Automating the Curation Process of Historical Literature on Marine Biodiversity Using Text Mining: The DECO Workflow

Abstract

Historical biodiversity documents comprise an important link to the long-term data life cycle and provide useful insights on several aspects of biodiversity research and management. However, because of their historical context, they present specific challenges, primarily time- and effort-consuming in data curation. The data rescue process requires a multidisciplinary effort involving four tasks: (a) Document digitisation (b) Transcription, which involves text recognition and correction, and (c) Information Extraction, which is performed using text mining tools and involves the entity identification, their normalisation and their co-mentions in text. Finally, the extracted data go through (d) Publication to a data repository in a standardised format. Each of these tasks requires a dedicated multistep methodology with standards and procedures. During the past 8 years, Information Extraction (IE) tools have undergone remarkable advances, which created a landscape of various tools with distinct capabilities specific to biodiversity data. These tools recognise entities in text such as taxon names, localities, phenotypic traits and thus automate, accelerate and facilitate the curation process. Furthermore, they assist the normalisation and mapping of entities to specific identifiers. This work focuses on the IE step (c) from the marine historical biodiversity data perspective. It orchestrates IE tools and provides the curators with a unified view of the methodology; as a result the documentation of the strengths, limitations and dependencies of several tools was drafted. Additionally, the classification of tools into Graphical User Interface (web and standalone) applications and Command Line Interface ones enables the data curators to select the most suitable tool for their needs, according to their specific features. In addition, the high volume of already digitised marine documents that await curation is amassed and a demonstration of the methodology, with a new scalable, extendable and containerised tool, “DECO” (bioDivErsity data Curation programming wOrkflow) is presented. DECO’s usage will provide a solid basis for future curation initiatives and an augmented degree of reliability towards high value data products that allow for the connection between the past and the present, in marine biodiversity research.

Unsupervised literature mining approaches for extracting relationships pertaining to habitats and reproductive conditions of plant species

“…The growing volume of scientific literature on biodiversity has led to a focus on the development of computational methods for extracting meaningful information from unstructured textual data (Farrell et al, 2022 ; Paragkamian et al, 2022 ). This computational task is known as text mining, and it has been used to identify trends, patterns, and relationships that would otherwise be difficult to detect.…”

“…Information extraction (IE) is an umbrella term for tasks that seek to automatically extract structured information from unstructured text. With the exponential growth of digitized literature over the years, IE has become increasingly pertinent, due to its role in (semi-)automatically populating databases with content (Ravikumar et al, 2015 ; Lee et al, 2018 ; Paragkamian et al, 2022 ). Relation extraction (RE) is an IE task that is concerned with the identification of semantic relationships between entities or concepts in text.…”