ISR10 Introduction 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. 1-6 experiments reported in this thesis. 3. A Speci£ic Model - The SMART System The experimental results to be presented here in connection * with the optimization and evaluation al[OCRerr]rithms were obtained by simulation, assumi[OCRerr] a speci£ic model £or a mechanized document * retrieval system. This m6del is based On the SMART retrieval system developed at Harvard University under the direction 0£ [OCRerr]o£essor [OCRerr]rard Saltone15 The primary £eatures 0£ those elements 0£ the SMART system 0£ interest here are brie£ly outlined, so that it may become possible to re£er to' [OCRerr]hem in succeedi[OCRerr] chapters. A more thorou[OCRerr]h outline 0£ the SMART system is given in Appendix A. A. [OCRerr]operty Vector Indexing * -***. * Index images 0£ source documents in the experimental system * are assumed to be property vector representations 0£ document content. - - For present purposes it is su££icient to assume that the index image 0£ a re£erence document is an n-dimensional `vector in a property space in * * * which the weight or magnitude 0£ a given component (or attribute) * re£lects the de[OCRerr]ee' to which that attribute characterizes the content * 0£ the source text. Speci£ically, the index i[OCRerr]es experimentally used were constructed by a thesaurus trans£ormation 0£ the input text. An attribute 0£ the resulting index space corresponds to a thesaurus *category (group.o£ semantically related natural langage terms), and attribute weight is derived £rom the £req'uency 0£ occurrence 0£ the