Thousands of medical questions each day are asked through online platforms. With a similar description of the same symptoms and questions from many patients, the process of answering the questions becomes repetitive and puts an intensive load on doctors. In this study, a Question-Answering (QA) system is created to automatically select accurate answers for new questions by leveraging previously answered questions. This system aims to assist doctors in making accurate diagnoses with reduced effort and quicker response times. Neural network models and natural language processing techniques are implemented and applied to Arabic medical QA dataset from an online medical consultation platform. The primary challenges in processing Arabic stem from the complexities inherent in the language itself and the scarcity of resources such as studies and datasets. Additionally, the dataset consists of various dialects, further complicating the matter. The proposed model consists of Bidirectional Gated Recurrent Unit (BiGRU), Stack Convolutional Neural Network (Stack-CNN), Dense Neural Network (DNN), and Cosine similarity metric. The model is trained based on pre-trained word embeddings for Arabic language. A comparison of multiple models, such as Long Short-Term Memory (LSTM), BiGRU, CNN, Dense Neural Network (DNN), and Ensemble Modeling, is conducted to compare the accuracy of the generated answer. The results showed that the proposed model outperformed the other models and got the best accuracy of 76.35%.
Siyam et al. (Thu,) studied this question.