학술논문
面向投稿选刊的学术论文多标签分类研究 / Research on Multi-label Classification of Academic Papers for Periodical Selection
Document Type
Academic Journal
Author
Source
现代情报 / Modern Information. 44(1):48-108
Subject
Language
Chinese
ISSN
1008-0821
Abstract
[目的/意义]学术论文投稿中面临期刊选择多样性和拒稿重投问题,研究利用深度学习和多标签分类技术,基于论文题录信息给出多标签的投稿选刊建议.[方法/过程]选取情报学领域 8 种CSSCI期刊近 20年的论文作为样本,采用TextCNN、TextRNN等深度学习模型和预训练语言模型BERT构建多标签分类方法进行实验,并对比不同特征组合和多标签设置策略下的实验效果.[结果/结论]多标签分类能够反映学术论文对不同期刊的适合度,预训练语言模型BERT表现最佳,F1 达到 68.99%.
[Purpose/Significance]The academic paper submission is faced with the problems of journal selection di-versity and re-submission,this paper studies the use of machine learning technology to give multi-label recommendations for periodical submission based on the content of the academic paper.[Method/Process]Papers from 8 CSSCI journals in the field of information science in recent 20 years were selected as samples,TextCNN,TextRNN,and pre-trained lan-guage model BERT were used for experiments,and the experimental effects under different feature combinations and multi-label setting strategies were compared.[Result/Conclusion]Multi-label classification can reflect the suitability of articles for different periodical,and the pre-trained language model BERT performs best,with F1 reaching 68.99%.
[Purpose/Significance]The academic paper submission is faced with the problems of journal selection di-versity and re-submission,this paper studies the use of machine learning technology to give multi-label recommendations for periodical submission based on the content of the academic paper.[Method/Process]Papers from 8 CSSCI journals in the field of information science in recent 20 years were selected as samples,TextCNN,TextRNN,and pre-trained lan-guage model BERT were used for experiments,and the experimental effects under different feature combinations and multi-label setting strategies were compared.[Result/Conclusion]Multi-label classification can reflect the suitability of articles for different periodical,and the pre-trained language model BERT performs best,with F1 reaching 68.99%.