A Deep Learning Approach for Arabic Text Classification

Advancement in information technology produced massive textual material that is available online. Text classification algorithms are at the core of many natural language processing (NLP) applications. There are several algorithms which have been implemented to tackle the classification problem for English and other European languages. Few attempts have been carried out to solve the problem of Arabic text classification. In this paper, we demonstrate a feed-forward deep learning (DL) neural network for the Arabic text classification problem. The first layer uses term frequency-inverse document frequency (TF-IDF) vectors constructed from the most frequent words of the document collection. The output of the first layer is used as an input to the second layer. To reduce the classification error rate, we used Adam's optimizer. We conducted a set of experiments on two multi-classes Arabic datasets to evaluate our approach based on standard measures such as precision, recall, F-measure, support, accuracy and time to build the model. We compared our approach with the logistic regression (LR) algorithm. Experiments entailed that the deep learning approach outperformed the logistic regression algorithm for Arabic text classification.