A New Fusion Method on Machine Reading Comprehension
Wenfang Tang, Hong Jiang, Kejia Xu · 2020
Machine reading comprehension, as one of the core tasks of natural language processing, has been a research focus in this field. This paper is mainly aimed at Dureader, a Chinese-oriented multi-document reading comprehension task dataset, and an end-to-end machine reading comprehension model based on neural network is proposed. In this paper, a combination of context-to-question attention and self-attention, and a new fusion method are used to fuse context and attention information. The comparison of experimental results shows that compared with the baseline model, by our methods, BLEU-4 index has been improved by 6% and ROUGE-L index has been improved by 3%.