Text REtrieval Conference (TREC)
System Description

Organization Name: IBM T. J. Watson Research Center Run ID: ibms97a
Section 1.0 System Summary and Timing
Section 1.1 System Information
Hardware Model Used for TREC Experiment: IBM RS/6000 m. 591
System Use: SHARED
Total Amount of Hard Disk Storage: 24.5 Gb
Total Amount of RAM: 256 MB
Clock Rate of CPU: 77 MHz
Section 1.2 System Comparisons
Amount of developmental "Software Engineering": ALL
List of features that are not present in the system, but would have been beneficial to have:
List of features that are present in the system, and impacted its performance, but are not detailed within this form:
Section 2.0 Construction of Indices, Knowledge Bases, and Other Data Structures
Length of the stopword list: 512 words
Type of Stemming: MORPHOLOGICAL
Controlled Vocabulary: NO
Term weighting: NO
  • Additional Comments on term weighting:
Phrase discovery: YES
  • Kind of phrase: bigrams
  • Method used: STATISTICAL
Type of Spelling Correction: MANUAL CORRECTION
Manually-Indexed Terms: NO
Proper Noun Identification: NO
Syntactic Parsing: NO
Tokenizer: YES
Word Sense Disambiguation: NO
Other technique: NO
Additional comments:
Section 3.0 Statistics on Data Structures Built from TREC Text
Section 3.1 First Data Structure
Structure Type: N-GRAMS, SUFFIX ARRAYS, SIGNATURE FILES
Type of other data structure used:
Brief description of method using other data structure:
Total storage used: 1.3 Gb
Total computer time to build: 84 hours
Automatic process: YES
Manual hours required: hours
Type of manual labor: NONE
Term positions used: NO
Only single terms used: YES
Concepts (vs. single terms) represented: NO
  • Number of concepts represented:
Type of representation:
Auxilary files used: NO
  • Type of auxilary files used:
Additional comments: The above specified time includes all the processing from the raw data to unigrams and bigrams.
Section 3.2 Second Data Structure
Structure Type: NONE
Type of other data structure used:
Brief description of method using other data structure:
Total storage used: Gb
Total computer time to build: hours
Automatic process:
Manual hours required: hours
Type of manual labor: NONE
Term positions used:
Only single terms used:
Concepts (vs. single terms) represented:
  • Number of concepts represented:
Type of representation:
Auxilary files used:
  • Type of auxilary files used:
Additional comments:
Section 3.3 Third Data Structure
Structure Type: NONE
Type of other data structure used:
Brief description of method using other data structure:
Total storage used: Gb
Total computer time to build: hours
Automatic process:
Manual hours required: hours
Type of manual labor: NONE
Term positions used:
Only single terms used:
Concepts (vs. single terms) represented:
  • Number of concepts represented:
Type of representation:
Auxilary files used:
  • Type of auxilary files used:
Additional comments:
Section 4.0 Data Built from Sources Other than the Input Text
Internally-built Auxiliary File

File type: NONE
Domain type: DOMAIN INDEPENDENT
Total Storage: Gb
Number of Concepts Represented: concepts
Type of representation: NONE
Automatic or Manual:
  • Total Time to Build: hours
  • Total Time to Modify (if already built): hours
Type of Manual Labor used: NONE
Additional comments:
Externally-built Auxiliary File

File is: NONE
Total Storage: Gb
Number of Concepts Represented: concepts
Type of representation: NONE
Additional comments:
Section 5.0 Computer Searching
Average computer time to search (per query): 1540 CPU seconds
Times broken down by component(s):
Section 5.1 Searching Methods
Vector space model: YES
Probabilistic model: YES
Cluster searching: NO
N-gram matching: NO
Boolean matching: NO
Fuzzy logic: NO
Free text scanning: NO
Neural networks: NO
Conceptual graphic matching: NO
Other: NO
Additional comments: The above specified time is for the first pass scoring only. It takes tens of hours to train the models we applied for rescoring, using the current (not very efficient) implementation of query expansion method.
Section 5.2 Factors in Ranking
Term frequency: YES
Inverse document frequency: YES
Other term weights: NO
Semantic closeness: NO
Position in document: NO
Syntactic clues: NO
Proximity of terms: NO
Information theoretic weights: NO
Document length: YES
Percentage of query terms which match: NO
N-gram frequency: YES
Word specificity: NO
Word sense frequency: NO
Cluster distance: NO
Other: NO
Additional comments:
Section 6.0 Query Construction
Section 6.1 Automatically Built Queries for Ad-hoc Tasks
Topic fields used:       DESCRIPTION    
Average computer time to build query    CPU seconds
Term weighting (weights based on terms in topics): NO
Phrase extraction from topics: NO
Syntactic parsing of topics: NO
Word sense disambiguation: NO
Proper noun identification algorithm: NO
Tokenizer: YES
  • Patterns which were tokenized:
Expansion of queries using previously constructed data structures: YES
  • Comment: please se comments in 5.0
Automatic addition of: NONE
Section 6.2 Manually Constructed Queries for Ad-hoc Tasks
Topic fields used: NONE        
Average time to build query?   minutes
Type of query builder: OTHER
Tool used to build query:
Method used in intial query construction?
  • If yes, what was the source of terms?
Total CPU time for all iterations:  seconds
Clock time from initial construction of query to completion of final query:   minutes
Average number of iterations:
Average number of documents examined per iteration:
Minimum number of iterations:
Maximum number of iterations:
The end of an iteration is determined by:
Automatic term reweighting from relevant documents:
Automatic query expansion from relevant documents:
  • Type of automatic query expansion:
Other automatic methods:
  • Other automatic methods included:
Manual methods used:
  • Type of manual method used:
Send questions to trec@nist.gov

Disclaimer: Contents of this online document are not necessarily the official views of, nor endorsed by the U.S. Government, the Department of Commerce, or NIST.