SP500207
NIST Special Publication 500-207: The First Text REtrieval Conference (TREC-1)
Application of the Automatic Message Router to the TIPSTER Collection
chapter
R. Jones
S. Leung
D.L. Pape
National Institute of Standards and Technology
Donna K. Harman
The system achieved a throughput of one document a second against 200 filters. This was
especially pleasing since the experiments were run on a relatively modest platform, an 80486133
processor with 8 Mbytes of memory under SCO UNIX.
As measured against the original objectives described above, three of the four have been
achieved: AMR techniques have proved to be as accurate as any other in current use, The tests
certinly proved that AMR is robust in running a very large collection, its performance was also
quite adequate. The system was tuned during the first set of runs, and the experience has proved
to be very valuable. The only major objective that remains to be achieved is to analyse AMR's
sensitivity to widely differing filter types. Work is proceeding on this.
References
(1) Jones R.L., Enhanced Retrieval Mechanisms for Free Text
Data Bases, Proceedings First Pan Pacific Computer
Conference, (1985), pp 134-143
248