Multiple Clicks Model for Web Search of Multi-clickable Documents

Léa Laporte (1) , Sébastien Dejean (2) , Josiane Mothe (1, 3)

1 IRIT-SIG - Systèmes d’Informations Généralisées
2 IMT - Institut de Mathématiques de Toulouse UMR5219
3 UT2J - Université Toulouse - Jean Jaurès

This paper presents a novel document relevance model based on clickthrough information. Compared to the models from the literature we consider the case when documents can be clicked several times in a given search session. This case occurs more and more frequently, specifically for multi-clickable documents such as maps in location search enginess. Considering a real system query log, we evaluate our model and show that SVM can learn with fewer errors and with better MAP when the various types of clicks are considered in the model.

Mots clés: Information retrieval; Web search; Clickthrough information; Evaluation; Document relevance model.

Domaines: Recherche d'information [cs.IR]

Léa Laporte, Sébastien Dejean, Josiane Mothe. Multiple Clicks Model for Web Search of Multi-clickable Documents. International Conference on Enterprise Information Systems (ICEIS 2013), Jul 2013, Angers, France. ⟨hal-01259547⟩