Description
Large language models (LLMs) have recently emerged as a novel tool for knowledge extraction and evidence synthesis in a wide range of scientific disciplines, including ecology and conservation science. Whilst many taxa have been classified based on traits or the provision of ecosystem functions, taxonomic and geographic coverage is sparse, and knowledge of interactions between species, particularly invertebrates, is limited. In this study, we use a pre-trained LLM to analyse abstracts of a corpus of published material on biological pest control to obtain a global database of pest herbivores and their natural enemies (DAPHNE). Our methodology resulted in the extraction of 179,001 species interactions from 112,830 publications, containing 17,219 unique animal taxa resolved at the species level. Interactions include herbivory (granivory, frugivory and gall-formation), predation, parasitism and hyperparasitism, and comprise a range of additional information, such as species' taxonomy, pest status, pest importance, natural enemy importance, provision of biocontrol, associated plants and industries, invasiveness and the vectoring of pathogens and diseases. Our study thoroughly evaluated the extracted information, and shows that most data columns were extracted with high recall and precision (>90%) as evaluated on a held-out test set. Comparison with other sources shows a large degree of overlap with DAPHNE, but also highlights that our dataset was able to capture important information that has been missed by other sources. DAPHNE is openly available and is hosted online as an interactive shiny app (https://ddx5w5-daan-scheepens.shinyapps.io/daphne-interaction-network/).
| Topic | Topic 4: Can knowledge graphs and AI-based synthesis improve ecological synthesis? |
|---|