Wikimedia Conference 2010/Developers' Workshop/Notes/Structured Data

This pad contains live notes from the Structured Data working Group at Wikimedia Conference 2010/Developers' Workshop.

Outline

The first session was used to identify main topics to discuss today and tomorrow. The resulting topics are:

• Using MW templates for managing structured data (do we need better declarations of template parameters? datatypes? data extraction from templates, ...)
• Data import/re-use from external sources (live import vs. data integration, caching, push vs. pull, versioning issues, trust and provenance)

Participants

(based on introductory round in the morning)

• Tatiana de la O (acracia): API, RDF storage
• Gregor Hagedorn: image metadata (?), large datasets, (Identification in biology, matrix keys for querying data)
• Simon GESIS: wiki for social science databases (import/export)
• Leszek Krupinski (leafnode): importing metadata properly, database access to wikis
• John Erling Blad (jblad): using external data in Wikipedia, re-using public data
• Daniel Kinzler (Duesentrieb): multi-lingual metadata, commons, metadata extraction
• APPER (Chris): PersonData, tool server
• Lars Aronson (LA2): practical use of metadata, personal metadata
• Jonathan Gray (OKF): Open Data, Open Content, bibliographic metadata, browsing
• Inez: structured data extraction, article recommendation, (WYSIWYG background)
• Sebastien: Automatic Wikification
• Anja Jentzsch (anjeve): DBpedia
• Robert Isele: DBpedia
• Kolossos: Maps, Template Tiger
• Markus Krötzsch: Semantic MediaWiki
• Jeroen De Dauw: Semantic Maps

Use Cases

• Image metadata on Commons
• PersonenDaten on German WP
• Geodata on WP
• Bibliographical records (e.g. using FRBR ontology)
  • online demo at http://bibliographica.org/

Using MW templates for managing structured data

• Should we use templates, or develop some other way of presenting/recording meta-data?

How to manage metadata about templates?

• Proposal by Daniel:

Declare template paremeters, including documentation, expressed relation (e.g. RDF property), optional-flag, etc.

• Proposal by Markus:

SMW already has a feature like the one proposed by Daniel. In addition, SMW declares properties on separate pages to have a local name and datatype (instead of just using a technical URI of some external ontology directly)

• Properties should be first-class objects (like in RDF), existing globally and possibly being used in more than one template
• Some property values can have multiple languages
  • we want to support this only for plain text values
  • we treat it as in RDF internally
• Problem: not all properties have reasonable one-to-one mappings to template fields, e.g. sometimes multiple fields have to be consolidated into one property value
• Possible solution 1: Have a parser function for declaring template fields to have some "meaning", do this with an extension and incrementally introduce it to a WP project; actual values are obtained by hooking into template transclusion and checking if the template has declared meanings for its parameters to process them (advantages: no changes in core, incremental adoption/extension possible)
• Possible solution 2: Have a parser function for processing instantiated template parameters; the parser function is inserted into the template code (wrapped around the value) and processed on the pages the template is used on (advantages: no addition database lookup when using templates)

Which information to declare?

Core information:

• parameter name
• datatype
• unique identifier for the property (possibly from a standard vocabulary, or from the wiki)
• human-readable documentation

Auxiliary information:

• field required or not (used for editing)
• information for sanitizing inputs
• list-related attributes (e.g. separators for lists)
• ...

Basic datatypes:

• text, multilingual text, dates, numbers, wiki page names, geo coordinates, URLs
• lists of <anything>
• Units of measurement?

Other issues (later)

• Data model?
• Data types?
• Mapping to external ontologies?