Automated Indexing for Publication Types and Study Designs

项目摘要/摘要 从书目数据库中检索生物医学文章需要对主题进行准确、详细的索引 以及它们的出版类型和研究设计。索引者很难跟上 鉴于出版文献的爆炸性增长，手工作业。尽管NLM最近聘请了 自动机器学习方法根据主要主题对文章进行索引讨论，目前还没有 自动为所有出版物类型和研究设计中的每篇文章编制索引。我们最近做了 创建了一个工作原型工具，多标记器，它为所有人分配了概率预测分数 针对50种不同出版类型和研究设计的PubMed文章(合计为PTS)。我们现在建议 开发多标签器2.0，以处理更广泛的研究设计、文章、用户和用例，并 确保以适合每种不同类型用户的形式传播数据。具体来说， 我们的目标是： 目标1.优化为PubMed和非PubMed分配出版物类型和研究设计的方法 PubMed生物医学文章、预印本和手稿。 目标2.详细评估PTS，考虑模型性能、用例和用户。 目的3.通过查询接口和API优化PT预测分数的发布。 目的4.探索如何将多标记器与其他工具集成，以实现证据合成的自动化。 拟议的研究将大大加强相关文章和预印本在多个 数据库，并将对广泛的生物医学最终用户(临床医生、研究人员、学生和 期刊编辑)以及包括系统审查小组、书目数据库管理员、 那些研究人类疾病的临床前动物模型的人，以及制药公司开发新的 药物治疗。因此，完善生物医学文献的基础设施将间接影响人类 健康。

Project Summary/Abstract Retrieving biomedical articles from bibliographic databases requires accurate, detailed indexing of the topics that are discussed as well as their publication types and study designs. It is difficult for indexers to keep up with manual assignments in view of the explosion of published literature. Although NLM has recently employed automatic machine learning methods to index articles according to the major topics discussed, there is still no automatic means of indexing each article across all publication types and study designs. We have recently created a working prototype tool, Multi-Tagger, which has assigned probabilistic predictive scores for all PubMed articles for 50 different publication types and study designs (collectively, PTs). We now propose to develop Multi-Tagger 2.0, to handle a wider variety of study designs, articles, users and use cases, and to ensure that the data are disseminated in a form that is appropriate to each different type of user. Specifically, we aim to: Aim 1. Optimize methods for assigning Publication Types and study designs to both PubMed and non- PubMed biomedical articles, preprints and manuscripts. Aim 2. Evaluate PTs in detail, taking into account model performance, use cases and users. Aim 3. Optimize dissemination of PT predictive scores by query interface and API. Aim 4. Explore how to integrate Multi-Tagger with other tools for automating evidence synthesis. The proposed studies will greatly enhance retrieval of relevant articles and preprints across multiple databases, and will be useful for a wide range of biomedical end-users (clinicians, researchers, students and journal editors) as well as user groups including systematic review groups, bibliographic database managers, those studying preclinical animal models of human disease, and pharmaceutical companies developing new drug treatments. Improving the infrastructure of the biomedical literature will thus indirectly impact on human health.