ISR11 A Modified Two-Level Search Algorithm Using Request Clustering chapter V. R. Lesser Harvard University Gerard Salton Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government. VII-7 [OCRerr]. Desib[OCRerr] of an Experiment to Compare the Modified with the Normal i'[OCRerr]To-Level Search Scheme A) Problem Areas In structuring this experiment, the following questions zinist be answered: 1) what criteria can be used to judge the relative merits of each procedure ? 2) How should these alternative search procedures be implemented, and what parameters [OCRerr][OCRerr]st be adjusted in using these procedures ? 3) what type of document and query collection will serve as an adequate data base in order to obtain valid conclusions ? B) Tests to Compare the Effectiveness of Each Search Procedure The main criterion for effectiveness is based on the number of documents which must be scanned in each procedure in order to obtain most of the relevant documents for each query. In a practical implementation of a normal two level search scheme, the number of subsets completely searched [OCRerr][OCRerr]ll be either a fixed number for all queries, or will depend on the correlations with the query of the centroid vectors of the document subsets. Neither of these procedures for determinin& the number of cate[OCRerr]ories to be searched completely can be used to compare the effectiveness of the modified two-level search scheme with the normal two- level search, since neither the number of clusters nor the size of each cluster is the same for both search schemes. These differences in the number of clusters, and the size of each cluster make it impossible to use the same parameters for determining the number of subsets of documents to be completely searched for each of the search procedures. Further,