Semantic relation clustering for unsupervised information extraction (Regroupement sémantique de relations pour l'extraction d'information non supervisée) [in French]

Semantic relation clustering for unsupervised information extraction Most studies in unsupervised information extraction concentrate on the relation extraction and few work has been proposed on the organization of the extracted relations. We present in this paper a two-step clustering procedure to group semantically equivalent relations : a first step clusters relations with similar expressions while a second step groups these first clusters into larger semantic clusters, using different semantic similarities. Our experiments show the stability of distributional similarities over WordNet-based similarities for semantic clustering. We also demonstrate that the use of a multi-level clustering not only reduces the calculations from all relation pairs to basic clusters pairs, but it also improves the clustering results. MOTS-CLÉS : Extraction d’Information Non Supervisée, Similarité Sémantique, Clustering.