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
胡献君 1杜莹 2林鑫3
作者信息
- 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. 华中师范大学信息管理学院,湖北武汉 430079;湖北省数据治理与智能决策研究中心,湖北武汉 430079
- 折叠
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.
Abstract
[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 manage-ment 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 ef-fect 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 experi-mental 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Key words
machine translation/BERT-Whitening/synonymous sentence recognition/synonym/deep learning引用本文复制引用
出版年
2024
NETL
NSTL
万方数据