基于Siamese-BIGRU-Attention的语义相似度分析研究  

Semantic Similarity Analysis based on Siamese-BIGRU-Attention

作　　者：陈鑫 邱占芝[1] CHEN Xin;QIU Zhanzhi(School of Mechanical Engineering,Dalian Jiaotong University,Dalian 116028,China)

机构地区：[1]大连交通大学机械工程学院,辽宁大连116028

出　　处：《大连交通大学学报》2022年第4期113-116,共4页Journal of Dalian Jiaotong University

基　　金：国家自然科学基金资助项目(61074029)。

摘　　要：提出了Siamese-BIGRU-Attention语义相似度计算方法,该方法在Siamese Network的基础框架上引入了BIGRU和attention模块,增强了前后文的捕捉能力并充分利用上下文信息挖掘因果关系,实现对输入文本更好的表征.实验表明:基于Siamese-BIGRU-Attention的语义相似度计算方法展现了更好的性能,其准确率及F1 score值均得到了一定的提升.The Siamese-BIGRU-Attention semantic similarity calculation method is proposed.It introduces BIGRU and Attention modules on the basis of Siamese Network,enhances the ability to capture contextual information and makes full use of contextual information to mining causal relationship to realize better representation of input text.Experiment results show that the semantic similarity calculation method based on Siamese-Bigru-attention presents better performance with significantly improved accuracy F1 score.

关 键 词：语义相似度 双向门控循环单元 注意力机制 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]