Almost in the very beginning ...
The seed which has grown into CeDAR - the Centre for Database Access Research was probably planted way back in 1973 at the early days of online searching. The Marconi Research Laboratories at Gt. Baddow in Essex had developed an Automated Ultrafiche Terminal capable of storing enormous quantities of information on high density microform. This device offered access for a wide variety of potential applications from telephone directories to criminal records, maps to images of grasses brought back by Darwin from Australia, learning programmes to literature abstracts. This was a rich area for research and development and the writing of software to make it possible to search what could amount to gigabytes of information on the desktop - yes this was 23 years ago. Alas this technology never made it from R & D to widespread application ... but that's another story ...
Seven years later ...
The CeDAR story resumes at the Polytechnic of Huddersfield where a project to investigate the information needs of cancer therapy clinicians began in 1980 in collaboration with the University of Leeds. What has followed is a succession of projects which have investigated and developed interfaces for end-user searching of databases. The evolution of these interfaces follows a logical progression from an expert systems approach to what is termed view-based searching where some of the key interface features have survived from the first prototype - CANSEARCH. What follows is a brief overview of the prototypes and projects highlighting what survived and what became extinct.
An expert systems approach - promising much and after several years delivering promise:
A touch screen display formed by selected hierarchies for potential facets expected in queries for cancer therapy literature. A rule-based program which controlled the interaction and formulated legal MEDLINE search statements.
Sometimes (10%ish) it performed better than the real thing - as judged by a real expert (i.e. not me) - at least in terms of the quality of the search statements generated. Like all good early expert systems (including MYCIN) CANSEARCH was never used in anger.
- Good features:
Easy to use interface which offered selections to the user rather than requiring them to type
Browsability of controlled vocabulary (Medical Subject Headings)
Automatic inclusive searching
Occasionally it even impressed it's author - once given a rule it never forgot it
- Not so good features (in hindsight):
Complicated and time consuming to build
Didn't perform as well as the professional
MenUSE - Menu-based User Search Engine (1987-1993ish)
The first MenUSE system was built at the National Library of Medicine in the USA and could be used to search the complete MEDLINE database - switching subject matter revealed major weaknesses in the expert systems approach. (beware advocates of intelligent agents - been there, done that, no T-shirt, good luck with the programming!)
The original target information was on biotechnology, and as no clear CANSEARCH-like model could be devised, a simpler menu-based approach using combinations of selected concepts (akin to Quorum searching) was used to present the user with several sets - one for each combination - as the outcome of their interaction. Browsable menus were eventually automatically generated from the Medical Subject Headings. Subsequent development for the INSPEC and EPOQUE(European Parliament Online Query system) databases - courtesy of funding from Huddersfield Polytechnic (as was) - demonstrated general applicability and significant potential for multilingual information retrieval
Never really tested - but it did work and it would always give a result - unlike other approaches available at the time which used a form-based model of interaction. The highlights included demonstrating a search in Japanese of INSPEC on DataStar from the SIGIR conference in Copenhagen in the Summer of '92 and a three day event in Brussels, including prime billing at the Council of Ministers Working Party on Legal Data Processing in Brussels in May '93.
- Good features:
As for CANSEARCH +
Extensive vocabulary with direct access to menus at all levels using entry terms
Relatively easy to replicate for other databases in different subject areas - using the INSPEC thesaurus and EUROVOC.
More informative screens incorporated the number of documents against each concept or term derived from inclusive searching
- Not so good features:
Limited use of interaction - the process was one of complete search specification and then a search
Required a thesaurus which would ideally have a limited number of top-level concepts
VUSE - View-based User Search Engine (1993ish - )
A feature was introduced to MenUSE for INSPEC and EuroMenUSE (yes we had to call the European Parliament System something like that) which came to shift the thinking radically. The feature was implicit Boolean searching through filtered views:
Browsable hierarchies, as in MenUSE, provided the means for users to specify the subject matter of their search, but this was then used to progressively refine the database. Subsequent views of the vocabulary would provide data on the number of documents matching the condition of the filter and each entry on the view. It seemed appropriate to drop the reference to menus as the user was no longer selecting components of a search statement, they were electing to see a breakdown of a selected subset of documents according to different sets of criteria.
