融合机器翻译与BERT-Whitening的同义句识别研究  

Synonymous Sentence Recognition Based on Machine Translation and BERT-Whitening

作　　者：胡献君[1] 杜莹 林鑫 HU Xianjun;DU Ying;LIN Xin(School of Electrical and Electronic Engineering,Naval University of Engineering,Wuhan 430079,China;School of Information Management,Central China Normal University,Wuhan 430079,China;Hubei Data Governance and Intelligent Decision Research Center,Wuhan 430079,China)

机构地区：[1]海军工程大学电子工程学院,湖北武汉430033 [2]华中师范大学信息管理学院,湖北武汉430079 [3]湖北省数据治理与智能决策研究中心,湖北武汉430079

出　　处：《情报科学》2024年第6期83-88,98,共7页Information Science

基　　金：国家社会科学基金重点项目“国家科研论文和科技信息高端交流平台构建研究”(22ATQ006)。

摘　　要：【目的/意义】构建机器翻译与BERT-Whitening结合的句子同义识别模型,可以提升同义句识别效果,为下游的信息资源管理与服务应用提供支撑。【方法/过程】首先对同义句的类型及特点进行分析,在此基础上构建融合机器翻译与BERT-Whitening的同义句识别模型,并通过实验对模型效果进行验证。其中,识别模型由句子预处理、候选同义句识别、嵌入式文本表示与基于相似度融合的同义判断等四个部分构成。【结果/结论】实验结果表明,机器翻译与BERT-Whitening结合模型的准确率、召回率和F1分别达到了0.840、0.859和0.849,明显高于对照组。【创新/局限】未在专业性较强的领域文本验证,普适性验证不足,且准确率、召回率提升空间较大。【Purpose/significance】 Constructing a sentence synonym recognition model combining machine translation and BERT-Whitening can improve the effect of synonym sentence recognition and provide support for downstream information resource management and service applications. 【Method/process】 Firstly, the types and characteristics of synonymous sentences are analyzed, and on this basis, a synonymous sentence recognition model integrating machine translation and BERT-Whitening is constructed, and the effect of the model is verified through experiments. The recognition model consists of four elements: sentence preprocessing, candidate synonym recognition, embedded text representation and synonym judgment based on similarity fusion. 【Result/conclusion】 The experimental results show that the accuracy, recall and F1 of the combined model of machine translation and BERT-Whitening are 0.840,0.859 and 0.849, respectively, significantly higher than the control Group. 【Innovation/limitation】 Lack of text verification in highly specialized fields, insufficient universality verification, and significant room for improvement in accuracy and recall.

关 键 词：机器翻译 BERT-Whitening 同义句识别 同义 深度学习 

分 类 号：TP391.2[自动化与计算机技术—计算机应用技术]