The SMART System:

1
The SMART System
Progress Report on System Acquisition and Set-Up
Danyel Fisher Jonathan Henke Jason Hong
Jonathan Huang Jeane Stetson

• March 8, 2000
• IS 240 Principles of Information Retrieval

2
Background

• Developed 1961-64 at Harvard
• Maintained at Cornell University
• Tested at every TREC conference
• Emphasis automatic retrieval (rather than
  interactive)
• Vector-based analysis, tf x idf weighting
• Current version 13.3 (we have 11.0)

3
Bibliography

• Salton, Gerard. The SMART retrieval system
  experiments in automatic document processing.
  Englewood Cliffs, N.J., Prentice-Hall. 1971
• Salton, Gerard. Developments in Automatic Text
  Retrieval. Science, 1991 Aug 30, v253
  n5023974-980.
• TREC Proceedings
• SMART Staff, User's Manual for the SMART
  Information Retrieval System. Technical Report
  71-95, Revised April 1974. Cornell University
  (1974).
• C. Buckley, Implemetation of the SMART
  Information Retrieval System. Technical Report
  85-686, Cornell University (1985).

4
Indexing (Creating a Collection)

• Document pre-parsing
• recognize document structure and convert to a
  standard format
• Finding handling indexable information
• parsing, stopword removal, stemming, term
  clustering, synonym dictionaries, etc.
• Query handling
• parsing, stopword removal, stemming, etc.
  (parallel to document handling)

5
Indexing (Creating a Collection)

• Retrieval methods
• term weighting and similarity evaluation
• Default standard tf x idf weighting, vector
  inner product
• Output format display

6
Indexing Customizable Elements

• Document location format
• Indexable information index format
• Query format
• Retrieval method (document/query comparison)
• Output/display format

7
System Architecture

• 350 source files
• 45,000 lines of code
• Can include user-programmed modules

8
Set-up Procedure

• Download source code
• ftp//ftp.cs.cornell.edu/pub/smart
• Compile
• Look for documentation
• Indexing completed using default settings
• Unable to complete query yet
• Unable to examine index
• Cannot verify success of indexing!

9
System Documentation

• Minimal
• Poorly explained
• Cryptic
• Uses their own specific terminology

10
Problems Faced

• Virtually every feature is customizable
• Somewhere there are people who know how to do the
  customization..
• SMART suffers from the advantages and
  disadvantages of most academic research software.
  It's designed to be extremely flexible (as long
  as you know what you're doing!) - SMART manual
• Documentation is too high level.

11
Further Steps

• Complete a query using default settings.
• Identify specific files for adjusting each
  customizable feature.
• Determine how to modify each feature.

12
Recommendations Advice

• Find someone who has actually worked with the
  system before.
• Understanding operation requires examination of C
  source code.
• Customization requires modifying / creating C
  code.