Une approche de recherche d'attributs pertinents pour l'agrégation d'information

Given a query, aggregated search aims at aggregating information nuggets retrieved from differents sources. The aggregated result is a coherent answer that can be seen as an alternative to the traditional ranked list of documents. We focus in this paper on relational aggregated search where information nuggets are mainly classes, instances and attributes. We present a result aggregation approach to answer to class queries. Our approach is based on relevant attributes detection and uses tables to present results. It is composed of three steps: instances and attributes selection, filtering and attributes ranking. In order to evaluate our approach we assess the quality of tables constructed from DBpedia. The evaluation shows that attributes scoring depends on the selected instances, and that the weighting function can affect significantly the quality of tabular results. MOTS-CLÉS : Recherche d’information, Agrégation de résultats , Recherche d’attributs