View article

Words co-occurrence networks capture relationships between words co-occurring in a given document or phrase: each node is a word, and an edge exists if two words both appear in the same document. Co-occurrences networks have been used, among other things, to study the structure of human language [5], to detect influencial text segments [7], to identify authorship signature in temporal evolving networks [1]. Here we are interested in the analysis of the mathematical content of scientific articles, in order to develop measures for visualising mathematical concepts that can be used to classify scientific production in a novel way. The use of co-occurrence networks in scientometrics is not new, for example [10] focus on weighted co-occurrences networks of keywords to study the expansion of knowledge in nanoEHS risk literature. Our work differs as we extend our analysis a large set of mathematical concepts, and because we use a simplicial complex framework, as in [4], to have a better understanding of the clustering of concepts within and across different articles. Modeling co-occurrence relations as a simplicial complex allows to go beyond the network description that reduces all the structural properties to pairwise interaction and their combinations, and provides a natural framework for studying higher-order relations. This approach has proved to be very useful when the data has inherently rich structure, as in [9].