CNN-BILSTM Based-Hybrid Automated Model for Arabic Medical Question Categorization

The digital revolution in the medical field has increased the demand for online medical consultations, as well as the growing reliance of patients on these online services, resulting in a large dataset of medical questions. As a result, automated question categorization has become essential. It improves the quality of service by ensuring accurate responses to patient questions. Hence, question categorization is essential for question-answering systems based on natural language processing. To this end, an Automated Arabic Medical Question Categorization System (AMQCS) is developed. AMQCS relies on the concatenation of two forms of feature extraction to address the complexity of Arabic language learning representation, which are Word2Vec and Weight2Vec. Then, the produced numerical representations are used as inputs to the CNN and RNN algorithms. Next, AMQCS concatenates the outputs of these deep learning architectures as one vector. This vector is fed as input to a fully connected layer to classify the question. In the experiment part, a database of 72,000 Arabic medical questions categorized into 18 specialties is used to validate the model. Besides, several metrics are used to evaluate the performance of the proposed model, including accuracy, precision, recall, and F1-score. The proposed model showed exceptional classification performance, scoring above 89% in all metrics used, indicating strong balance across all metrics. Consequently, the findings offer a prospect to leverage AMQCS in online Arabic medical services, which can greatly assist physicians in delivering accurate responses and managers in enhancing the administration.