ISR11 Relevance Feedback in an Information Retrieval System chapter W. Riddle T. Horwitz R. Dietz Harvard University Gerard Salton Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government. vI-~ results are presented; finally, conclusions are drawn from the study. 2. Principal Methods The relevance feedback information is used iteratively to perturb the query vector in the following manner: [2] Lu where =L] +[OCRerr] [OCRerr]2 ].[i nxl m = number of indexing terms n = number of retrieved document R = mxn matrix in which the ith column is the concept vector of the document of rank i W = nxl vector of relevance weighting factors It is assumed that query [OCRerr] has caused a set, R, of n documents to beretrieved. Using the a priori jud[OCRerr]ents of the relevance of the retrieved documents, the relevance weighting factors are determined, and are used to define W, a vector of length n, so that the relevance weight of the document of rank j is the jth element. The relevance weight of a document reflects the relevance of that document to the query (the determination of the maanitude of these weights is discussed later). [OCRerr] is a multiplier control- ling the strength of the perturbation to the query [OCRerr]. The end result of this modification is that scme linear combination of the kth elements of the retrieved documents is added to the kth element of the query vector Q, for all k of the indexing terms, thus producing a new query [OCRerr] In order to determine the effects of varying the pammeters [OCRerr], n, and the relevance weighting factors, sample runs are made using selected