Deep Learning approach for Arabic Healthcare: MedicalBot

Abstract Intelligent healthcare has made significant progress using artificial intelligence (AI) methods. Since the COVID-19 pandemic, healthcare service has gained more attention, particularly remote and automated healthcare consultation. Medical bots are becoming increasingly popular for obtaining medical advice and support. Medical bots offer numerous benefits, making them an attractive option for patients and healthcare providers. They provide 24/7 access to medical advice, reduce appointment wait times by providing quick answers to common questions or concerns, and cost savings associated with fewer visits or tests required for diagnosis and treatment plans. The outcome of the medical bot depends upon the learning quality. Thereby depends on the appropriate corpus within the domain of interest. Arabic is one of the most commonly used languages for sharing users' internet content. Developing a medical bot on Arabic faces many challenges: including the morphological composition of the language, the diversity of dialects, and the need for appropriate corpora in the domain. Thus, this paper's main contribution is introducing the largest Arabic Healthcare Q&A dataset, called (MAQA). The dataset consists of more than 430k questions distributed into 20 medical specializations. Also, the paper adopts three deep learning models, LSTM, Bi-LSTM, and transformers, for experimenting and benchmarking the proposed corpus MAQA. The experimental results indicate that the recent Transformer model, with an average cosine similarity of 80.81% and BLeU score of 58%, outperforms the traditional deep learning models.