Soft textual cartography based on topic modeling and clustering of irregular, multivariate marked networks

Abstract

Soft textual cartography is an original approach aimed to study communities on spatially embedded and textually defined complex weighted networks. The present approach relies on the integration of topic modeling and soft clustering procedures. These two aspects can be combined using topic distances, and weighted unoriented networks representing the spatial configuration; their synergy is promising in topic interpretation and geographical information retrieval. This paper proposes an unified formalism, underlining the compatibility of the two aspects, as illustrated on the textual descriptions of the municipalities of the canton of Vaud, Switzerland. It also points to possible extensions and applications of the method, potentially useful for dealing with the ever growing amount of georeferenced textual content.