基于依存句法的初等数学分层抽样应用题题意理解

数学应用题自动求解,即利用计算机对自然语言描述的应用题进行自动理解和作答,一直是人工智能领域研究的重难点和核心目标之一。针对应用题语义复杂、上下文情景多变、关键参数难以准确识别的问题,提出一种基于依存句法的初等数学分层抽样应用题题意理解方法。通过构建一个面向初等数学分层抽样类应用题的句模库,并结合依存句法来实现分层抽样应用题解题信息的自动抽取。实验研究发现,与仅基于句模的信息抽取方法相比,该方法对不同语义角色的句子的信息抽取准确率均有一定提升,整题理解的准确率从40%上升至68%。

STRATIFIED SAMPLING WORD PROBLEM UNDERSTANDING OF ELEMENTARY MATHEMATIC BASED ON DEPENDENCY PARSING

Automatic math word problem solving,which is to comprehend and solve the word problems that described in natural language automatically by computer,has always been one of the key points in the field of artificial intelligence.There are many problems in automatic problem solving,such as complex semantics,variable context and impalpable parameters.Focus on these problems,we proposed stratified sampling problem understanding of elementary mathematics based on dependency parsing.Combining the construction of sentence template library for elementary mathematics stratified sampling problems with the dependency parsing,the automatic extraction of question information was realized.The experimental results indicated that,comparing with the extraction method only based on sentence templates,the proposed method can generally improve the information extraction accuracy for sentences with different semantic roles,and the accuracy of comprehension of the whole question increased from 40%to 68%.