Document Ranking Methodology for Conventional Boolean Retrieval Systems

Abstract

One of the most desirable features of a document retrieval system is that the items in the system’s output are ranked in descending order by their likelihood of relevance. In this way, the system’s user can be presented with a manageable set of those documents in the collection which are most likely to be useful in satisfying his information needs. However, commercially available retrieval systems are usually based on conventional set theory and Boolean logic, and thus they generally do not have sound ranking capabilities. In other words, a document retrieval system founded on a typical Boolean search strategy provides the user with a number of documents listed in an arbitrary order, i. e., no indication of document relevance is offered. Since in operational situations such unranked output is often very large, especially when the query is a broad one, a great burden may be imposed on the user

The objective ofthis paper is to discuss the results of the author’s recent investigation into rigorous methods for incorporating a weighting mechanism into conventional Boolean retrieval systems in order to provide lists of ranked output documents. This has been achieved by developing a formal theory linking Boolean retrieval procedures applicable in the case of queries as well as documents represented by simply by sets of unweighted index terms. Implementation of the resulting document ranking achemes would not require any changes in the principles upon which operational Boolean retrieval systems are based; some additional software is all that would be needed. This indicates the great potential of these ranking schemes for significantly improving the effectiveness of conventional document retrieval systems.