Building Online Social Network Dataset for Arabic Text Classification

Abstract

Social networking sites have spread widely in recent years, and through them, a large amount of data is shared in all its forms: text, photo, voice, and video. It also allows communication with users through different forms such as chat, comments, and Posts, and the most exchanged content is in the form of text data. These results a large volume of data displayed to each user. This encouraged and attracted the attention of researchers to make an effort to analyze and work on this large amount of data available for free on the online social networks, most efforts focus on Twitter and English data. Building dataset is the most time-consuming and the most important part of the text classification process. Despite the increase in the number of Arabic users and the increase in Arabic content on online social Networks (OSN), there is a scarcity in Arabic datasets collected from social networks for text classification purpose. So In this paper, Arabic social dataset was built to be used in text classification purpose. our dataset was gathered from Facebook, it consists of 25,000 posts were collected from different Facebook pages and were classified into ten categories, politics, economics, sport, religion, technology, TV, ads, foods, health, and porno. The dataset was assessed to ten Arabic local speakers and Facebook users to evaluate the validity of the dataset made. We used a RapidMiner tool to evaluate and compute the performance of our dataset. We obtained a classification accuracy of 95.12%.