Pashto Sentiment Analysis Dataset

Pashto Sentiment Analysis Dataset

Overview

The Pashto Sentiment Dataset is a large-scale manually reviewed corpus developed for sentiment analysis and text classification in the Pashto language.

The dataset contains 10,229 Pashto sentences annotated with one of three sentiment labels:

• Positive
• Negative
• Neutral

Pashto language is a low-resource language with limited publicly available NLP resources. This dataset aims to support research, education, and industrial applications in sentiment analysis, natural language understanding, and machine learning for Pashto language.

---

Why This Dataset?

Despite being spoken by tens of millions of people, Pashto language remains underrepresented in modern NLP research. The lack of high-quality annotated datasets presents a significant challenge for building robust language technologies.

This dataset was created to help researchers and developers:

• Build sentiment classification systems
• Fine-tune transformer models
• Benchmark NLP algorithms
• Develop Pashto language technologies
• Support low-resource language research

---

Key Features

✅ 10,229 manually reviewed samples

✅ Three sentiment categories

✅ Balanced class distribution

✅ Real-world Pashto language content

✅ Suitable for machine learning and deep learning

✅ Ready-to-use CSV format

✅ Designed for academic and industrial research

---

Dataset Statistics

---

Class Distribution

The dataset provides a relatively balanced distribution across sentiment classes, making it suitable for supervised learning experiments.

Class Distribution

print(df["sentiment"].value_counts())

---

Dataset Schema

---

Example Records

---

Data Collection

The dataset was compiled from publicly available Pashto-language content covering multiple domains, including:

• Politics
• International Affairs
• Economy
• Sports
• Society
• Technology
• Culture
• Health

The diverse coverage helps improve the dataset's usefulness across different NLP applications.

---

Annotation Process

The annotation workflow consisted of:

Step 1 — Data Collection

Pashto text samples were gathered from publicly available sources(BBC Pashto, VOA Pashto, Shamshad News ....)

Step 2 — Initial Labeling

Each sentence was assigned a sentiment category based on its overall semantic meaning.

Step 3 — Manual Review

All labels were reviewed and validated by the dataset creator.

Step 4 — Quality Assurance

The final dataset underwent cleaning and validation procedures before release.

---

Data Quality

Several quality-control measures were applied:

• Duplicate removal
• Missing-value removal
• Text normalization
• Formatting consistency checks
• Label verification
• Manual review

These steps were performed to improve dataset reliability and usability.

---

Potential Applications

Natural Language Processing

• Sentiment Analysis
• Opinion Mining
• Text Classification
• Language Understanding
• Content Analysis

Machine Learning

• Multi-Class Classification
• Deep Learning
• Transformer Fine-Tuning
• Transfer Learning

Research

• Low-Resource NLP
• Cross-Lingual Learning
• Pashto Language Technology
• Benchmark Evaluation

---

Getting Started

Load the Dataset

import pandas as pd

df = pd.read_csv("pashto_sentiment_dataset.csv")

print(df.head())

Class Distribution

print(df["sentiment"].value_counts())

---

Recommended Models

Researchers may use this dataset with:

• BERT
• RoBERTa
• XLM-RoBERTa
• mBERT
• DistilBERT
• LLaMA-based classifiers
• Traditional machine learning models

---

Dataset Characteristics

• The dataset is written in Standard Pashto.
• The corpus is derived from professionally written Pashto-language content.
• The dataset covers a diverse range of topics, including politics, economics, sports, technology, health, society, culture, and international affairs.
• The language follows commonly accepted Pashto writing conventions and editorial standards.
• The dataset is suitable for sentiment analysis, text classification, transfer learning, and Pashto NLP research.
• The corpus provides broad linguistic coverage across multiple news domains.

---

Intended Use

This dataset is intended for:

• Research
• Education
• Benchmarking
• Model Development
• Academic Projects
• Industrial NLP Applications

---

Citation

If you use this dataset in your research, publications, or projects, please cite:

@dataset{khail2026pashto,
  author       = {Khairullah Ibrahim Khail},
  title        = {Pashto Sentiment Analysis Dataset},
  year         = {2026},
  publisher    = {Kaggle},
  url          = {https://www.kaggle.com/datasets/khairullahsahil/pashto-sentiment-dataset},
  version      = {1.0}
}

Khairullah Ibrahim Khail

Pashto Sentiment Analysis Dataset

Version 1.0. Kaggle, 2026.

---

Author

Khairullah Ibrahim Khail

---

Version History

Version 1.0

• Initial public release
• 10,229 manually reviewed Pashto sentences
• Three sentiment classes
• CSV format

---

Acknowledgments

This dataset was created to support the growth of Pashto Natural Language Processing and encourage further research on low-resource languages.

Contributions, feedback, and research collaborations are welcome.Khairullah Ibrahim Khail