NADiA: News Articles Dataset in Arabic for Multi-Label Text Categorization
NADiA Dataset is the largest, to the best of our knowledge, source for Arabic textual data that can be used in any NLP related task such as text classification. We chose the abbreviation NADiA as it is a common Arabic name. The data was collected by scraping ‘SkyNewsArabia’ and ‘Masrawy’ news websites using Python scripts that are fine-tuned for each website. SkyNewsArabia will be referred to as NADiA1, while the latter would be NADiA2. NADiA1 is a big dataset containing 37,445 files, while NADiA2 is a huge dataset that contains 678,563 files. However, after filtering and cleaning we reduced the numbers to 35,416 and 451,230 for NADiA 1 and 2, respectively. NADiA1 consists of the following categories (24, displayed in English for easy referencing): News, North Africa, Levant, Middle East, The Americas, Research, Finance & Economy, War & Terrorism, Gulf, Europe, Political Figures, Iran, Technology, Russia, Sports, Tennis, Football, English League, Arabian Sports, Spanish League, Health, East Asia, Environment, Other Countries NADiA2 consists of the following categories (28, displayed in English for easy referencing): Politics, Middle East, Asia, Africa, United States, Europe, Other Countries, Leaders, Sports, Arabian Sports, Football Clubs, Spanish League, Egyptian League, Finance, Arts, Cinema & TV, Fashion, Health, Pregnancy & Delivery, Cancer, Obesity, Social Media, Technology, Religion, Islamic, Fatawa, Worship, Prophet Biography
Steps to reproduce
Notes to reproduce the tags: 1- The main tags (24/28) are introduced while reading the dataset. 2- The code for producing these tags is provided and fine-tuned per dataset in its zipped file. 3- In each article, the file starts with a line containing multiple tags in Arabic & English. The first tag (regardless of language) is obtained from the source portal (URL Address), while the rest of the tags are extracted from the article's keywords.