现代情报2025，Vol.45Issue(10)：3-15,13.DOI:10.3969/j.issn.1008-0821.2025.10.001
基于深度域适应方法的非结构化摘要功能识别研究
Research on Non-Structural Abstract Function Recognition Based on Deep Domain Adaptation Method
摘要
Abstract
[Purpose/Significance]The identification of structural functions in academic abstracts involves recognizing specific functions such as research background,methods,results,and conclusions from abstracts.This study addresses the issues of heavy data annotation work,insufficient domain transfer capability,and poor model interpretability that are commonly present in the task of non-structural abstract function recognition.[Method/Process]Using a deep domain adap-tation method based on the concept of transfer learning to construct a model for recognizing structural functions in abstracts.Specifically,employed ALBERT,a pre-trained language model to extract contextual features from abstracts.Then,under the joint action of the feature extractor,domain discriminator,and category classifier,the structural func-tional feature knowledge was transferred from the source domain with labeled data to the target domain with unlabeled data,aiming to achieve the cross-domain transferability without the need for annotated training samples.In addition,the SHAP method was used to provide interpretability analysis for the model's output results,and experiments were conducted on both the target domain(COVID-19 dataset)and the source domain(PubMed 20K dataset).[Result/Conclusion]The model based on the deep domain adaptation method achieves better recognition performance than the baseline models,with the superior performance and better interpretability on the"methods"and"results"functions.The experimental results indi-cate that the optimal model can achieve knowledge transfer from the source domain to the target domain in an unsupervised learning manner,reducing the model's dependence on data annotation and enhancing the model's portability for this task.关键词
摘要结构/功能识别/领域自适应/深度学习/无监督学习Key words
abstract structure/function recognition/domain adaptation/deep learning/unsupervised learning分类
社会科学引用本文复制引用
施顺顺,苟震宇,张琳,黄颖..基于深度域适应方法的非结构化摘要功能识别研究[J].现代情报,2025,45(10):3-15,13.基金项目
国家自然科学基金面上项目"从测度到理解:跨学科研究的成果分类、合作模式与影响扩散研究"(项目编号:72374160). （项目编号:72374160）
