IRS13
Scientific Report No. IRS-13 Information Storage and Retrieval
Summary
summary
Harvard University
Gerard Salton
Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government.
search and retrieval methods follows in subsequent reports in this series.
The analysis of information search procedures and the measurement
of retrieval performance are tasks for which a clearly established method-
ology does not as yet exist. For this reason, it becomes necessary to
consider several different procedures, each designed to reveal a different
aspect of retrieval evaluation: the total system viewpoint, the viewpoint
of the user who insists on high precision, and that of the user who requires
high recall. In each case, the aim of the studies included in the present
report has been to reach conclusions which may be of practical help to the
designer of automatic information systems. The analysis used is thus "insight
oriented" rather than "proof-oriented" in the sense that a selective manual
analysis of a few typical requests is used in order to gain an understanding
of the general process. Formal statistical significance computations of the
evaluation results are contained in section III of report ISR-12.
All but three of the ten sections of this report have been prepared
by E. M. Keen. Section I is devoted to a detailed examination of the test
environment used, including a description of the document collections and
search requests, of the manner of obtaining relevance decisions, of the
variables entering into the evaluation process, and of the differences in the
query and document make-up for the three collections in use.
The evaluation parameters are examined in detail in section II. The
viewpoint used in generating the performance measures is first described.
This is followed by an introduction of some performance measures which are
particularly useful for systems producing a ranked document output in de-
creasing order of correlation between documents and search requests. Two
types of evaluation measures are used including global measures where a
xii