On models of information retrieval processes

摘要

A model called the binary independence model is presented as a generalization of a few models which have applied to the analyses of a clustered search process, some indexing strategies and a relevance feedback process. This model, together with the Swets model, the linked-2-Poisson model, the 2-Poisson model, the binary limited-dependence model, the tree dependence model, the binary dependence model and the non-binary independent model are compared and contrasted. Despite the fact that these models are intended for different applications, three aspects of modelling are identified, based on which, these models are compared. The three aspects are the class concept, the distribution of similarities and the relation between a matching function and its retrieval effectiveness. As results of the comparison between the models, more insight into the models is gained and a set of guidelines are suggested to help a system designer to choose a model.