A Semantic Modelling Approach to Biological Parameter Interoperability

1
A Semantic Modelling Approach to Biological
Parameter Interoperability
Ocean Biodiversity Informatics
Roy Lowry Laura Bird British Oceanographic Data
Centre Pieter Haaring RIKZ, Rijkswaterstaat, The
Netherlands
2
Presentation Overview

• The nature of the problem
• Dictionaries and data models
• The starting position
• Manual mapping
• Automation through semantic matching
• From dictionary to semantic model
• Mapping semantic models
• Semantic model applications
• Conclusions and lessons learned

3
The Nature of the Problem

• BODC and Rijkswaterstaat both have marine
  databases holding a wide range of physical,
  chemical and biological parameters
• Both were to be included pan-European
  metadatabases (EDIOS and SEA-SEARCH CDI) using a
  common discovery vocabulary
• BODC set up the vocabulary and obviously included
  a mapping to the BODC Parameter Dictionary
• Problem arose of how to provide a similar mapping
  for the Rijkswaterstaat
• If the Rijkswaterstaat data markup vocabulary
  could be mapped to the BODC Parameter Dictionary
  then the BODC discovery vocabulary mapping could
  be used

4
Dictionaries and Data Models

• BODC systems have roots in the GF3 model, which
  means
• Data values are linked to a parameter code
• Parameter code is defined in a Parameter
  Dictionary
• The parameter code specifies more than one
  metadata item for the data value
• For chemical and biological data more than one
  becomes a lot

5
Dictionaries and Data Models

• Rijkswaterstaat uses data models (DONAR becoming
  WADI)
• Measurements are accompanied by attributes
  containing specific atomic metadata items
• Each attribute is populated from a controlled
  vocabulary
• DONAR constrains attribute term combinations
  using a parameter dictionary concept
• WADI reduces maintenance overheads by allowing
  any combination

6
The Starting Position

• BODC
• Parameter Codes defined by two plain-text fields
• Related semantic information not necessarily in
  the same field
• Fields would not concatenate sensibly
• OK for humans, but not for machines
• Rijkswaterstaat
• Consistently located semantics
• Metadata fields that concatenate sensibly in both
  Dutch and English

7
Manual Mapping

• Manual mapping protocol
• For each entry in the Rijkswaterstaat
  dictionary spreadsheet
• Look up code with identical meaning using BODC
  Dictionary search tools (Access Filter by Form)
• If found
• Copy BODC code from Access and paste into
  spreadsheet
• Else
• Prepare dictionary update record and submit for
  QA and load
• Error prone and 500 entries is pushing the limit
  of human endurance!

8
Semantic Matching

• When code lists run into thousands, automation is
  required
• Rijkswaterstaat developed a semantic matching
  tool to pull matching terms (preferably one) from
  the BODC dictionary
• Defeated by the lack of standardisation in the
  BODC plain-text fields e.g.
• Calanus abundance
• Abundance of Calanus
• Calanus count
• Number of Calanus

9
Dictionary to Semantic Model

• Became apparent that the BODC Dictionary required
  significant improvement if it was to support
  mapping automation
• Development strategy was to model the parameter
  code in the same way DONAR models a measurement
• Semantic model developed to cover all codes in
  BODC Dictionary

10
Dictionary to Semantic Model

• Semantic model developed from DONAR with an
  increased semantic element count to overcome
  shoe-horning
• Principle that semantic elements may be combined
  automatically to produce text descriptions
  maintained
• Currently implemented as three sub-models
• Element superset will ultimately be created as a
  single model

11
Dictionary to Semantic Model

• Biological sub-model semantic elements
• Parameter (Abundance, Biomass)
• Taxon_code (ITIS code)
• Taxon_name
• Taxon_subgroup (gender, size, stage)
• Parameter_compartment_relationship (per unit
  volume of the, per unit area of the)
• Compartment (water column, bed, sediment)
• Sample_preparation
• Analysis
• Data_processing
• Needs further refinement e.g. subdivide
  Taxon_subgroup

12
Mapping Semantic Models

• Two stage process
• First map the semantic elements
• DONAR Parameter BODC Parameter
  Parameter_compartment_relationship
• DONAR Compartment BODC Compartment
• Then map vocabularies for mapped elements
• Surface water water column
• Relational database designers will recognise this
  as normalisation

13
Mapping Semantic Models

• Number of look-ups required is reduced by an
  order of magnitude
• Vocabulary elements have simple semantics so
  automation is possible
• Approximately 90 of the Rijkswaterstaat to BODC
  mapping accomplished by a single SQL statement
• Straightforward extension of vocabulary maps
  (different names for same thing) sorted out most
  of the rest
• Thesauri could help reduce the need for this

14
Mapping Semantic Models

• Hard Core problems required manual resolution
• Unclear or ambiguous semantics in Rijkswaterstaat
  element vocabularies (residual beta)
• Problems with Dutch to English translation
• Some mapping errors were detected
• Caused by homonyms (Branchiura)
• Emphasises the need for more than just a name for
  a taxon (reference or ITIS code)

15
Semantic Model Applications

• Semantic modelling is a lowest common denominator
  approach to metadata
• This is what makes it good for mapping
• The approach also offers the basis for
  user-controlled data discovery and
  interoperability
• User chooses the semantic element subset
• User data selection interaction based on the
  subset vocabulary
• Automated interoperability requires more
  sophistication (thesauri, ontologies)

16
Conclusions

• Dont even think about manual mapping of large
  parameter dictionaries
• 99 of a map is completed in the first 10 of the
  time
• More standardisation means fewer errors and
  problems
• Semantic model vocabularies need ontologies and
  thesauri to achieve their full interoperability
  potential

17
Conclusions

• Semantic modelling works for mappings between
  dictionaries and data models
• It also has great potential for parameter
  discovery and interoperability