THE USE OF AUTOMATIC RELEVANCE FEEDBACK IN BOOLEAN RETRIEVAL SYSTEMS

A technique is described for automatic reformulation of boolean queries. Based on patron relevance judgements of an initial retrieval, prevalence measures are derived for terms appearing in the retrieved set of documents that reflect a term's distribution among the relevant and non‐relevant documents. These measures are then used to guide the construction of a boolean query for a subsequent retrieval. To illustrate the technique, a series of tests is described of its application to a small data base in an experimental environment. Results compare favourably with feedback as employed in a SMART‐type system. More extensive testing is suggested to validate the technique.