A taxonomy of representations in information retrieval system design

Abstract

In IR there is a growing body of empirical data and practical experience with various representations. Taxonomy may be defined as 'all the various activities involved in the construction of classificatory systems.' There are three stages in our research: identification, char acterization, and comparison/evaluation. Identification in volves an enumeration both of the categories of representations which could in principle be part of an IR system and of the different members in each category. Characterization involves identification of properties which could be used to characterize the members of each category. Comparison/evaluation in volves identification of measures which could be used in com paring the members within each category and/or in evaluating the impact of variations in representations on system perfor mance.