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. express quite explicit formulations of their information nee[OCRerr]s ano can be expectea, therefore, to obtain relevant source accuments with a high probability of success. At the other ena of the spectrum, however, there are those users with less explicit information neeas or with only vagne familiarity with the subject area being searchea. Clearly, this class of users[OCRerr]is less likely to be able to formulate search requests which will retrieve useful references. Unaer these assumptions it is then pertinent to consiaer techniques for reaucing search request variance in two &istinct contexts. First, from an operational viewpoint, one woul& like to process search requests which are optimized with respect to the cost of retrieval, the cost of optimization, and the value of the information to the user. Second, in the context of retrieval evaluation, it is desir/ble to isolate explicitly the effect of the request formulation from the effects of indexing and request-accument matching. [OCRerr]ormally, in testing indexing devices one compares gross retrieval results obtained for some sample set of search requests. [OCRerr]he comparisons so obtained reflect the joint behavior of the test queries and the indexing scheme, but do not provide an explicit comparison of the indexing methods alone. If it were possible, however, to def irie an optimal search request (for a fixed indexing technique), corresponding to any given test query, the comparative retrieval results for the optimal requests would provide a much clearer evaluation of the power of the indexing methods, since performance variations due to request malformationwould be eliminated.