In order to train the reference extraction model the training data should be prepared considering the following guideline.

For each reference found in the document, it should be enclosed in <ref> & </ref> tags.

Example. (the following snippet is from the complete pdf to text document.)

<ref> Taggert, Paul; Szczerbiak, Aleks (2004): Contemporary Euroscepticism in the Party Systems of the European Union Candidate States of Central and Eastern Europe. European Journal of Political Research, 43, 1, S. 1-27.</ref>

<ref>Tömmel, Ingeborg (2006): Die Reform der Strukturpolitik der EU – Eine Reform europäischer Governance? In: Kleinfeld, Ralf; Plamper, Harald; Huber, Andreas (Hrsg.): Regional Governance. Steuerung, Koordination und Kommunikation in regionalen Netzwerken als neue Formen des Regierens. Göttingen: V&R unipress. S. 183-200.</ref>

Additionally the layout information of the PDF document is also required. This can be obtained using the Cermine api which generates an .CSV file containing the co-ordinates for each lines. Add link to the RefExt and a code snippet. For each line found in the PDF document the co-ordinates value for five different columns is generated. They are as follows

• txt: the text of each line.
• hsl: the horizontal space line.
• vsl: the vertical space linw within the page.
• vw: the vertical width of the line.
• hl: the horizontal length of the line.
• pn: the number of the paragraph.

Please refer to the image for better understanding of the layout information.