three pillars

  1. metadata modelling
  2. metadata workflow: production, consumption, generation (mostly multimedia)
  3. experiments and conclusion

Use Case: the MICO Project

goal: mash-up a super-cute dog video

semantic search would be awesome: Kunststueck,

Remember: Tiere erkennen ist schwer! (Snapshot Serengheti)


  • Video Analyzer
  • Audio Analyzer
  • Text Analyzer
  • Tag Analyzer

Drawback: huge data to pre-train

Outcome: metadata background for dog video as RDF


knowledge statements are represented as triples:

<subject> <predicate> <object>

Representation as Knowledge Graph Query language: SPARQL

question: dual graph

Web Annotation Data Model (WADM)

W3C-standard for the re-use of ontologies; these contain restrictions

Why RDF?

Because , otherwise relational data base

MICO Metadata Model (MMM)

  • combine all extractor data

This is not learning – this is knowledge representation!

How to evaluate?