A non-classical logic for information retrieval

Implicit in many information retrieval models is a logic. These logics are hardly ever formalised. This paper formalises a non-classical logic underlying information retrieval. It shows how a particular conditional logic is the ‘right’ logic to do Information Retrieval. Its relationship to existing retrieval mechanisms is investigated. The semantics of the logic are expressed in probability theory, and evaluated through a possible-world analysis, thus establishing an intensional logic. In doing so, we motivate a new principle, the logical uncertainty principle, which gives a measure of the uncertainty associated with an inference.