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-5 queries. The results of these sample runs, presented in the following paragraphs, are used to develop strategies which are then applied to the entire set of queries. A) Determination of the [OCRerr]umber of Documents Retrieved The number of documents, n, that are returned to the user is set at fifteen. The saznple runs show that if this number is reduced to eight, the effectiveness of the updating process is dIminished. In the case cited in Figure 1, returning fifteen documents leads to the retrieval of four relevant documents after three modifications are made, while returning only eight documents leads to the final retrieval of only two relevant documents after the same number of modifications. This implies the need to return initially as many relevant documents as possible so that more information can be used in the updating procedure. (The number of relevant documents initially retrieved also depends on the correlationfunction, as is discussed later.) Further, in determining the number of documents to be retrieved, a co[OCRerr]promise must be made between the desirability of retrieving a large number of documents and the desirability of not imposing a large reading task on the user. B) The Effect of the Correlation Function The result of an iteration is a list of n documents ranked by their correlations with the query. These correlations are determined by one of the following correlation functions: Cosine correlation function:£7] m i[OCRerr]l (q[OCRerr]d[OCRerr]) \/ ([OCRerr]mz=1[OCRerr][OCRerr][OCRerr] ) x ([OCRerr]m[OCRerr]1didi