Automatic search session detection exploiting results similarity from an external document collection (Détection automatique des sessions de recherche par similarité des résultats provenant d'une collection de documents externe) [in French]

Automatic search session detection exploiting results similarity from an external document collection Search engines users apply complex search behaviours such as query reformulation and multitasking search to satisfy their information needs. These search behaviours may be observed through query logs, and constitute clues allowing a better understanding of users’ needs. In this perspective, it is decisive to group queries related to the same information need into a unique search session. We propose an automatic session detection method exploiting the WIKIPEDIA documents collection, based on the similarity between the results returned for each query pair to estimate the similarity between queries. This method shows better performance than both temporal and lexical approaches traditionally used for successive session detection, and can be applied as well to multitasking search session detection. These experiments were conducted on a dataset originating from the OpenEdition Web portal. MOTS-CLÉS : Recherche d’information, détection automatique de sessions de recherche, analyse de journal de requêtes.