Label**: One of the four classes — `Bias`, `Normal`, `Misinformation`, `Hate Speech`.

## 📜 License

This dataset is licensed for **research and educational purposes only**. Ensure proper citation and ethical use in any public project or publication.

## 📧 Contact

For inquiries, contributions, or collaborations, please contact:  
**Utkarsha Khanal**  
📬 utkarsha.khanal@gmail.com