Biotech Glossary interface

We looked at the problem considering the various layers:

  1. 1.      Data layer

a.      Could use VocBench but is often slow – might be best to leave the editing and dissemination environments separate
b.      Could use another triple store such as Allegrograph or Sesame – anything with a SPARQL endpoint.

  1. 2.      Service layer

a.      Could use VocBench web services
b.      Could write custom web services or try someone else’s service library
c.       Could directly interact with SPARQL endpoint

  1. 3.      View layer

a.      Could use out-of-the-box Drupal and write some thesaurus view code inside it and skin for biotech.
b.      Could try to modify the EuroVoc code and skin for biotech (poor docs, seems model-dependent).
c.       Could try some other out-of-the-box SKOS-based thesaurus viewer, even commercial
d.      Could use some other framework of their choosing and write view code inside it and skin for biotech
We also briefly considered the possibility of Drupal 7 and import the entire glossary into it to make the entire site based on the biotech RDF but this a) still doesn’t create a nice navigable interface, b) scalability unknown – Burley failed to do it with AGROVOC, c) maintenance issues are unclear.
Our feeling was that the choice we were most confident and knowledgeable about was: 1.b + 2.c + 3.a
This also would get us part way to the code we need for AGROVOC.
Our best guess for the Biotech people was the following:

  • 1 week to skin Drupal to their desired look and feel
  • 2 weeks to get the data working nicely in a triple store
  • 2 weeks to build a nice glossary front-end inside the skinned Drupal

We should just add a few week for “known unknowns” while were at it. How does 2 person months sound?
Then the big part that really needs to be done is on their side. The import has shown that there is a lot of dirty data that needs cleaning by a domain expert. We estimate about 3 months time.
Of course, in the end, they get quite a lot:

  1. A clean glossary.
  2. An environment in which to edit it.
  3. An entire customizable website that can not only display their glossary, but they can build an entire community if they so desire.
  4. Their glossary published using latest good practices of LOD.

We also took a good look at what we liked and didn’t like about thesaurus interfaces. In fact the current AGROVOC interface in AIMS is horrible. Gudrun will begin making an assessment of existing interfaces and compile a report of the good, the bad and the ugly so that we can be sure and design the best possible end-user experience, not just for biotech, but for AGROVOC. One thing we noted as a STRONG commonality across thesauri interfaces is, they are stand-alone. Part of why AGROVOC is so horrible in AIMS is that…it is in AIMS. A couple things we noted right away:

  • Should be stand-alone
  • Needs basic and advanced search
  • Multi-lingual visualization is tricky – we have one of the broadest thesauri in amount of languages
  • We like many aspects of the EuroVoc, NALT and CABI interfaces


Start date: 

People involved: 





© 2007 - 2020 Valeria Pesce
Twitter icon
Facebook icon
LinkedIn icon icon
StumbleUpon icon
Digg icon
Reddit icon
Technorati icon