The inverted file index common to many full-text information retrieval systems presents unusual and challenging data management requirements. These requirements are usually met with custom data management software. Rather than build this custom software, we would prefer to use an existing database management system. Attempts to do this with traditional (e.g., relational) database management systems have produced discouraging results. Instead, we have used a persistent object store, Mneme, to support the inverted file of a full-text information retrieval system, INQUERY. The result is an improvement in performance along with opportunities for INQUERY to take advantage of the standard data management services provided by Mneme. We describe our implementation, present performance results on a variety of document collections, and discuss the advantages of using a persistent object store to support information retrieval. This work is supported by the NSF Center for Intelligent Information Retrieval at the University of Massachusetts.
Available at: http://works.bepress.com/eliot_moss/50/