RECODE - Relational Ecological COrpus for Data Extraction

Area of Science:

• Ecology
• Conservation Biology
• Taxonomy

Background:

• Ecological research relies heavily on species location and trait data.
• This crucial data is often unstructured text in publications, especially for invertebrates.
• Automated data extraction using AI is challenging due to the need for labeled training corpora.

Purpose of the Study:

• To introduce RECODE, a manually annotated corpus of ecological and taxonomic literature.
• To facilitate the training and fine-tuning of AI models for automated data extraction.
• To address the lack of standard datasets for ecological and taxonomic text mining.

Main Methods:

• Manual annotation of ecological and taxonomic literature by subject matter experts.
• Validation of annotations by experts familiar with the traits of spiders and insects.
• Development of a corpus for training AI models in named entity recognition and relation extraction.

Main Results:

• Creation of RECODE, a manually annotated corpus of scientific literature.
• The corpus contains validated occurrence and trait data for spiders and insects.
• RECODE provides a standardized dataset for AI model development in ecology.

Conclusions:

• RECODE is a valuable resource for advancing automated data extraction in ecology and conservation.
• The availability of such annotated corpora is essential for improving AI model performance.
• This work supports the development of more efficient methods for ecological data mining.