Communities of related terms in Karst terminology co-occurrence network

Karst science is an attractive field of interdisciplinary research with rich terminology. This study was performed as part of a project aiming at developing novel approaches to terminology extraction and visualization, in line with the understanding of knowledge, as represented in texts, as conceptually dynamic and linguistically varied. The aim of this paper is to investigate how powerful graph-based methods can be used for visualizing and analysing domain terminology. In order to detect communities in karst terminology, we analyse the frequently cooccurring karst terms in a scientific corpus of karstologic literature. The most frequent cooccurrence pairs, which included ten or more co-occurrences within the whole corpus, are delivered as input to the Louvain community detection algorithm and visualized as a domain graph. The resulting data was evaluated by domain experts who found that the detected term groups are meaningful and correspond to different types of karst phenomena. The results are further discussed in relation to more standard topic modelling approaches, using Latent Dirichlet Allocation and Non-negative Matrix Factorization algorithms.