The necessity for adaptation in modified boolean document retrieval systems

A document retrieval system may be described by three formal characteristics: the syntax employed to describe documents (keywords or vectors of weights, for instance), the form of machine-processable queries it accepts as valid (unordered sets of keywords, keywords with Boolean connectives or weighted vectors, for example), and the retrieval rules used to rank or retrieve documents. This article argues that the interdependence among document descriptions, queries, and retrieval rules requires adaptation for the system to perform effectively when one of its components changes. Recently, suggestions have been made to modify traditional Boolean document retrieval systems to allow more flexible queries and ranked document output. However, these new forms of queries and retrieval rules likely require that documents be described differently than they are in existing, commercial Boolean retrieval systems. A “genetic algorithm” is discussed as a means for redescribing documents. This probabilistic algorithm uses feedback along with alternative descriptions of a single document and takes account of the dependency structure of subject terms.