Resumen
Regardless of the clear growth of Arabic texts on social networking sites (SNSs), it is still difficult to understand or summarize users' opinions or perspectives on a specific topic. Accordingly, Arabic text classification is one of the most challenging topics. This is because of several issues related to the nature of the Arabic language and words that have different variation in meaning. In this paper, after tokenizing the Arabic words, we investigate the role of several pre-processing techniques before classifying Arabic text into different categories. Arabic words were converted into vectors using the term frequency-inverse document frequency (TF-IDF) technique. The findings show that applying Linear Support Vector Machine (LSVC) with stop words and without stemming techniques can outperform the application of Decision Tree (DT) and Random Forest (RF) methods. It was found that the effectiveness of the proposed LSVC is 99.37%. These outcomes are significant to identify users' opinions on SNSs and can have many implications on political, social, economic, and business sectors.