Improved bibliographic reference parsing based on repeated patterns

Parsing details like author names and titles out of bibliographic references of scientific publications is an important issue that has received considerable attention recently. However, most existing techniques are tailored to the highly standardized reference styles used in the last two to three decades. They do not perform well with the wide variety of reference styles used in older, historic publications. Thus, they are of limited use when creating comprehensive bibliographies covering both historic and contemporary scientific publications. This paper presents a generic approach to bibliographic reference parsing, named RefParse, which is independent of any specific reference style. Its core feature is an inference mechanism that exploits the regularities inherent in any list of references to deduce its format. In addition, our approach learns names of authors, journals, and publishers to increase the accuracy in scenarios where human users double check parsing results to increase data quality. Our evaluation shows that our approach performs comparably to existing ones with contemporary reference lists and also works well with older ones.