Blood Disease Prediction from Arabic Medical Dataset Using Arabert through Name Entity Recognition رسالة ماجستير

Abstract

Natural Language Processing (NLP) tasks have been showing state-of-the-art results since the appearance of Bidirectional Encoder Representations from Transformers (BERT). However, literature review shows that BERT performs very effectively at modeling domain-specific datasets when Fine-Tuned and modified. This was not the case before BERT since specific-domain datasets were very challenging to NLP task due to the type of these datasets where they are usually rich with scientific and technical jargon. This thesis will use AraBERT, a special variant of BERT that is trained to understand Arabic language. AraBERT will be used in a different context by altering and modifying the model to help us achieve a successful prediction from a specific-domain Arabic dataset and answer our research questions. The result of this work shows big improvements in prediction accuracy for specific-domain Arabic medical dataset. The final results demonstrate the model's capability and robustness to predict blood disease symptoms