La presente invention concerne des systemes, des procedes et des supports de stockage lisibles par ordinateur permettant d'ajuster les caracteristiques de la presentation d'une page de resultats d'un moteur de recherche (SERP) par un dispositif client sur la base d'une intention d'interrogation d'un utilisateur. Le client peut reacheminer un prefixe de recherche vers un service de recherche et, en reponse, recevoir une ou plusieurs suggestions d'interrogation et un ou plusieurs algorithmes d'apprentissage automatique concus pour chaque suggestion d'interrogation. L'utilisateur execute une interrogation de recherche contenant l'intention d'interrogation en selectionnant une des suggestions d'interrogation. Le dispositif client calcule des scores pour chaque groupe de resultats en utilisant l'algorithme d'apprentissage automatique concu pour la suggestion d'interrogation selectionnee. Au moins une caracteristique de presentation de la SERP est ajustee de telle maniere qu'au moins un groupe de resultats est mis en valeur par rapport a un autre sur la base des scores du groupe de resultats respectifs representant une pertinence par rapport a l'intention d'interrogation.

In information retrieval, relevance of documents with respect to queries is usually judged by humans, and used in evaluation and/or learning of ranking functions. Previous work has shown that certain level of noise in relevance judgments has little effect on evaluation, especially for comparison purposes. Recently learning to rank has become one of the major means to create ranking models in which the models are automatically learned from the data derived from a large number of relevance judgments. As far as we know, there was no previous work about quality of training data for learning to rank, and this paper tries to study the issue. Specifically, we address three problems. Firstly, we show that the quality of training data labeled by humans has critical impact on the performance of learning to rank algorithms. Secondly, we propose detecting relevance judgment errors using click-through data accumulated at a search engine. Two discriminative models, referred to as sequential dependency model and full dependency model, are proposed to make the detection. Both models consider the conditional dependency of relevance labels and thus are more powerful than the conditionally independent model previously proposed for other tasks. Finally, we verify that using training data in which the errors are detected and corrected by our method, we can improve the performance of learning to rank algorithms.