Improving Healthcare Question Answering System by Identifying Suitable Answers
Baivab Das, S. Jaya Nirmala · 2022 IEEE 2nd Mysore Sub Section International Conference (MysuruCon) · 2022
Healthcare question answering is an important task in the field of Natural Language Processing. As the era of the internet has already begun, textual data related to the healthcare field has also increased rapidly. There is a huge opportunity to utilize the full potential of deep learning-based methods to achieve better accuracy in healthcare question answering. Unfortunately, the open-domain pre-trained models do not perform well in healthcare question answering. As the terminologies used in the healthcare field are unique, the concept of understanding them is hard. To resolve this issue, in this project, the BioBERT model is used to answer questions related to the healthcare field. BioBERT is based on BERT (Bidirectional Encoder Representations from Transformers) and it is pre-trained on large biomedical corpora which gives it the ability to understand healthcare-related terminologies. As the BERT model has already proved its efficiency as a bidirectional language model, the BioBERT model here is trained with the SQUAD question-answering dataset. The model is able to answer questions related to the healthcare field with an F1 score of 88.