This paper presents an information retrieval model based on the Bayesian network formalism. The topology of the network (representing the dependence relationships between terms and documents) as well as the quantitative knowledge (the probabilities encoding the strength of these relationships) will be mined from the document collection using automatic learning algorithms. The relevance of a document to a given query is obtained by means of an inference process through a complex network of

more »

... ences. A new inference technique, called propagation + evaluation, has been developed in order to obtain the exact probabilities of relevance in the whole network efficiently.