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. 1-8 thereo£, a matching £unction is desired which is independent 0£ the vector ma[OCRerr]nitudes involved. Under these circumstances it is natural to assume that the in£ormation carried by the index vector is contained in its angular position (i.e. its orientation in the property space). The matching £unction assumed, there£ore, is the angular distance or a monotonic £unction 0£ this distance between the search request vector and the source document vector, wherein decreasing distance is assumed to indicate increasing probability 0£ relevance. I). Terminolo[OCRerr] In dealing with the £ore[OCRerr]oing model, the £ollowing de£initions are required: 1) Let[OCRerr]= [OCRerr] represent the set 0£ source documents in'the natural langu[OCRerr] comprising the re£erence collection. 2) [OCRerr] represent a set 0£ sample search requests in the natural langu[OCRerr]e comprising a test set 0£ retrieval queries. 3) Let T represent the index trans£ormation £rom the natural language to the index language. The index image 0£ a document [OCRerr] d[OCRerr]=T(I)[OCRerr])[OCRerr] and the index image 0£ a search request [OCRerr]. E [OCRerr] is q.=T([OCRerr].). [OCRerr][OCRerr]urther let l--'{d ,d[OCRerr], 1 1 1 2 where alternative models are considered, e.g. lndex images represented[OCRerr]by `sets rather than vectors, the required notation will,be introduced following the framework defined here.