UESTION-ANSWERING MACHINE LEARNING MODEL FOR COVID-19

Covid-19 pandemic outbreak has affected everyone of our lives directly or indirectly. There is a lot of information and an equally good amount of misinformation filled out there on the internet. The only way we can mitigate the Covid-19 crisis and keep ourselves in a healthy condition is to be informed with the right information. Whatsapp forwards, YouTube clickbait videos have made the situation worse by propagating misinformation. Governments and NGOs are trying to give out the right information but these misinformation carriers have got more reach and popularity. In this thesis, I tried to come up with a machine learning model that would act as a question answering engine for Covid-19 queries. The Question answer engine is built on the Covid-19 Open Research Dataset (CORD-19 dataset in short), the largest open dataset corpus of Covid- 19 related research papers which is developed by Allen Institute for AI as one of the training datasets. Bidirectional Encoder Representations from Transformers (BERT), Global Vectors for Word Representation (GloVe) and Generative Pre-trained Transformer 3 (GPT-3) models were used to create required embeddings. The model combines Covid-19 related research texts’ mining and leverages the famed GPT-3 models capabilities with Natural Language Processing(NLP) tools to give a better set of answers than the previously known information retrieval Question Answering Systems. Evaluation of the model is done through comparison with a manually created dataset with accurate information based on reliable sources like WHO. It is hoped that the Machine Learning Model created through this research study would be able to help researchers and common people alike in their search for accurate knowledge and information..