基于命题逻辑与数据增强的逻辑推理答题方法  

Logical reasoning answering method based on propositionallogic and data augmentation

作　　者：张杰 张虎[1] 王宇杰 ZHANG Jie;ZHANG Hu;WANG Yu-jie(School of Computer and Information Technology,Shanxi University,Taiyuan 030006,China)

机构地区：[1]山西大学计算机与信息技术学院,山西太原030006

出　　处：《计算机工程与设计》2025年第1期190-197,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(62176145)。

摘　　要：为解决现有面向机器阅读理解的逻辑推理方法对文本中存在的深层逻辑关系利用不足与现有的逻辑推理数据集规模较小的问题,提出一种基于命题逻辑与数据增强的逻辑推理答题方法。将命题逻辑引入到逻辑推理任务,构建符合文本逻辑推理的新规则,将其应用到文本逻辑图的扩充过程中,增加文本中可利用的深层逻辑关系;结合文本中存在的逻辑关系,提出并列关系替换、推理词和推理词组替换的数据增强策略。实验结果表明,该方法比以往的方法取得了更好的效果。To solve the problem that the existing logical reasoning methods for machine reading comprehension make insufficient use of the deep logical relationships existing in the text and the small scale of the existing logical reasoning datasets,a logical reasoning answering method based on propositional logic and data enhancement was proposed.The propositional logic was introduced into the logical reasoning task,a new rule in line with text logical reasoning was constructed,and it was applied to the expansion process of text logic graph,which increased the deep logical relationships available in the text.Combined with the logical relationships existing in the text,the data augmentation strategies of juxtaposition relationship substitution,inference word and inference phrase substitution were proposed.Experimental results show that this method achieves better results than previous methods.

关 键 词：阅读理解 逻辑推理 逻辑关系 命题逻辑 文本逻辑图 数据增强 神经推理 符号推理 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]