Chatbot as a Virtual Assistant to Retrieve Information from Datasheets Using Memory Controllers Domain Knowledge
Haidy Elsayed, Minna Tuallah Hussien, Esraa Mohsen Salah, Anss Khaled Farouk, Abdelrahman Tarek Ahmed, Mona Mahmoud Abd El-Hafez, Khaled Salah · 2023
This paper introduces an end-To-end approach for implementing a chatbot system that helps developers to find the information they need from protocol specifications (pdf documents). The proposed chatbot is a retrieval-based model which uses machine learning algorithms in predicting answers based on the provided training-set. The proposed system is based on the RASA framework and cdQA library using BERT model. We discuss the pre-processing phase for our dataset, the implementation phase, and how we fine-tune the model to have better performance mentioning all the used tools and the methodology our work is based on. The proposed chatbot can get users questions in voice and text forms and return answers in both forms too. The user should provide the protocol name as a keyword, so the chatbot can retrieve the answer, otherwise the chatbot asks the user to provide the protocol name. The response time of the chatbot is very reasonable. The model performance was enhanced after deploying fine–tuning.