MESMUSES methodology

1
MESMUSES methodology

• Lessons learned and open issues
• Alain Michard
• Florence, June 2003

2
MESMUSES broad vision

• Just like several other projects
• SW is all about semantic interoperability
• Sharing machine-readable terminologies and
  classification schemes
• Science and culture are collective and
  international
• Semantic Web methodology should be highly
  relevant for managing and sharing scientific and
  cultural information

3
Some key ST issues in the Project

• Model is RDFS / OWL-Lite adequate ?
• Schema authoring method and tools needed !
• Metadata where does it come from ?
• Automatic Indexing experiments with a
  categorizer

4
The basic SW model
Type texte imprimé, monographie Auteur(s)
Zola, Émile (1840-1902) Titre(s) L'assommoir
Texte imprimé / par Emile Zola Edition 50e
éd. Publication Paris G. Charpentier,
1878 Description matérielle 111-569 p. Notice
n FRBNF35963044
Real-world entities
5
Model and Schema Language

• Typed attributes are needed
• XML-Schema types
• Derived types (e.g. Celsius temperature,
  Gregorian date, etc.)
• Enumerated types, thesauri
• Time-stamping
• Cardinality constraints
• Explicit transitivity of properties (e.g.
  geographic inclusion)

6
Schema authoring issues (1)

• Find the right level of abstraction
• Is  Glucid  a class or an instance ?
• Or is it sometime a class and sometime an
  instance ?
• Avoid the  KR  attitude and practices !
• Its all about indexing resources with shared
  terminologies, not about representing human
  knowledge !

7
Schema authoring issues (2)
8
Schema authoring issues (3)
9
Schema authoring issues (4)

• Authoring tools are badly needed
• Graphical representation of the schema
• Zooming on sub-graphs (hierarchies)
• Versioning
• Consider using UML authoring environment ?
• Established methodology and tutorials are needed

10
Creating Surrogates

• Data extraction and fusion from structured
  sources
• R-DB, XML-DB, LDAP
• Updating
• When ?
• Should not create duplicates !
• Detect cross-references
• Authority lists
• Thesauri
• Lexical distance
• ???

11
Automatic Categorization

• Automatic indexing
• By extracting metadata from resources
• By automatic categorization
• Define hierarchies of  concepts  inside the
  schema
• Seeding with representative documents
• Machine learning to create categorizers
• Pros enriched search functionality
• Cons hierarchies of categories are static
• Adding a category may change the categorizers of
  the others

12
Bottom-line

• RDFS schema authoring may be more difficult than
  E-R modelling
• Debates on syntactic features are irrelevant
• Should be grounded on real-world implementations
  and testbeds
• A new query language (e.g. RQL) is not high
  priority
• We have not addressed the  logical rules  layer
• Semantic Web vs. Community Webs