Large-Scale Information Retrieval with Latent Semantic Indexing

Todd A. Letsche and
Michael W. Berry
Department of Computer Science
107 Ayres Hall
University of Tennessee
Knoxville, TN 37996-1301
letsche@cs.utk.edu
berry@cs.utk.edu

Preprint submitted to Information Sciences - Applications, 1996.

• Abstract
• 1. Introduction
• 2. Vector-Space Models for Information Retrieval
  • 2.1 Introduction to Vector-Space Models
  • 2.2 Latent Semantic Indexing
    • 2.2.1 Term-Document Representation
    • 2.2.2 Weighting
    • 2.2.3 Computing the SVD
    • 2.2.4 Query Projection and Matching
    • 2.2.5 Relevance Feedback
• 3. The Design and Implementation of a Latent Semantic Indexing Server
  • 3.1 Functional Specification
  • 3.2 Original Implementation
  • 3.3 A New Implementation
    • 3.3.1 Serial Implementation
    • 3.3.2 Distributed Implementation
    • 3.3.3 C++ classes
• 4. A Client-Server Interface
  • 4.1 The LSI WWW Interface
    • 4.1.1 Features
    • 4.1.2 LSI:Book
    • 4.1.3 LSI:Query
  • 4.2 Implementation
• 5. Results and Conclusions
  • 5.1 Computing Environment
  • 5.2 Methodology
  • 5.3 Serial Results
    • 5.3.1 Timing on the Network
    • 5.3.2 LsiQuery versus LsiFinder
    • 5.3.3 LsiBackend versus LsiFinder
  • 5.4 Distributed Results
    • 5.4.1 Execution Time
    • 5.4.2 Speedup and Scalability
    • Speedup
    • Scalability
  • 5.5 Future Work
  • 5.6 Conclusions
• 6. Biographies of Authors
• References
• About this document ...

This research was supported by the National Science Foundation under grant Nos. NSF-ASC-92-03004, NSF-ASC-94-11394, and NSF-CDA-95-29459.