Index Expression Belief Networks for Information Disclosure

It is widely accepted that to extend the eeectiveness of information disclosure beyond the limitations of current empirically-based retrieval systems, some notion of document semantics has to be incorporated into the retrieval mechanism. A recent approach to bringing semantics into play is to found the retrieval mechanism on the notion of logical inference. In this paper, we build on this approach and describe a promising new mechanism for information disclosure, called the Reenement Machine. The Reenement Machine features the language of index expressions as a language for characterizing information objects and a deduction mechanism driven by rules of inference. Two types of inference rule are distinguished. The rules of strict inference follow the line of traditional logical deduction. As the characterizations of objects are incomplete and requests are typically partial descriptions of the information need, the rules of strict inference are supplemented with a rule of plausible inference. This rule of plausible inference is motivated by recent work in the area of plausible reasoning in knowledge-based systems and, in particular, is derived from the work on belief networks. Besides giving details of the Reenement Machine, this paper also presents some preliminary experimental results.