Semi-automatic Metadata Extraction from Scientific Journal Article for Full-text XML Conversion

摘要

By the increasing continuous academic researches, the volume of scientific articles has dramatically reached unpredictable level. To facilitate archive and publication, many scientific journals in Korea are actively adapting Open Access (OA) policy. In addition, it has more attractable than commercial printing of companies that freely provide the full text of article published in scholarly journal through web to user. Because of difficulty to convert automatically unstructured format such as pdf document into full-text, which is structured with accuracy, the most full text conversion works in scholarly journal publisher have been conducted with human interaction. To deal with the problem of low reliability in the automatic metadata extraction and help with minimum human interaction, we propose semi-automated metadata extraction method based on rule-based method and machine learning method. In this experiment, we verified the performance under 26 different journals in Open Access Korea Central (OAK Central). We only cover two part (elements of front and back) as part of an effort to convert full-text xml based on JATS v1.0. As a result, our proposed method reached F1 = 94.1% in front and F1 = 92.5% in back.