Salvato in:
| Autore principale: | |
|---|---|
| Natura: | Recurso digital |
| Lingua: | |
| Pubblicazione: |
Zenodo
2025
|
| Accesso online: | https://doi.org/10.5281/zenodo.17961963 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
Sommario:
- <p>The present research focuses on the two most promising tools for bibliographic reference extraction<br>and parsing, i.e. Anystyle (Keil, 2023) and GROBID (Lopez, 2023a), and another tool, OUTCITE.1<br>The first two tools have been selected starting from a previous study (Cioffi and Peroni, 2022), in<br>which the available reference extraction tools were analysed to understand their performances against<br>a corpus of 56 PDF articles (the previous Gold Standard) published in 27 subject areas (Computer<br>Science, Arts and Humanities, Mathematics, etc.). The third tool, OUTCITE, is an extension of<br>another tool tested in the previous research, EXCITE. As a first step, a series of tests have been<br>conducted to check if the two software have been recently improved. Keeping in mind the differences<br>emerged during this phase, the code used for testing and comparing the reference extraction tools,<br>from the previous study, has been extended and modified to make it available also for others to be<br>reused for similar analysis (Pagnotta, 2024a). Another important step has been the creation of a new<br>Gold Standard, composed of 112 PDF documents (an extension of the prior one) together with the<br>training of these tools against part of the Gold Standard. A second testing phase has followed, in<br>which the trained versions of the tools have been evaluated against the remaining part of the Gold<br>Standard (different configurations have been tested) to check if the training was successful and<br>improved the performances. One specific training configuration has given promising results and has<br>been used to compare the performances of the original version of GROBID (i.e. out of the box), and<br>the latest version of OUTCITE at the time of the experiments, i.e. October 2023 (Pagnotta, 2023).<br>The final step of the research work has been to develop a reference extraction service (Pagnotta and<br>Paolini, 2024) – after having chosen the tool which outperformed the others after the training - which<br>enables a user to provide a PDF of a scholarly article as input and to have, in return, citation data and<br>bibliographic metadata from all the references that are cited by the given article in a format that<br>enables their ingestion in OpenCitations (Peroni & Shotton, 2020).</p>