Introduction

Plans and Goals

The Natural Language Processing and Information Retrieval Group at NIST has developed a Z39.50-1995 UNIX client and server for use initially with our IR application prototype PRISE but designed to minimize the work of incorporating other IR engines.

The NIST PRISE system treats documents and queries as lists of words and responds to a query with a list of documents ranked in order of their statistical similarity to the query. A basic version of the PRISE search engine (without the interface and without any client/server mechanism) has been available for research use for several years (contact Donna Harman, donna.harman@nist.gov for more information).

The goals of the current work with Z39.50 include the following:

• to promote the availability of information retrieval services by publishing source code for a working Z39.50 client/server - a set of building blocks to give commercial, academic, and government IR developers/researchers a headstart.

  ZPRISE 2.0 provides a basis for experimentation with relevance feedback using positive and/or negative user relevance judgements.

• to increase our understanding of the Z39.50 standard and our ability to influence and encourage its development/use.

  We are particularly interested in the development and testing of the proposed Ranked List Query (Type 102) and in the use of Z39.50 to enable focussed comparisons of IR clients and search engines.

• to provide an enhanced, user-friendly version of our PRISE application within a Z39.50 interface.

ZPRISE 2.0

ZPRISE 2.0 is a public domain starter set implementation of Z39.50-1995 with the latest version of PRISE for TREC/TIPSTER-sized document collections and contains support for query enhancement by means of relevance feedback.

Overview of Function Supported

• Z39.50 versions
• 3,2,1


• Options
  • Init
  • Search
  • Query Types and Attributes

  When using the ZServer default translator, Type-1 queries are accepted but the query is treated as a list of words (terms). Operators are ignored.

  When using the ZServer paired with PRISE and its query translation routine, Type-101 is accepted as well with the following attributes:

  • Structure (word, phrase, wordlist)
  • Truncation (do not truncate)
  • Position (any position)
  • Use (depends on database)

  A small part of the experimental Type-102 query (as of September 1995) has been implemented to support relevance feedback between the ZClient and the ZServer/PRISE when Z39.50 Version 3 is in force.

  • Present
    • Element-sets: Full (Doc#, Title, Body) and Brief (Doc #, Title)
    
    
    • Record-Syntax: Simple Unstructured Text Records Syntax (SUTRS)
  • Close
  
  
• Proximity, Phrase, Attribute Search

ZClient and ZServer/PRISE support specification and search by

• term proximity
• phrases
• some document attributes
• Highlighted query terms in document display

The ZClient, when used with ZServer/PRISE, will present a list of matching query terms for each document displayed and highlight matching terms when displaying a full document.

• Index and Search TIPSTER/TREC-sized document collections

PRISE supports the indexing and timely searching of TIPSTER/TREC-sized document collections (about a million documents).

Dependencies

• Solaris 2.5

  The ZPRISE 2.0 package makes and runs on SunOS 4.1.x but only the version for Solaris will be supported.

• Tk 4.0 / Tcl 7.4 (available from NIST ftp site)
• nsgmls parser
• ANSI-C
• X11 R3 or later
• Note: ZPRISE 2.0 runs on the TREC Spanish collections but is probably still not entirely 8-bit clean.

Availability

ZPRISE 2.0 is in the public domain and is available from NIST at no cost. Contact Paul Over (over@nist.gov) with cc to Willie Rogers (wrogers@nist.gov) for more information.