Biomedical Question Answering: A Survey of Methods and Datasets

摘要

Thousands of biomedical research papers are published each day. Now, it takes more time than ever for researchers and healthcare information professionals to find relevant information from this huge literature. Classical information retrieval (IR) systems such as PubMed, while helpful, still provide a large number of search results that needs to be examined manually. Biomedical Question answering (QA) systems on the other hand can extract and provide one direct and exact answer to a question. While huge progress has been made recently in general-domain QA. Biomedical QA still has a long way to go. The number one reason for the slow progress of biomedical QA compared to general-domain QA, is the limited number of biomedical QA datasets and the even smaller number of annotated training instances. The goal of this survey is to list and compare the available biomedical QA datasets, and to provide an overview of traditional and end-to-end neural-based biomedical QA systems.