ISR11
Scientific Report No. ISR-11 Information Storage and Retrieval
Relevance Feedback in an Information Retrieval System
chapter
W. Riddle
T. Horwitz
R. Dietz
Harvard University
Gerard Salton
Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government.
v'-13
ieedback process is studied in detail, the validity of this assumption is
demonstrated repeatedly; for practically every query tested, the recall is
eventually increased.
The explicit assumption basic to this investigation is that the
relevance feedback and query modification proccesses can be optimized in
some sense. Although the analysis demonstrates that no uniformly best
process exists, the investigation of the relevance weighting factors, the
search correlation function, and the iteration strategy shows that same
combinations of these parameters produce methods that work well in many
cases.
Specifically, the follo[OCRerr]ng choices of the parameters constitute an
optimal system since[OCRerr]each optimizes that part of the modification process
which it affects most directly. First, the sL[OCRerr]le vector matching correlation
function is optimal in that it retrieves more relevant documents on the
initial search than do either the cosine or co-occurrence correlation
functions. Second, the iteration strategy in which [OCRerr] is increased `[OCRerr]th
every iteration is optimal in the sense that the scale of the query
updating information is effectively equal to the scale of the current query.
Finally, the use of zero and one as relevance weighting factors, including
the negative relevance weighting factor heuristic method, is optimal in
the sense that it adequately describes the feedback information provi[OCRerr]ded
by the user. Therefore, this combination of parameters yields a retrieval
system which is optimized for high precision and recall through the use of
relevance feedback information. Of course, it can be argued that some of
these choices for the parameters may lead to inefficiencies in the retrieval
process, but this objection is of no consequence in this investigation