ISR10
Scientific Report No. ISR-10 Information Storage and Retrieval
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.
feedback modified Queries.
Figures [OCRerr].8 and [OCRerr] compare the retrieval results of two
additional original and relevance feedback queries. [OCRerr]ote that for the
query "Analog-Digital1 shown in Figure 3.[OCRerr], one of the relevant
documents namely document 46, experiences a decrease in its retrieved
rank from 21st to 46th while the ranks of the other relevant documents
are substantially improved. This may be interpreted as implying that
the index image of document 46 is less associated (in terms of angular
distance in the index space) with the other relevant documents than it
is with the original query. In general, this effect occurs, whenever
the index images of the documents relevant to a search request form
distinct clusters in the index space, and when the set of relevant
documents identified by the relevance feedback consists substantially
of members from only one of these clusters. In some cases it will be
possible to identify such situations, and autqmatically to generate
multiple queries for such search requests. For the case in point,
however, the single document (no. 46) is assumed not to be identified
by the original query. In this instance, then, there is no effective
way to increase the probability of retrieving it. Such situations must
then be interpreted (assuming that there are no grounds on which to
question a user1s relevance judgments), aQ arising from deficiencies
in the indexing[OCRerr] process or from the inherent information loss which
necessarily accompanies it.
B. Average Results and Successive Iterations
The query modification procedure as illustrated in the