uruviel2001:
twitter.com/joelkuiper_/status/1255567049591570432covid "art" ... The way it’s constructed is by looking through all literature in the CORD19 corpus, then extracting mentions of interventions as predicted by a BiLSTM-CRF PICO classifier. These mentions are then mapped to medical concepts in RDF. To visualise the relations, the concepts are projected onto a 2-dimensional Poincaré manifold as described by “Poincaré Embeddings for Learning Hierarchical Representations” (arxiv.org/pdf/1705.08039.pdf) using the Gensim implementation rare-technologies.com/implementing-poincare-embeddings/