基于外部知识检测的数学几何单词问题求解  

Solving Mathematical Geometry Word Problem Basedon External Knowledge Detection

作　　者：魏亚琴 刘斌[1] 张倩 崔学英[1] 谢秀峰[1] WEI Ya-qin;LIU Bin;ZHANG Qian;CUI Xue-ying;XIE Xiu-feng(School of Applied Science,Taiyuan University of Science and Technology,Taiyuan 030024,China)

机构地区：[1]太原科技大学应用科学学院,太原030024

出　　处：《太原科技大学学报》2024年第5期507-513,共7页Journal of Taiyuan University of Science and Technology

基　　金：国家自然科学基金(11701406,11901134,12061091,11972019);山西省基础研究计划项目(202103021224274,201901D111261);山西省社会经济统计科研课题(KY[2022]73);山西省省筹资金资助回国留学人员科研项目(2022-163)。

摘　　要：随着人工智能技术的不断发展,深度学习方法在数学单词问题智能求解方面得到了广泛应用。基于图到树的编解码网络表现出良好的性能,但外部知识缺乏使得模型求解准确率提升受到限制,构建一种基于几何知识库的检测模块,将其融入编解码器网络模型,提高了模型的预测能力。通过在中文数学问题数据集GeometryQA上进行验证,模型表现出更高的准确率,具有一定的优越性。With the continuous development of artificial intelligence technology,deep learning methods have been widely used in the intelligent solution of math word problem.Graph-to-tree based encoder-decoder networks show good performance,but the lack of external knowledge limits the improvement of model solving accuracy.A detection module based on a geometric knowledge base was constructed and incorporated into the encoder-decoder net-work model to improve the predictive capability of the model.Through validation on the Chinese mathematical problem dataset GeometryQA,the model exhibits higher accuracy and has certain superiority.

关 键 词：自然语言处理 数学单词问题 几何知识库 检测模块 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]