自动数学应用题解算器研究综述  

Review of research on automatic math word problem solver

作　　者：张珑 杨波 罗琨杰 ZHANG Long;YANG Bo;LUO Kunjie(College of Computer and Information Engineering,Tianjin Normal University,Tianjin 300387,China)

机构地区：[1]天津师范大学计算机与信息工程学院,天津300387

出　　处：《天津师范大学学报（自然科学版）》2021年第2期1-9,共9页Journal of Tianjin Normal University：Natural Science Edition

基　　金：国家自然科学基金面上资助项目(61771173).

摘　　要：从2个层面综述近年来自动数学应用题解算器的相关研究.首先,从数据准备层面总结应用于解算器设计的数据集的特征,以及数学应用题的自动生成方法;其次,从数学应用题解算方法层面分类介绍解算器模型,包括基于模板匹配的方法、基于统计分类的方法、基于树或图的图形方法和基于深度学习框架的方法,并分析了各类型的核心算法及性能,此外,介绍了解算器性能的评估策略;最后,指出目前研究存在的问题并对该领域的发展提出了可能的研究方向.The related researches on automatic math word problem solver in recent years are reviewed from two aspects.Firstly,the characteristics of data sets commonly used in solver designing are summarized from the data preparation level,and the automatic generation methods of math word problem are also summarized.Secondly,the solver models which are classified into four categories are introduced from the level of solving methods,including solvers based on template matching,solvers based on statistical classification,solvers based on tree or graph and solvers based on deep learning.The core algorithms and performance of each category are analyzed.Moreover,the evaluation strategies for performance of automatic math word problem solver are introduced.Finally,the existing problems in the present researches are pointed out,and the possible research directions for the development of this domain are proposed.

关 键 词：数学应用题解算器 模板匹配 表达式树 单位依赖图 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]