BEN-FND: A 16K Multilingual (Bangla–English) Fake News Detection Dataset

BEN-FND is a curated multilingual fake news detection dataset containing 16,349 Bangla and English news samples, each labeled as "Fake" or "Real." The dataset was created by combining data from multiple publicly accessible online sources, including news portals, fact-checking websites, and publicly released fake-news datasets. Only selected portions from each source were used.

All samples were extensively cleaned, deduplicated, normalized, and restructured to form a unified dataset suitable for machine learning and natural language processing research. The content was collected over a long period, and partial samples were taken from each source, resulting in a diverse representation of news topics and misinformation patterns.

Each record includes the following fields:
id – unique numeric identifier
title – headline or title of the news item
news_details – short article text or detailed summary
language – Bangla or English
category – news domain (e.g., Politics, Business, National, Sports, etc.)
label – Fake or Real

The dataset is designed for research in fake news detection, misinformation classification, multilingual NLP, and cross-lingual analysis. It can be used for training and benchmarking machine learning models, developing text classifiers, and exploring linguistic patterns in Bangla and English fake news.

Provenance Note:
This dataset is a curated compilation created from multiple publicly accessible sources. Due to incremental collection and partial sampling, not all original URLs or dataset identifiers are available. The structure, cleaning, preprocessing, and compilation are original contributions by the author.

License:
All curation and preprocessing work is released under CC0 1.0 Public Domain Dedication.
Original text excerpts remain the copyright of their respective publishers and are provided strictly for research and educational purposes.