VUSE for INSPEC was tested at the conclusion of research which examined ranking and relevance feedback extensions to a view-based system - and found to perform slightly better in respect of recall and precision. Yet these sessions were mainly mediated - somewhere in the design process we introduced too much complexity. Highlights included watching MEP Assistants (including Mr Kinnock Jr) in Brussels competing to demonstrate whose MEP was most active in what area of the Parliaments business.
- Good features:
More powerful searching - akin to an analysis e.g. profiling sets of documents by year or treatment code
More interesting to use, the approach lent itself to exploration of the databases.
- Not so good features:
Difficult to understand and use - more so VUSE for INSPEC than VUSE for EPOQUE
Proposals to enhance the forms-based Watch-CORDIS interface using VUSE are still under consideration at the European Commission in Luxembourg (I hope!).
HIBROWSE - High-resolution Interface for Database Specific BROWsing and Searching (honest) (1993ish - )
The usability problems of VUSE were serious - a parallel development in CeDAR demonstrating improved access to an ORACLE database - HIBROWSE for Hotels - offered features which might overcome these problems. A British Library Funded project which went live on March 1st 1995 provided the opportunity to examine usability issues in applying HIBROWSE search techniques (View-based searching) to the EMBASE database published by Elsevier Science Publishers BV in Amsterdam :
In the early attempts at design the user was to be presented with multiple views for a chosen set of search parameters - browsable indexing vocabulary (EMTREE), year of publication, author, journal, institution. This just would not work (in spite of the confidence with which it was presented at the Online 95 conference!). The current approach - illustrated below - enables the user to select multiple views from the vocabulary according to the major divisions (facets) of EMTREE (and more recently EPOQUE - we just can't leave it alone). The chosen views serve to mutually filter each other. Each view (or facet) can be refined to reduce the number of documents or expanded to increase them - forcing reciprocal updating of the numbers of documents against each term or concept on the presented views.
Looks promising but we have to develop the prototypes (one a Visual Basic front-end to EMBASE on STN-International and EPOQUE, the second a development in Natural for Windows accessing a subset of EMBASE held as an ADABAS database on our own Sun Server - with co-sponsorship from Software AG of the UK and Sun MicroSystems) to a level that we can put them in front of real users. But you can bet all the money you have the outcome will be very superior than what is currently on offer to the end-user.
- Good features:
Even more powerful searching - a much more drivable interaction
Even more interesting to use, the approach lends itself to deeper exploration of the databases where we can programme the full HIBROWSE functionality and find out, for example, which Institutions write the most papers on Alzheimer's disease
- Not so good features:
We'll tell you after the usability testing
- High points so far:
Presenting at the INFO 96 conference at the Tel Aviv Hilton in May 1996
Presenting at the ISKO 96 conference at the Library of Congress in July 1996
- Best description in HTML
View-based Searching on Intranet Systems (1996-1999) - A VACANCY for a RESEARCHER
Doing it for Web browsers - hopefully with collaboration from two large and well respected organisations - there is a University Bursary, which will hopefully be enhanced, to recruit a student who would register for a PhD. Anyone interested please contact Steve Pollitt to discuss the opportunity.
Steve Pollitt is the UK coordinator for
ISKO - The International Society for Knowledge Organisation.
HIBROWSE, VUSE and View-based Searching are Trade Marks of the University of Huddersfield
CeDAR - Centre for Database Access Research
School of Computing and Mathematics
University of Huddersfield
UK HD1 3DH
Tel: +44 (0) 1484 472147/472248
Fax: +44 (0) 1484 421106
The Web pages for CeDAR can be found at: http://www.hud.ac.uk/schools/cedar