Citation Style Classification: a Comparison of Machine Learning Approaches

Citation style classification is a task that aims to detect a citation style according to which a given bibliographic entry is formatted. There are more than a hundred of recognized citation styles available, including popular ones such as ACM, IEEE, MLA, APA, and even more exotic ones. Automatic detection of citation style can be used in document linters, such as those intended for conference and term papers. Using automatic style classification enables people who assess articles in large quantities reduce cognitive load and increase efficiency. Apart from this, automatic detection of citation style can be used for reference parsing, topic classification, and reference extraction.In this paper we propose two novel approaches to citation style classification using both deep and classic machine learning methods. We evaluate them using a specially designed dataset, consisting of 6 million bibliographic records and spanning 91 citation styles. Our experiments showed that the proposed approaches significantly outperform the existing state-of-the-art solution, while also supporting almost five times more citation styles.