ISR10 Search Request Formulation 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. Since operationally it is expected that only a few iterations ½ Would ever be used, the differences between these alternative formulations is not of major significance. If the user is satisfied with the relevant documents identified by the previous iterations and would, in effect, like to find others which are closely related to these documents, queries produced by equatioxi ([OCRerr].15) would be more suitable. If on the other hand, he is interested in maintaining a broader search, the iterations produced by equation (3.14) will not be as dependent on the relevant documents previously identified (members T of [OCRerr] Average performance results for a second iteration of relevance feedback produced by each of these alternatives are shown in Figure 3.12. The results obtained with the original and first iteration queries are included for comparison. As can be seen by these graphs, the results obtained from using the iteration formula of equation (3.15) are somewhat better than when the second iteration starts from the original query. Rowever, in comparing the behavior of these alternatives on individual queries, there are some cases in which the reverse is true. Figure 3.18 illustrates an example of this. In this case it is clear that documents 315 and 264 are not clustered in the index space with the other relevant documents; and therefore, these documents suffer more drastically from successive iterations (equation (3.15) ) than from successive modifications to the original query (equation (3.14) ).