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.
5-25
or
n +1
recall error = - 0 (5.18)
2
Thus the integral of the difference between the recall function for a
perfect retrieval and the recall function of an actual retrieval oper-
ation is the difference between the actual average rank ([OCRerr]) of the
members of the set of relevant documents and the average rank
(n0+1.)/2 which would obtain under perfect retrieval.
This parameter may be normalized to the range C -
ering the case fQr which the rank of every member of D is
greater than every member of D This is clearly the case
error; therefore:
1 by consid-
numerically
of maximum
n
0
1 \[OCRerr] - n +1
max recall error =7 ½ N - (ii' 2
i=1
Hence:
= n10[OCRerr] no (N + N - n0 + i)$ - n0 +2 1
N-n
0
- n[OCRerr] +2 1
re (5.19)
n N-n
0
* is a normalized index of the recall error. As this parameter measures
recall error, it' is desirable to reverse it. Therefore:
r = 1 h[OCRerr]-_(nO +1
fl)
n ½ N-n
L `0
*([OCRerr].2o)