Validity of Q&A metrics through NLP measurement patterns

Nowadays the Artificial Intelligent (AI) systems as chatbots are becoming more popular in both business and private life. The main idea of the chat-bots is to recognize the text or the speech of the user, extract the needed information from the database and give back to the user the information he was looking for. This type of automatic question answering (QA) system can be implemented in the education field. The goal of this work is to prepare the Natural Language Processing (NLP)approach to develop in the future the chat-bot for the student's need. Specifically to understand how the already pretrained BERT large model finetuned on SQuAD can be efficient on the different Conversational Question Answering (COQA) dataset. This thesis can be considered as successful experimental work with a well-performed result of 84% of Accuracy.