Daniel at the FinSBD-2 Task: Extracting List and Sentence Boundaries from PDF Documents, a model-driven approach to PDF document analysis

Emmanuel Giguet (1, 2, 3, 4, 5) , Gaël Lejeune (6, 7)

1 Equipe Hultech - Laboratoire GREYC - UMR6072
2 GREYC - Groupe de Recherche en Informatique, Image et Instrumentation de Caen
3 UNICAEN - Université de Caen Normandie
4 CNRS - Centre National de la Recherche Scientifique
5 NU - Normandie Université
6 STIH - Sens, Texte, Informatique, Histoire
7 STIH-LC - Équipe Linguistique computationnelle

In this paper, we present the method we have designed and implemented for identifying lists and sentences in PDF documents while participating to FinSBD-2 Financial Document Analysis SharedTask. We propose a model-driven approach for the French and English datasets. It relies on a top-down process from the PDF itself in order to keep control of the workflow. Our objective is to use PDF structure extraction to improve text segment boundaries detection in an end-to-end fashion.

Domaines: Informatique [cs]; Traitement du texte et du document

Emmanuel Giguet, Gaël Lejeune. Daniel at the FinSBD-2 Task: Extracting List and Sentence Boundaries from PDF Documents, a model-driven approach to PDF document analysis: Proceedings of the Second Workshop on Financial Technology and Natural Language Processing. https://aclanthology.org/2020.finnlp-1.11/, pp.67-74, 2021, Proceedings of the Second Workshop on Financial Technology and Natural Language Processing. ⟨hal-02927304⟩