ISR10
Scientific Report No. ISR-10 Information Storage and Retrieval
Evaluation of Document Retrieval Systems
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.
5-50
retrieved sequence, exibits a reasonable range for the search requests
examined to date, and typically varies from .6 to 1.0. In practice then,
to produce a useful range of values for the recall index, one is forced
to expand its scale A scale expansion of 5, introduced so as to main-.
tam an upper bound of 1.0, produces an observed range for the scaled
recall h.ndex similar to that of the precision index. The scaled index
is defined as:
rn = 1.0 - 5(i.o[OCRerr]- rn)
5
where r is the normalized rank recall defined by equation (5.20).
n
Two related performance indices may be derived from the two
which have been considered. These are useful in the case where a par-
ticular query is subjected to a set of retrieval operations (varying
some system parameter for example) which are to be compared. The recall
error, equation (5.18) was derived as:
n +1
recall error = - 0
2
Since 0 = (n+1) / 2, a positive index with an upper bound of 1.0 may
mm .0
be defined as:
n +1
.0
2
rank recall =
A similar observation for the case of the derived precision error,
equation (5.21), produces the index: