Reference Metadata Extraction from Korean Research Papers

摘要

A large amount of research papers are published in various fields and the ability to accurately extract metadata from a list of references is becoming increasingly important. Moreover, metadata extraction is crucial for measuring the influence of a particular study or researcher. However, it is difficult to automatically extract data from most lists of references because they consist of unstructured strings with bibliographies structured in various formats depending on the proceedings. Thus, this paper presents an effective and accurate method for extracting metadata, such as author name, title, publication year, volume, issue, page numbers, and journal name from heterogeneous references using the conditional random fields model. To conduct an experiment measuring the effectiveness of the proposed model, 1,415 references from 93 different academic papers published in Korea were used and a high accuracy of 97.10% was obtained.