Ordinal regression based model for personalized information retrieval

Abstract

Retrieving relevant items as a response to a user query is the aim of each information retrieval system. But 'without an understanding of what relevance means to users, it is difficult to imagine how a system can retrieve relevant information for users' [1]. In this paper, we try to capture what relevance is for a particular user and model his profile implicitly considering his non declared preferences that are inferred from a ranking of a reduced set of retrieved documents that he produces. We propose an ordinal regression based model for interactive ranking which uses both the information given by this subjective ranking, as well as the multicriteria evaluation of these ranked documents, to adjust optimally the parameters of a ranking model. This model consists of a set of additive value functions which are built so as they are as compatible as possible with the subjective ranking. The preference information used in our model requires reasonable cognitive effort from the user. © 2009 Springer Berlin Heidelberg.