Objective: To investigate a new approach for query expansion based on
retrieval feedback. The first objective in this study was to examine a
lternative query-expansion methods within the same retrieval-feedback
framework. The three alternatives proposed are: expansion on the MeSH
query field alone, expansion on the free-text field alone, and expansi
on on both the MeSH and the free-text fields. The second objective was
to gain further understanding of retrieval feedback by examining poss
ible dependencies on relevant documents during the feedback cycle. Des
ign: Comparative study of retrieval effectiveness using the original u
nexpanded and the alternative expanded user queries on a MEDLINE test
collection of 75 queries and 2,334 MEDLINE citations. Measurements: Re
trieval effectivenesses of the original unexpanded and the alternative
expanded queries were compared using 11-point-average precision score
s (11-AvgP). These are averages of precision scores obtained at 11 sta
ndard recall points. Results: All three expansion strategies significa
ntly improved the original queries in terms of retrieval effectiveness
. Expansion on MeSH alone was equivalent to expansion on both MeSH and
the free-text fields. Expansion on the free-text field alone improved
the queries significantly less than did the other two strategies. The
second part of the study indicated that retrieval-feedback-based expa
nsion yields significant performance improvements independent of the a
vailability of relevant documents for feedback information. Conclusion
s: Retrieval feedback offers a robust procedure for query expansion th
at is most effective for MEDLINE when applied to the MeSH field.