现代情报2025，Vol.45Issue(8)：18-30,13.DOI:10.3969/j.issn.1008-0821.2025.08.002
面向学术文本的语义增强自然语言推理模型
A Semantic-Enhanced Natural Language Inference Model for Academic Texts
摘要
Abstract
[Purpose/Significance]The paper aims to generate high-quality synonymous sentences for academic texts utilizing large language models and enhance the performance of natural language inference model through the implementation of semantic enhancement strategies.[Method/Process]Based on the utilization of large language model to generate synony-mous sentences for academic texts,the paper proposed a semantic-enhanced natural language inference model,SENLI.The model consisted of a representation module,a semantic enhancement module,and an inference module.Specifically,the representation module was responsible for capturing the semantic features of academic texts and their corresponding syn-onymous sentences.The semantic enhancement module integrated the semantic features of the synonymous sentences into the original semantic features of the academic texts,thereby obtaining semantic-enhanced features.Finally,the inference module predicted the semantic relationship between pairs of academic texts based on both the original semantic features and the semantic-enhanced features.The study conducted an empirical study by applying the proposed model to the SciTail,SciNLI,and ZwNLI datasets.[Result/Conclusion]The experimental results show that the SENLI model achieves accuracy rates of 95.11％,79.20％,and 98.43％ on the SciTail,SciNLI,and ZwNLI datasets,respectively.Compared to the baseline models,the improvements are at least 1.27％,1.08％,and 0.92％,demonstrating the effectiveness of the pro-posed model.The utilization of synonymous sentences generated by large language models for semantic enhancement can en-hance the performance of natural language inference model.The research contributes to advancing the field of natural lan-guage inference and provides potential technical support for applications such as information retrieval and academic literature mining.关键词
自然语言推理/学术文本/语义增强/深度学习/大语言模型Key words
natural language inference/academic text/semantic enhancement/deep learning/large language model分类
信息技术与安全科学引用本文复制引用
张贞港,余传明,王静楠..面向学术文本的语义增强自然语言推理模型[J].现代情报,2025,45(8):18-30,13.基金项目
国家自然科学基金面上项目"基于知识增强的科技文献创新识别与评价模型研究"(项目编号:72374219) （项目编号:72374219）
"面向跨语言观点摘要的领域知识表示与融合模型研究"(项目编号:71974202) （项目编号:71974202）
中南财经政法大学中央高校基本科研业务费(项目编号:202411401). （项目编号:202411401）
