ISR10 Synopsis synopsis 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. The indexing component of the model is discussed in chapter 2 which is primarily descriptive in nature. The work of a number of researchers in this area is cited in ciescrioi[OCRerr]C [OCRerr]e development and current trends of a[OCRerr]tomatic content analysis. Particular emphasis is placed on concept vector indexing techniques w;[OCRerr]ich incorporate thesaurus type semantic associations. The S[OCRerr]T system generates document and query index images by such techniques and these were used in obtaining the experimental results presented in ehapters 3 and 4. The[OCRerr]original contributions in chapter 2 are related to the techniques proposed for index image optimization. A search request optimization algorithm analytically derived from an assumed optimalitycriterion is presented.in[OCRerr]hapter 3. The optimization algorithm and the[OCRerr] notion of request optimization by an iterative sequence of retrieval operations are original with the author. In addition the notion of testing index language devices by the use of optimal search requests is original. Experimental results illustrating the optimization process are presented. These were derived by a simulation which was coded and run on the IBM 7O[OCRerr]4 in cbnjunction with the S[OCRerr]ART retrieval system. A search request formulation based on this optimization technique offers the promise of improving' the performance of document retrieval sYstems, given[OCRerr]the current state' of development of computer time sharing and man-machine communication technology. Chapter 4 pre[OCRerr]ents `an original automatic document [OCRerr]clas'sification algorithm, heuristically motivated by considerations of search'e'fficiency and by the functional nature of query-document matching operations. This algorithm was coded in Fortran and run on xvi