The problem of the mathematical framework for the documents-to-queries matching portion of an information retrieval system is examined. Assuming some universe in which probabilities can be computed, the fundamental notion is that the degree of support for a query of provided by a document is . Through a series of examples, the authors then show that this probabilistic inference model is similar to other models for information retrieval (Boolean, fuzzy-set, vector space, and so on).

Although this paper is interesting and easy to read, what is most striking about it is the discrepancy between the broad claim that this method “provides a common conceptual and mathematical basis” for other retrieval models, which are called “special cases,” and the many caveats that appear when the individual models are examined. For example, the authors show that different variations in the new model lead to approximations to the vector space model under a given similarity measure, but some well-known similarity measures (like the cosine measure) cannot be interpreted thus. Similarly, the new model is the same as the standard Boolean model or the fuzzy-set model, provided one views the latter two in a particularly narrow way.