Matching Heterogeneous Textual Data Using Spatial Features

Mapping Heterogeneous Textual Data: A Multidimensional Approach Based on Spatiality and Theme

“…To compare documents through the spatial dimension, a common representation is required. To this end, in previous work, we defined the STR [6], or Spatial Textual Representation, a graph structure generated from text features. A STR is composed of two components: spatial entities (i.e., vertices) and spatial relations (i.e., edges).…”

“…As geographical index, we adopt Geodict 3 [7], generated from three sources: Wikidata, OpenStreetMap and Geonames. In [6], we conducted experiments with graph matching algorithms [5] to match spatial features integrated into the STR. We defined three STR transformations (two STR Abstractions that abstract spatial entities in the STR -Uniform Abstraction and Bounded Abstraction and one STR Extension) and evaluate the associated matching quality through four criteria (See Section 2.2).…”

“…We thus propose a multidimensional mapping approach for heterogeneous textual data that exploits firstly the spatial dimension and secondly the thematic dimension. For the spatial mapping of textual data, we have already proposed a representation of document: the Spatial Textual Representation (STR ) [6]. STR is a graphical representation, composed of connected spatial entities (i.e.…”

Exploitation de l’hétérogénéité dans les données textuelles

“…Pour mesurer la similarité entre les STRs, nous utilisons des algorithmes de graph matching (Fize et al, 2018). On distingue deux grandes catégories d'algorithmes : structure-based et pattern-based.…”

“…Tableau 3.1. Couple de mesure (Fize et al, 2018) Cependant, on obtient une valeur de précision moyenne sur les critères ESP ou ESCC dû à leur rareté dans les correspondances. Parmi les combinaisons dominantes, on souligne que l'utilisation de versions transformées de la STR, ici l'extension, améliore globalement la précision du système.…”