Intelligent Arabic-Based Healthcare Assistant

Text classification has been one of the most common natural language processing (NLP) objectives in recent years. Compared to other languages, this mission with Arabic is relatively restricted and in its early stages, and this combination in the medical application area is rare. This paper builds an Arabic health care assistant, specifically a pediatrician that supports Arabic dialects, especially Egyptian accents. The proposed application is a chatbot based on Artificial Intelligence (AI) models after experimenting with Two Bidirectional Encoder Representations from Transformers (BERT) models, a pre-trained BERT and Logistic regression TF-IDF and Doc2vec. These models were applied to the Arabic dataset with different dialects from different couturiers such as Egypt, Saudi Arabia, and Iraq. The proposed system consists of 4 stages: scrapping and collecting data, then wrangling it, data preprocessing, data extraction, trained models with new data, and connect the model to the database that contains the answers. Experimental tests showed that the BERT model outperformed the others by getting a 95% Accuracy. Logistic regression with Doc2vec was the second best with 71% F-measure and 73% Accuracy.