“…For example, after training on a large corpus of English language documents, given vectors representing words that are countries and capitals, Madrid→−Spain→+France→ will result in a vector that is similar to Paris→true→, more than other vectors in the corpus (Mikolov et al , 2013). This type of representation has led to better performance in downstream classification problems, including in biomedical literature classification (Chen et al , 2018; Minarro-Giménez et al , 2014), annotations (Duong et al , 2018; Zwierzyna and Overington, 2017) and genomic sequence classifications (Dutta et al , 2018; Du et al , 2018; Mejia Guerra and Buckler, 2017; Zhang et al , 2018).…”