Information Storage and Retrieval. Document Retrieval Based on Clustered Files.

摘要

A retrieval system is considered in which document descriptions are stored and accessed in groups called clusters. All items in a cluster meet common similarity criteria and are represented by a composite entity called a profile. Among the problems investigated in clustered file organizations are profile definition,updating,hierarchy storage,and secondary profile uses. A comparison with an inverted file is also included. Most of the work is experimental in nature and uses the SMART information retrieval system. The report covers basic and advanced concepts in document retrieval,file organization,and clustered files. It includes a new evaluation scheme for cluster searches based on the 'precision floor'and 'recall ceiling'measures, a description of the preparation of unbiased,economical profiles, a study of several profile types (standard,rank value,frequency rank,shortened) and weighting schemes (none,partial,full) and descriptions of several experiments designed to improve the construction and use of clustered files. (Author)