Herzog Christian, Lunn Brian Kierkegaard
Digital Science, 90 York Way, London, N1 9AG UK.
Scientometrics. 2018;117(1):641-645. doi: 10.1007/s11192-018-2854-z. Epub 2018 Jul 27.
With Dimensions, Digital Science provides the research community a new approach on research related information, bringing formerly siloed content types such as grants, patents, clinical trials with publications and citations together, making it as openly available as possible (see app.dimensions.ai). Due to the different content types, (controversial) journal based classifications were not an option since it would not allow to categorise grants etc. Hence Digital Science opted for applying a categorisation approach using machine learning and based on the content of the documents and well established classification systems for which a training set was available. The implementation at launch was a first step and requires to be improved-although we observe a reliability comparably to manual coding for grants, the implementation at launch comes with some shortcomings as observed by Bornmann (2018), mostly due to challenges with the training set coverage. To overcome the shortcomings of the initial categorization approach we implemented an improvement process with the research community and Lutz Bornmann's analysis presented a great opportunity to provide more transparency and insights in the ongoing improvements.
借助Dimensions，数字科学为研究界提供了一种处理研究相关信息的新方法，将以前孤立的内容类型（如资助、专利、临床试验以及出版物和引用文献）整合在一起，并尽可能使其公开可用（见app.dimensions.ai）。由于内容类型不同，基于期刊的（有争议的）分类方法不可行，因为它无法对资助等进行分类。因此，数字科学选择采用一种基于机器学习、依据文档内容以及有可用训练集的成熟分类系统的分类方法。发布时的实施是第一步，需要改进——尽管我们观察到在资助方面其可靠性与人工编码相当，但正如博尔曼（2018年）所指出的，发布时的实施存在一些缺点，主要是由于训练集覆盖方面的挑战。为了克服初始分类方法的缺点，我们与研究界共同实施了一个改进过程，而卢茨·博尔曼的分析为在持续改进中提供更多透明度和见解提供了绝佳机会。
