Classifying papers into subfields using Abstracts, Titles, Keywords and KeyWords Plus through pattern detection and optimization procedures: An application in Physics
提出一种利用论文元数据（摘要、标题、关键词等）和机器学习模式检测来将论文自动分类到物理学子领域的方法，可帮助科研评价、期刊编辑和数据库优化分类。
Abstract Classifying papers according to the fields of knowledge is critical to clearly understand the dynamics of scientific (sub)fields, their leading questions, and trends. Most studies rely on journal categories defined by popular databases such as WoS or Scopus, but some experts find that those categories may not correctly map the existing subfields nor identify the subfield of a specific article. This study addresses the classification problem using data from each paper (Abstract, Title, Keywords, and the KeyWords Plus) and the help of experts to identify the existing subfields and journals exclusive of each subfield. These “exclusive journals” are critical to obtain, through a pattern detection procedure that uses machine learning techniques (from software NVivo), a list of the frequent terms that are specific to each subfield. With that list of terms and with the help of optimization procedures, we can identify to which subfield each paper most likely belongs. This study can contribute to support scientific policy‐makers, funding, and research institutions—via more accurate academic performance evaluations—, to support editors in their tasks to redefine the scopes of journals, and to support popular databases in their processes of refining categories.