Multiscale Visual Analysis of Lexical Networks

A lexical network is a very useful resource for natural language processing systems. However, building high quality lexical networks is a complex task. "Jeux de mots'' is a web game which aims at building a lexical network for the French language. At the time of this paper's writing, "jeux de mots'' contains 164 480 lexical terms and 397 362 associations. Both lexical terms and associations are weighted with a metric that determines the importance of a given term or association. Associations between lexical terms are typed. The network grows as new games are played. The analysis of such a lexical network is challenging.The aim of our work is to propose a multi-scale interactive visualization of the network to facilitate its analysis. Our work builds on previous work in multi-scale visualization of graphs. Our main contribution in this domain includes (1) the automatic computation of compound graphs, (2) the proximity measure used to compute compound nodes, and (3) the computation of the containment relation used to exhibit the dense relation between one important node and a set of related nodes.