SANAD: Single-Label Arabic News Articles Dataset for Automatic Text Categorization

Published: 2 September 2019| Version 2 | DOI: 10.17632/57zpx667y9.2


SANAD Dataset is a large collection of Arabic news articles that can be used in different Arabic NLP tasks such as Text Classification and Word Embedding. The articles were collected using Python scripts written specifically for three popular news websites: AlKhaleej, AlArabiya and Akhbarona. All datasets have seven categories [Culture, Finance, Medical, Politics, Religion, Sports and Tech], except AlArabiya which doesn’t have [Religion]. SANAD contains a total number of 190k+ articles. How to use it: ___________ 1. Unzip compressed resources. 2. Each folder contains 6-7 sub-folders which are labeled by the category's name. 3. Each sub-folder contains a set of article files corresponding to its category. SANAD_SUBSET is a balanced benchmark dataset (from SANAD) that is used in our research work. It contains the training (90%) and testing (10%) sets. How to use it: ___________ 1. Unzip the compressed file. 2. There are 3 main folders containing the 3 datasets: Akhbarona, Khaleej, and Arabiya. 3. Each dataset-folder contains 2 sub-folders: training and testing. 4. The training and testing folders include the balanced categories sub-folders.



University of Sharjah


Natural Language Processing, Machine Learning, Classification System, Arabic Language, Categorization, Text Processing