Intelligent Question Answering System Based on Machine Reading Comprehension

Abstract Question answering(Q&A) system is important for accelerating the landing of artificial intelligence. This paper makes an improvement on Q&A which uses method retrieval-machine reading comprehension (MRC). In retrieval phase, we use BM25 to recall some documents and split these into paragraphs, then reorder paragraphs according correlation with question, so as reduce number recalled improve speed MRC. MRC stage, design a multi-task structure, can judge whether paragraph contains answer locate accurately. Besides, modify loss function fit sparse labels during training. The experiments are carried out multiple data sets verify effectiveness improved system.