Recommender systems for fossil community distribution modelling

We propose to leverage recommender systems from machine learning build large-scale community distribution models for the mammalian fossil record. Recommender are behind most online life today, shopping news personalisation, dating, or selection of study programmes fastest routes. Many work by predicting user preferences items that occur together in profiles. Technically, this setting closely resembles co-occurrence species natural environments. Here we frame modelling as a task, tailor existing techniques purpose and optimisation criteria fitting ecological context. The predictive power comes co-occurrences. demonstrate potential approach analysing past ecosystems on case Miocene sites Europe, where use proposed reconstructing companionships relative abundances large mammals. modelling, although not climatically explicit, can help reconstruct analyse their structure dynamics over time space. It also allows, even coarsely, predict presence–absence data. More generally, perspective is means analysis communities relationships between contexts.