PDFdigest: an Adaptable Layout-Aware PDF-to-XML Textual Content Extractor for Scientific Articles

PDFdigest: an Adaptable Layout-Aware PDF-to-XML Textual Content Extractor for Scientific Articles Daniel Ferr´es∗, Horacio Saggion∗, Francesco Ronzano†, and `Alex Bravo∗ ∗ Large Scale Text Understanding Systems Lab TALN Research Group, DTIC Department Universitat Pompeu Fabra, Barcelona (Spain) {daniel.ferres,horacio.saggion,alex.bravo}@upf.edu † Integrative Biomedical Informatics Group (IBI) Research Programme on Biomedical Informatics (GRIB) IMIM-UPF, Barcelona (Spain) francesco.ronzano@upf.edu Abstract The availability of automated approaches and tools to extract structured textual content from PDF articles is essential to enable scientific text mining. This paper describes and evaluates the PDFdigest tool, a PDF-to-XML textual content extraction system specially designed to extract scientific articles’ headings and logical structure (title, authors, abstract,...) and its textual content. The extractor deals with both text-based and image-based PDF articles using custom rule-based algorithms implemented with existing state-of-the-art open-source tools for both PDF-to-HTML conversion and image-based PDF Optical Character Recognition. Keywords: Language Resources, Scientific Text Mining, Digital Libraries, Information Extraction, PDF Conversion 1. Introduction Nowadays automated approaches to mine scientific litera- ture are essential to support researchers - as well as any other interested actor - in taking full advantage of the huge number of articles available online: (Ware and Mabe, 2015) estimate that more than 2.5 million papers are published on the Web every year, and the percentage of publications distributed as Open Access content is constantly increas- ing. Even if XML-based formats are emerging, the major- ity of scientific articles are still accessed as PDF files. As a consequence, effective tools to extract structured textual content from PDF files represent a key technology to en- able scientific text mining (Ronzano and Saggion, 2016). Such tools are essential components to develop a varied range of applications in different academic and industrial contexts, thus supporting the implementation of new, intel- ligent patterns to access scientific information (Accuosto et al., 2017). Once textual contents are extracted from scien- tific publications, they can be mined and enriched using ex- isting Natural Language Processing (NLP) techniques such as Named Entity Recognition and Classification, Parsing, Relation Extraction, among others, thus helping researchers to improve their access to the scientific knowledge. In this paper we present PDFdigest, a tool that extracts structured textual contents from scientific articles in PDF format. PDFdigest performs the conversion of files from PDF to XML format while preserving both the textual con- tent and the layout details of the input PDF document. The contributions of this paper can be summarized as fol- lows: • PDFdigest, a PDF to XML system that extracts the content of both text-based and image-based PDF files1 by: 1Web-service: http://taln.upf.edu/pdfdigest – dealing with specific paper layouts as well as with text in multiple languages; – storing the layout of the original PDF in an HTML file, thus enabling the possibility of Vi- sual Analytics over the original PDF document; – implementing a customizable approach to adapt the content extraction process to different PDF paper layouts by manually changing the extrac- tion offsets and thresholds. • An evaluation of the effectiveness of the text-based and image-based PDF textual content …

Content truncated to 500 words