System Summary and Timing
  Organization Name: Logicon
  List of Run ID's: losPA2, losPA3

  Construction of Indices, Knowledge Bases, and other Data Structures 

    Methods Used to build Data Structures 

    - Length (in words) of the stopword list: 580 
    - Controlled Vocabulary? : No 
    - Stemming Algorithm: No              
    - Term Weighting: No 
    -  Phrase Discovery? :              
    -  Tokenizer? : Yes            
      - Patterns which are tokenized: All alphanumeric strings were tokenized.
        Only alphabetic tokens of length > 1 were kept.  The following were 
        then eliminated: stopwords; tokens serving as SGML tags; tokens 
        occurring <= 2 times on each CDROM. 

    Statistics on Data Structures built from TREC Text

    - Inverted index           
    - Clusters           
    - N-grams, Suffix arrays, Signature Files           
    - Knowledge Bases            
      - Use of Manual Labor                  
    - Special Routing Structures           
    - Other Data Structures built from TREC text           
      - Run ID : losPA2, losPA3 
      - Type of Structure: Corpus word-frequency table 
      - Total Storage (in MB): 15 
      - Total Computer Time to Build (in hours): 9 
      - Automatic Process? (If not, number of manual hours): Yes
      - Brief Description of Method: Process analyzed all documents on all 3 
        CDROMs.  Each document was tokenized, and a count was kept with each 
        token indicating the number of documents in which it occurred. 

  Query construction

    Automatically Built Queries (Routing)

    - Topic Fields Used: None 
    - Average Computer Time to Build Query (in cpu seconds): CPU time was not 
      measured.  Average wall-clock time was 324 seconds. 
    - Method used in Query Construction          
      - Terms Selected From            
        - Only Documents with Relevance Judgments: Yes 
      - Term Weighting with Weights Based on terms in            
        - All Training Documents: Yes 
        - Documents with Relevance Judgments: Yes 
      - Phrase Extraction from            
      - Syntactic Parsing            
      - Word Sense Disambiguation using            
      - Proper Noun Identification Algorithm from            
      - Tokenizer             
        - Patterns which are tokenized (dates, phone numbers, common patterns, 
          etc): Same as tokenizing rules for Data Structures. 
        - from Documents with Relevance Judgments: Yes 
      - Heuristic Associations to Add Terms from            


      - Expansion of Queries using Previously-Constructed Data Structure:              
      - Automatic Addition of Boolean connectors or Proximity Operators using 
        information from            
        - Documents with Relevance Judgments: Yes.  Any 2 of the 10 "most 
          descriptive" tokens were specified in an AND condition for document 
          selection.  The 1000 "most descriptive" tokens were then used for 
          document ranking. 

  Searching

    Search Times

      - Run ID : losPA2, losPA3
      - Computer Time to Search (Average per Query, in CPU seconds): CPU time 
        was not measured.  Average wall-clock time was 0.45 seconds per 
        document, or 0.009 seconds per profile per document. 

    Machine Searching Methods

      - Boolean Matching? : Yes 
      - Free Text Scanning? : Yes. Search engine was the Logicon Message
        Dissemination System (LMDS), a COTS product. 

    Factors in Ranking

      - Term Frequency? : Yes.  Term frequency in the set of relevant training 
        documents vs. the set of all training documents. 
      - Proximity of Terms? : Passage-level searching was employed as part of 
        the scoring algorithm.  A passage consisted of 100 contiguous non-
        stopwords, and each new passage began 50 non-stopwords after the 
        beginning of the previous passage. 
      - Document Length? : Yes.  Score from term weights was multiplied by the 
        ratio of unique query tokens in document or passage to total unique 
        tokens in document or passage. 

    Machine Information

    - Machine Type for TREC Experiment: Sun SPARCstation 10 
    - Was the Machine Dedicated or Shared: Dedicated 
    - Amount of Hard Disk Storage (in MB): 5,000 MB 
    - Amount of RAM (in MB): 32 MB
    - Clock Rate of CPU (in MHz): 50 MHz 

    System Comparisons 

    - Amount of "Software Engineering" which went into the Development of the 
      System: The search engine was the Logicon Message Dissemination System 
      (LMDS), a COTS product.  The query creation and document ranking software
      were research prototypes integrated via its API.   
    - Given appropriate resources            
      - Could your system run faster? : Yes 
      - By how much (estimate)? : 50 percent