Figure 1

Construction of the knowledge space. a. We use the metadata from 1.45 million articles posted on the arXiv, corresponding to the article field tags, authors, and timestamp. We build a high-dimensional 175 space where each article is uniquely mapped through field tags corresponding to orthogonal dimensions. This high-dimensional space is finally embedded within a 2-dimensional knowledge space using the tSNE algorithm. Each point represents an article. Colors correspond to major academic fields in arXiv based on the first tag (i.e primary field) of the articles. b. Confusion matrix resulting from a k-Nearest Neighbors prediction of a location’s major field tag in the knowledge space based on the major field tags of its \(k=10\) nearest neighbors. We use 70% of the data for training and the remaining 30% for testing. The values show the proportion of predictions in the test set falling within a given category (columns) given the observed major field (rows). “Electrical Engineering” is short for “Electrical Engineering and Systems Science”