Search

With the accelerating development of science and technology, the academic papers being published in various fields are increasing rapidly. Academic papers specially in science and technology fields are a crucial media for researchers who develop new technologies by identifying knowledge regarding the latest technological trends and conduct derivative studies in science and technology. Therefore, the continual collection of extensive academic papers, structuring of metadata, and construction of databases are significant tasks. However, research on automatic metadata extraction from Korean papers is not being actively conducted currently owing to insufficient Korean training data. We automatically constructed the largest labeled corpus in South Korea to date from 315,320 PDF papers belonging to 503 Korean academic journals and this labeled corpus can be used for training the models of automatic extraction for 12 metadata types from PDF papers. This labeled corpus is available at https://doi.org/10.23057/48. Moreover, we developed inspection process and guidelines for the automatically constructed data and performed a full inspection of the validation and testing data. The reliability of the inspected data was verified through the inter-annotator agreement measurement. Using our corpus, we trained and evaluated the BERT based transfer learning model to verify its reliability. Furthermore, we proposed new training methods that can improve the metadata extraction performance of Korean papers, and through these methods, we developed KorSciBERT-ME-J and KorSciBERT-ME-J+C models. The KorSciBERT-ME-J showed the highest performance with an F1 score of 99.36%, as well as robust performance in automatic metadata extraction from Korean academic papers in various formats.

Bibliographic references containing citation information of academic literature play an important role as a medium connecting earlier and recent studies. As references contain machine-readable metadata such as author name, title, or publication year, they have been widely used in the field of citation information services including search services for scholarly information and research trend analysis. Many institutions around the world manually extract and continuously accumulate reference metadata to provide various scholarly services. However, manually collection of reference metadata every year continues to be a burden because of the associated cost and time consumption. With the accumulation of a large volume of academic literature, several tools, including GROBID and CERMINE, that automatically extract reference metadata have been released. However, these tools have some limitations. For example, they are only applicable to references written in English, the types of extractable metadata are limited for each tool, and the performance of the tools is insufficient to replace the manual extraction of reference metadata. Therefore, in this study, we focused on constructing a high-quality corpus to automatically extract metadata from multilingual journal article references. Using our constructed corpus, we trained and evaluated a BERT-based transfer-learning model. Furthermore, we compared the performance of the BERT-based model with that of the existing model, GROBID. Currently, our corpus contains 3,815,987 multilingual references, mainly in English and Korean, with labels for 13 different metadata types. According to our experiment, the BERT-based model trained using our corpus showed excellent performance in extracting metadata not only from journal references written in English but also in other languages, particularly Korean. This corpus is available at http://doi.org/10.23057/47.