大概内容:首先简要介绍一下常用的一些框架及标准,其次参照PRISMA以及SPAR-4-SLR,从searching, screening, data management /analysis 三个维度介绍一些方便的工具。
参照:
Tools (ioe.ac.uk)
Software and Tools - Systematic Reviews - LibGuides at Lancaster University
Text mining for searching and screening the literature - Guides at McGill Library
0. Protocol/framework
这个在我的一篇公众号文章中 如何写文献综述: 概念 分类及规则 中提到了一些常用的,比如
PRISMA ( Preferred Reporting Items for Systematic Reviews and Meta-Analyses), 提供了一系列的checklist及可供研究者自行创造流程图: PRISMA Flow Diagram (shinyapps.io). 还有一些extension.
PICO Framework. Population - Characteristics of the population or sample; Intervention - Treatment or exposure being studied; Comparison - Alternative treatment or exposure as control; Outcome - Results or endpoints measured.
Cochrane Library. 提供了很详细的handbook来操作SR,常用于healthcare.
EPPI-Centre. Developed by the Evidence for Policy and Practice Information Centre
SPAR-4-SLR. Structured Procedure for Aiding the Reproduction of Systematic Literature Reviews. Business, social 可考虑
ADO. Antecedents, Decisions, and Outcomes framework. Paul and Benito (2018)
ROSES pro forma and flow diagram. RepOrting standards for Systematic Evidence Syntheses, 常用于环境科学。ROSES Flow Chart (shinyapps.io)
</br>
1. Searching/Identification
关键在于 定义研究概念 -> 确定研究问题 -> 计划搜索策略
对于领域尚在发展 或没有什么已定的terminology, ontology, model, dictionaries 的情况下,最好是LLM (NLP) 进行 search keywords 的 探索性搜索 (当然,要基于较为完整的理论知识)。
SR-Accelerator. 可识别 n-gram
AntConc, 可n-gram, 聚类,常用场景是已识别出高频词后 为了增加搜索的precision (如用在SR-Accelerator之后)
Termine. 比较old了,不大推荐使用
LitSearchR, 综合了RAKE keyword frequency/n-gram, grouping, network analysis 等的R package. 有提供shiny界面,但shiny会受限于上传文件的默认大小 (5MB)。
Yale MeSH Analyzer. 依旧适用于MeSH terms 也就是医学主题词, 受限于MeSH database.
PubMed PubReMiner: a tool for PubMed query building and literature mining (amc.nl) 适用于pubmed的搜索,免费
Ananse. 参考 An automated method for developing search strategies for systematic review using Natural Language Processing (NLP) - ScienceDirect,用了litsearchr 类似的techniques
</br>
2. Screening
这里的screening包含 semi-automation screening tool, 以及方便统一管理进行screening, inclusion & exclusion 的软件。
Abstrackr semi-automated, 注册后免费使用
ASReview,半自动化screening process, 用的是主动学习,可下载portable
SWIFT-Review. Uses statistical text mining and machine learning to rank order documents in the screening process
revtools | An R package for screening in evidence synthesis projects
DistillerAI Paid (free for librarians; contact them directly)
Covidence. free for Cochrane authors and fee-based for others
Rayyan. completely free tool,相对covidence来讲,没有prescribed workflow,
EPPI-Reviewer. knowledge synthesis. Paid (free for Cochrane collaborators).
</br>
3. Data management/analysis
包括文献管理、后续分析、报告、描述、可视化等等。
文献管理自然是诸如 endnote, mendely, zotero 之类的,可以实现remove duplicates, unify, bulk export 等等
分析自是可以根据自己需求,做SR SLR, 文献分析,文本分析等等,描述报告方式根据方法不同略有差异
可视化可使用之前介绍过的一些知识图谱/网络的方法,可用工具或者python, R 等实现,这里不赘述。