当前位置: X-MOL 学术Appl. Soft Comput. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
A Multi-Channel Convolutional Neural Network approach to automate the citation screening process
Applied Soft Computing ( IF 8.7 ) Pub Date : 2021-07-30 , DOI: 10.1016/j.asoc.2021.107765
Raymon van Dinter 1 , Cagatay Catal 2 , Bedir Tekinerdogan 1
Affiliation  

The systematic literature review (SLR) process is separated into several steps to increase rigor and reproducibility. The selection of primary studies (i.e., citation screening) is an important step in the SLR process. The citation screening process aims to identify the relevant primary studies fairly and with high rigor using selection criteria. Through the study selection criteria, reviewers determine whether an article should be included or excluded from the SLR. However, the screening process is highly time-consuming and error-prone as the researchers must read each title and possibly hundreds to thousands of abstracts and full-text documents. This study aims to automate the citation screening process using Deep Learning algorithms. With this, it is aimed to reduce the time and costs of the citation screening process and increase the precision and recall of the relevant primary studies. A Multi-Channel Convolutional Neural Network (CNN) is proposed, which can automatically classify a given set of citations. As the architecture uses the title and abstract as features, our end-to-end pipeline is domain-independent. We have performed six experiments to assess the performance of Multi-Channel CNNs across 20 publicly available systematic literature review datasets. It was shown that for 18 out of 20 review datasets, the proposed method achieved significant workload savings of at least 10%, while in several cases, our model yielded a statistically significantly better performance over two benchmark review datasets. We conclude that Multi-Channel CNNs are effective for the citation screening process in SLRs. Multi-Channel CNNs perform best on large datasets of over 2500 samples with few abstracts missing.



中文翻译:

一种多通道卷积神经网络方法来自动化引文筛选过程

系统文献审查 (SLR) 过程分为几个步骤,以提高严谨性和可重复性。主要研究的选择(即引文筛选)是 SLR 过程中的一个重要步骤。引文筛选过程旨在使用选择标准公平且严格地识别相关的初级研究。通过研究选择标准,审稿人决定是否应将一篇文章纳入或排除在 SLR 之外。然而,筛选过程非常耗时且容易出错,因为研究人员必须阅读每个标题,可能还要阅读成百上千的摘要和全文文档。本研究旨在使用深度学习算法自动化引文筛选过程。有了这个,它旨在减少引文筛选过程的时间和成本,并提高相关初级研究的精确度和召回率。提出了一种多通道卷积神经网络(CNN),它可以自动对给定的引用集进行分类。由于架构使用标题和摘要作为特征,因此我们的端到端管道与领域无关。我们已经进行了六个实验来评估多通道 CNN 在 20 个公开可用的系统文献综述数据集中的性能。结果表明,对于 20 个评论数据集中的 18 个,所提出的方法实现了至少 10% 的显着工作负载节省,而在一些情况下,我们的模型在统计上比两个基准评论数据集产生了显着更好的性能。我们得出结论,多通道 CNN 对 SLR 中的引文筛选过程是有效的。多通道 CNN 在 2500 多个样本的大型数据集上表现最佳,几乎没有缺少摘要。

更新日期:2021-08-05
down
wechat
bug