摘要
机器阅读理解的目标是使机器更好地理解自然语言文本并在此基础上回答提出的问题,是自然语言处理领域热门的研究方向之一。早期,由于受到了数据集的约束,对机器阅读理解的认识大多仅限于一个单跳式的问答。随着最近几年多跳机器阅读理解数据集的发展,多跳式机器阅读理解得到广泛的研究,极大地推动了机器阅读理解领域的发展。从以下几个方面对基于多跳式的机器阅读理解进行归纳总结:介绍了机器阅读理解任务定义与发展历程;阐述了多跳式机器阅读理解任务定义并梳理总结相关数据集;详细整理了多跳式机器阅读理解基于注意力机制和图神经网络以及问题分解相关模型方法的研究进展;最后,对多跳式机器阅读理解未来研究重点和所面临的研究挑战进行展望。
The goal of machine reading comprehension is to enable machines to better understand natural language texts and answer questions on this basis,which is one of the hot research directions in the field of natural language processing.In the early days,due to the constraints of data sets,the understanding of machine reading comprehension was mostly limited to a single-hop question and answer.With the development of multi-hop machine reading comprehension data sets in recent years,multi-hop machine reading comprehension has been widely studied,which has greatly promoted the development of the field of machine reading comprehension.We summarize the multi-hop machine reading comprehension from the following aspects:Introduce the definition and development of machine reading comprehension task,expound the definition of multi-hop machine reading comprehension task,and sort out the related data sets.The research progress of multi-hop machine reading comprehension based on attention mechanism,graph neural network and problem decomposition related models is sorted out in detail.Finally,the future research emphases and challenges of multi-hop machine reading comprehension are prospected.
作者
仇亚进
奚雪峰
崔志明
盛胜利
周悦尧
QIU Ya-jin;XI Xue-feng;CUI Zhi-ming;SHENG Sheng-li;ZHOU Yue-yao(School of Electronic and Information Engineering,Suzhou University of Science and Technology,Suzhou 215000,China;Suzhou Key Laboratory of Virtual Reality Intelligent Interaction and Application Technology,Suzhou 215000,China;Suzhou Smart City Research Institute,Suzhou 215000,China;Texas Institute of Technology,Lubbock 79401,USA)
出处
《计算机技术与发展》
2023年第2期9-16,23,共9页
Computer Technology and Development
基金
国家自然科学基金(61876217,62176175)
江苏省“六大人才高峰”高层次人才项目(XYDXX-086)。
关键词
多跳式机器阅读理解
注意力机制
图神经网络
问题分解
数据集
muti-hop machine reading comprehension
attention mechanism
graph neural network
question decomposition
dataset
