Link prediction in ecological meta‐networks under extreme taxonomic bias

Abstract Ecological networks offer powerful insights into community function but, without first characterizing these networks accurately, our ability to detect and interpret changes under environmental stress is limited. We develop an extension of the Covariate‐Informed Link Prediction (COIL) framework to reduce bias in ecological link prediction when interaction data are derived from studies focused on a small number of species. Because the absence of an observed interaction is only informative when the species involved actually co‐occur, uncertainty in species occurrence can strongly influence how non‐interactions are interpreted in meta‐network datasets. Our extended model (COIL+) employs a latent factor structure that borrows information across species, incorporates species traits and phylogeny, and integrates observations from multiple studies to account for uncertainty in species occurrence under extreme taxonomic bias (i.e. studies focussing on only a small subset of species). We additionally introduce a trait‐matching procedure that quantifies how the influence of species traits on interaction probability varies across partner species. We illustrate the use of the model with a literature‐based dataset of 268 sources reporting Afrotropical frugivory and compare performance with and without correction for occurrence uncertainty. COIL+ substantially improves link prediction by increasing out‐of‐sample discrimination and reduces sampling bias, revealing 5637 likely but unobserved frugivory interactions (a median of nine additional interactions per frugivore). Newly predicted interactions are concentrated among poorly sampled frugivores, such as the water chevrotain ( Hyemoschus aquaticus , a small forest‐dwelling ungulate) and the rufous‐bellied helmetshrike ( Prionops rufiventris , a passerine bird of East African tropical forests). Additionally, the method better distinguishes true interactions from non‐interactions compared to existing approaches under strong taxonomic bias and narrow study focus. This framework generalizes to diverse meta‐network contexts and provides a useful tool for link prediction in the face of biased interaction data.