Machine reading comprehension to answer COVID19 queries using Bio-Bert and multi-task learning

Abstract

In recent years, with the development of artificial intelligence technology in many fields, the question-answering system has brought significant change to knowledge acquisition mechanism. The question-answering system based on machine reading comprehension can obtain short and accurate answers compared with the traditional retrieval question-answering system. This thesis designs an intelligent question-answering method based on information retrieval and multi-document machine reading comprehension. Firstly, a two-stage information retrieval recall strategy is designed. After the information retrieval by using Bm 25, a structure called Dual Bio-Bert Retrieval is designed, which uses two Bio-Bert to extract semantic features in questions and paragraphs respectively in the training stage. Then, the information between the two Bio-Bert is interacted by utilizing Performer. At the reading comprehension stage, an additional method—Matching Tech is designed to improve the model. Compared with other common methods, the results of the information retrieval and reading comprehension shows that the models designed in this research have good performance.