计算机工程与应用 ›› 2022, Vol. 58 ›› Issue (17): 23-33.DOI: 10.3778/j.issn.1002-8331.2201-0384

• 热点与综述 • 上一篇    下一篇

问答系统中复杂问题分解方法研究综述

冯钧,李艳,杭婷婷   

  1. 河海大学 计算机与信息学院 水利部水利大数据重点实验室,南京 211100
  • 出版日期:2022-09-01 发布日期:2022-09-01

Survey on Question Decomposition Method in Question Answering System

FENG Jun, LI Yan, HANG Tingting   

  1. Key Laboratory of Water Big Data Technology of Ministry of Water Resources, School of Computer and Information, Hohai University, Nanjing 211100, China
  • Online:2022-09-01 Published:2022-09-01

摘要: 问答系统可以针对用户提出的自然语言问题给出精准的答案,是自然语言处理领域中一个重要的研究方向。对于具有复杂语义结构和句法结构的多跳问题,模型需要强大的自然语言理解能力。问题分解作为问题理解的一种技术,有着不可估量的作用。阐述了问题分解的研究背景与意义;根据问题特征提取的方式,将现有的方法分为传统机器学习方法和深度学习方法两大类,传统机器学习方法以规则模板匹配和基于分割的方法为主,深度学习方法以基于Transformer、图神经网络、注意力机制、查询图和强化学习为主,并分别从模型架构、优势、劣势等方面进行分析。结合目前研究的动态,初步展望了未来的研究方向。

关键词: 问答系统, 复杂问题, 问题分解, 机器学习, 深度学习

Abstract: Question answering system can give accurate answers to natural language questions raised by users, and is an important research directions in the field of natural language processing. For multi-hop questions with complex semantic and syntactic structures, models require strong natural language understanding capabilities. As a technique of question understanding, question decomposition plays an immeasurable role. Firstly, the research background and significance of question decomposition are introduced. Then, according to the method of question feature extraction, the existing methods are divided into two categories:traditional machine learning methods and deep learning methods. Traditional machine learning methods are divided into rule template matching and segmentation-based methods. Deep learning methods are divided into Transformer, graph neural network, attention mechanism, query graph and reinforcement learning methods. And, it analyzes from the aspects of model architecture, advantages and disadvantages. Finally, combined with the current research trends, the future research directions are discussed.

Key words: question answering system, complex question, question decomposition, machine learning, deep learning