学术文本结构功能深度学习识别方法的多学科对比分析  

Multi-disciplinary Comparative Study on Methods of Academic Text Structure Function Recognition Based on Deep Learning Model

作　　者：李楠[1] 方丽 张逸飞 Li Nan;Fang Li;Zhang Yifei(Institute of Information Science and Technology Information,East China University of Science and Technology,Shanghai 200237,China;School of Information Science and Engineering,East China University of Science and Technology,Shanghai 200237,China)

机构地区：[1]华东理工大学科技信息研究所,上海200237 [2]华东理工大学信息科学与工程学院,上海200237

出　　处：《现代情报》2019年第12期55-63,87,共10页Journal of Modern Information

摘　　要：[目的/意义]学术文本的结构功能识别可视为多类别文本自动分类问题,借助深度学习技术能够获得良好的自动识别性能,然而目前缺少其在不同学科适用性的对比研究。[方法/过程]选择医学、图情、数据、出版、经济5个学科方向5种期刊的6 452篇结构式摘要为基础语料,设计并实现了基于Magpie深度学习组件的学术文本结构功能识别实验,通过对比分析同一分类模型在不同学科领域实验语料上的性能表现及其影响因素,揭示机器学习方法的学科适用性规律。[结果/结论]实验结果显示,学科差异性对于机器学习效果有显著的影响,其中医学领域学术文本的结构功能识别效率明显高于其他学科,常见的学术文本功能结构框架中"方法"和"结果"的机器学习识别效果更佳。[Purpose/Significance]As an automatic classification task of multi-category,the structure function recognition of academic text can achieve good recognition performance by using deep learning method.However,there is a lack of comparative research on its applicability in different disciplines.[Method/Process]This paper built a multi-disciplinary academic text data sets,composed by 6452 structured abstracts of academic articles from 5 journals in different disciplines including medical science,library and information science,data science,publishing science,economics.And the experiment of structure function recognition of academic text based on Magpie,an open-source deep learning component,was conducted to investigate the applicability of the same classification model in different disciplines by analyzing the experimental performance and influencing factors.[Result/Conclusion]The results showed that discipline differences had a significant impact on the performance of deep learning method,among which the structural function recognition efficiency of academic texts in the medical discipline was significantly higher than that of other disciplines,and the recognition efficiency of"method"and"result"in the common functional structure framework of academic texts was better.

关 键 词：文本结构功能识别 深度学习 多学科 文本分类 MAGPIE 

分 类 号：G203[文化科学—传播学]