IRS13
Scientific Report No. IRS-13 Information Storage and Retrieval
An Analysis of the Documentation Requests
chapter
E. M. Keen
Harvard University
Gerard Salton
Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government.
X-20
possibly a strong correlation between these characteristics and real breadth
of requests. The superiority of the specific, long and low frequency requests
is again seen in performance figures for the 19 requests that exactly fall
into the expected combinations as seen in Figure 10. The thesaurus dictionary
is seen, as expected, to give the most improvement to the general short and
high frequency requests. Further analysis of this type awaits suitable
c[OCRerr][OCRerr]puter programs, since the hand analysis methods used are too time consuming.
C) Comparison of Requests of the Two Preparers
Since two persons were responsible for request preparation, any
variation in the measurable characteristics of generality length and fre-
quency[OCRerr] already noted may be correlated with the different preparers. Figures
Il, 12 and 13 show that a quite strong correlation does exist, since the
requests from preparer I[OCRerr]A?? are on average more specific, longer and hence
have lowermean frequencies thau requests from preparer IIBI? (Figure 11).
Figure 12 repeats the data of Figure 9, adding the request preparer distin-
ction, and Figure 13 shows that if the eight sets of results in Figure 12
are divided into two sets of four each by the diagonal line in Figure 12,
correspondence is quite marked and is probably statistically significant.
The previously examined subject request characteristics such as
studies of unclear requests, requests having a multiple need, and requests
containing identifiable important words (see part 5D) are almost equally
divided among requests of the two preparers; thus, although the requests
prepared by person 11A?' are expected to give the better performance, it is
not correct to assume that `1A11 did a better quality job than "B11. The
six requests judged difficult for the system (Part [OCRerr]) comprise five "A'1
requests and one "B", but as has been noted only three of these requests