ISR10 Search Request Formulation chapter Joseph John Rocchio Harvard University Gerard Salton Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government. 5-21 would be produced by the equation n2 q1 = n2q0 - L i=1 (5.i[OCRerr]) where S =[OCRerr][OCRerr]1'[OCRerr]2 [OCRerr] 5n} constitutes those documents originally 2 retrieved but judged nonrelevant. In this case the modified query vector q1 represents a vector in the index space which is both related to the original query, and also moved from that region of the index space which was identi- fied by the original retrieval operation. A sequence 0£ such modifications allows controlled pertubations within the region 0£ the index space 0£ interest. Thus, even in the case where an initial search provides no relevant documents, the relevance feedback algorithm can be used to provide helpful query modifications to an iterative search process. 6. Experimental Results A. Some Sample Search Requests To test the effectiveness of the search request modification process based on relevance feedback as outlined above, some experi- ments were conducted using the SMART automatic document retrieval system. A set of sample search requests and a collection of 405 abstracts from the computer literature originally published in the (1 - -. - - . - -- - - [OCRerr] - [OCRerr]-¼-.-.[OCRerr]-:: ......- [OCRerr] --`-- L.i..- -.