A Supervised Approach for Detecting Allusive Bibliographical References in Scholarly Publications

Exploiting the links between content is crucial in recommendation approaches. In the case of a scientific article library, bibliographic references serve as a major link source. Among them, some are explicit references as we can find at the end of articles or books, while other references are scattered in the text or in the footnotes, according to a more or less strong implicit degree. We propose to focus on the detection of this type of references that we call allusive, in scientific articles from the field of Human and Social Sciences. To overcome the inherent difficulties raised by such reference detection, we present a method which aims at (i) identifying paragraphs that contain references via a classification process and (ii) at applying CCRFs (Cascaded Conditional Random Field) in order to detect more accurately the bibliographic entries and consequently annotate their contents.