NLP and Graph embeddings: TextGraphs.jl

Abstract

Natural language processing (NLP) techniques analyze formal properties of speechin humans. One way of analyzing language is through graph embeddings. Graphs are cunningly simple abstractions, which have vertices and nodes as basic elements. Nevertheless, graphs were revealed to be very useful abstractions to describe associations in real-world phenomena whenever there are coupled elements. Graph embeddings associate words with nodes and adjacency in a phrase with links (edges). When words are repeated, the respective node is linked twice. Some attributes of these “speech graphs” are correlated with mental states, such as connectivity (e.g. largest connected components).The first section of this article reviews theoretical basis for the analysis of text-based graph embedding (”Text Graphs” or ”Speech Graphs”). We present standard procedures of text embedding and discuss graph properties (e.g. connected components) in this context.Some features brought by TextGraphs.jl (e.g. density, centrality) were not used before in literature. The second section covers applications, including installation, basic features, and vignettes with example cases