| 摘 要: 编写机器阅读理解软件中，一个基本步骤就是对于给定问题先在文档中找到和答案相关的语句。目前该 领域大部分算法都使用递归神经网络，但由于很难序列并行化，这类算法在长文档上运行很慢。受人类在首次浏览文章 时识别与问题相关的段落或语句，并仔细阅读这些内容得到答案的启发，本文采用一个粗糙但快速的模型用于答案相关 句的选择。实验在WIKIREADING LONG 数据集上取得了较好的结果。 |
| 关键词: 机器阅读理解软件；神经网络；答案相关句 |
|
中图分类号: TP391.1
文献标识码: A
|
|
| A Study of the Question-Aware Passage Extraction in Machine Reading Comprehension Software |
|
LIU Haijing
|
( Department of Computer Engineering, Taiyuan Institute of Technology, Taiyuan 030008, China)
|
| Abstract: In the programming of machine reading comprehension software,one basic step to extract the relevant sentences from the whole passage according to the given question.Recently,Recurrent Neural Network (RNN) is applied in most approaches,but it runs excessively slow when dealing with long documents because it is difficult to parallelize sequences.Inspired by the process that people first skim the document,identify relevant parts,and carefully read these parts to produce an answer,the paper proposes a rough but efficient model to extract the relevant sentences.Experiment results demonstrate the good performance on WIKIREADING LONG datasets. |
| Keywords: machine reading comprehension software;RNN;relevant sentences |
