PhD (2009-2013)#
Metrology of terrain networks, application to lexical resources enrichment and to information retrieval
(original French title: Métrologie des graphes de terrain, application à la construction de ressources lexicales et à la recherche d’information)
Prepared at IRIT and defended on 4 November 2013 at INP Toulouse.
Supervisors: Bruno Gaume and Henri Prade
Committee:
- Clémence Magnien (Université Pierre et Marie Curie)
- Pierre Fraigniaud (Université Paris 7)
- Pierre Zweigenbaum (LIMSI)
- Éric Gaussier (Université Grenoble 1)
- Frédéric Amblard (Université Toulouse 1)
Download: manuscript · defense slides · bibtex
External links: HAL (TEL) · OATAO (INPT)
Abstract#
Terrain networks (also called real-world or complex networks) are graphs built from real data: social networks, the internet, protein interactions, links between documents, lexical networks… Despite their varied origins, they share common properties: globally sparse but locally very dense, very short paths between vertices, and a highly heterogeneous degree distribution.
The thesis addresses three problems on these graphs: measuring the proximity between two vertices, comparing graphs that share the same vertices, and partitioning a bipartite graph. For proximity, it introduces a new measure, confluence, based on short-time random walks.
This work is applied to the semi-automatic enrichment of lexical resources (Wisigoth, for Wiktionary) and to the automatic organisation of search results (Kodex).