Arabic Multi-Topic Labelling using Bidirectional Long Short-Term Memory

Abstract

The number of text documents on the internet is rapidly increasing. As a result, there is a growing demand for methods that can automatically organize and identify electronic documents (instances). The multi-label classification task has been used in a variety of applications and is commonly used in real-world problems. It simultaneously assigns multiple labels to each text. In the Arabic language, there have been few and inadequate research studies on the multi-label text classification issue. In this paper, I proposed a deep learning model using bidirectional long short-term memory (BiLSTM) for multi-class topic classification using Mowjaz Multi-Topic Labelling Task dataset. The BiLSTM model consists of 4 layers only which can be considered as light weight model, these layers are input layer, bidirectional LSTM layer, and two dense layers. The results show that the model successfully to classify topics with F1-Socre of 0.8089 on the testing dataset.