Deciding Term Aboutness Probabilistically

Information retrieval is the quest to find those information objects relevant to a given information need. Relevance is a difficult notion to define operationally. As a consequence information retrieval mechanisms are typically driven by the decision of when one information carrier (e.g. a document) is about another (e.g. a query). As documents and queries are typically complex representations built up from so called terms, the aboutness decision relies on determining whether one term is about another. This article addresses the question of term aboutness within a framework provided by so called Information Containment Belief Networks. The latter are a computationally efficient subclass of belief network motivated from an underlying algebraic theory of information. Various properties of Information Containment Belief Networks are presented, with particular emphasis on those properties dealing with their computational efficiency. Some probabilistic strategies for computing term aboutnes..