Agenda
- Motivation
- Activities
- Interlinking Multimedia (iM)
- Challenges
Activities
- research projects
- W3C
- other activities
Activities - Research Projects I
Knowledge Space of Semantic Inference for Automatic Annotation and Retrieval of Multimedia Content (K-Space), FP6 NoE, 2006 - 2008
- VAMP, a semantic validation service for MPEG-7 profiles
- COMM, a core ontology for multimedia (based on MPEG-7 and DOLCE)
New Millennium, New Media (NM2), FP6 IP, 2004 - 2007
Activities - Research Projects II
Understanding Advertising (UAd), Austrian national FIT-IT, 2007/2008
- riese, linked data version of Eurostat data
- UCI demonstrator I R S, linked data version of Eurostat data
- discussion tracer (Firefox plugin), analyser
Other Activities - ramm.x
, RDFa-deployed multimedia metadata
details see specification ...
Other Activities - ramm.x
ramm.x client: processing
Other Activities - ramm.x
use cases
- annotate and share photos
- buy music
- description of video structure
- publishing professional content with metadata
- rights information for media asset (DRM)
- detailed description of large media assets
- cultural heritage applications
Other Activities - outreach, tutorials, etc.
Interlinking Multimedia (iM)
Why?
- based on our analysis ...
- from end-user's point-of-view multimedia content is easy to produce, store and share
- from professional user's point-of-view multimedia content requires lot of manual work (editing, distribution, etc.)
- to this end, metadata (or even ontologies) have not been a big success story w.r.t. multimedia
- some tasks (such as summarisation, surveillance, etc.) are hard if not impossible without solid metadata
Interlinking Multimedia (iM)
Ok. But Why iM?
- currently mainly textual resources interlinked (have a look at the LOD cloud)
- to date only global interlinking of multimedia content (cf. BBC)
- we need more to reuse existing multimedia metadata and use linked data
Interlinking Multimedia (iM)
An Attempt to Resurrect Hypermedia?
This is not about trying to resurrect hypermedia (it's dead, remember? ;) rather, iM
is a light-weight bottom-up approach to interlink multimedia content on the Web of Data. Basic building blocks are:
- the linked data principles
- common (multimedia) vocabularies, such as Exif, ID3, etc. - possibly along with ramm.x
- set of interlinking methods, such as User Contributed Interlinking UCI (see also human computation)
Interlinking Multimedia (iM)
CaMiCatzee I
Interlinking Multimedia (iM)
CaMiCatzee II
system architecture (more ...)
Interlinking Multimedia (iM)
Current Possibilities and Limitations
annotating youtube videos
Interlinking Multimedia (iM)
Current Possibilities and Limitations
BBC music beta site
Challenges
- discovery and usage of datasets (voiD)
- performance & scalability
- quality
- privacy & trust
- partners & use cases
Challenge - Discovery and Usage of Datasets
- voiD, the vocabulary of interlinked datasets
- light-weight vocabulary to describe datasets and their interlinking
- work started at ESWC08, involved are people from DERI (Richard), Talis (Keith), University of Oxford (Jun), and myself
- currently we work on the guide and the core vocabulary
Challenge - Discovery and Usage of Datasets
usage of voiD in a distributed setup
Challenge - Discovery and Usage of Datasets
:DBpedia a void:Dataset ;
foaf:homepage <http://dbpedia.org/> ;
dc:subject <http://dbpedia.org/resource/DBpedia> ;
void:containsLinks :DBpedia2DBLP .
:DBLP a void:Dataset ;
foaf:homepage <http://www4.wiwiss.fu-berlin.de/dblp/all> ;
dc:subject <http://dbpedia.org/resource/Computer_science> ;
dc:subject <http://dbpedia.org/resource/Journal> ;
dc:subject <http://dbpedia.org/resource/Proceedings> .
:DBpedia2DBLP a void:Linkset ;
void:target :DBLP .
Challenge - Discovery and Usage of Datasets
SELECT ?dataset {
?dataset a void:Dataset;
dc:subject <http://dbpedia.org/resource/Proceedings> .
?datasetSrc a void:Dataset;
foaf:homepage <http://dbpedia.org/> ;
void:containsLinks ?linkset .
?linkset void:target ?dataset.
}
demo ...
Challenge - Performance & Scalability
- currently, we have global descriptions and interlinking of multimedia assets
- iM is all about fine-grained (spatio-temporal) interlinking of multimedia assets
- whatever solution we're after for linking multimedia assets to other datasets, it MUST perform well and scale to the size of the Web
Challenge - Quality
- nowadays, mainly automatic approaches for interlinking are used yielding rather low-quality interlinks
- UCI might be a good choice, but still needs serious research, esp. regarding motivation
- whatever solution we're after for linking multimedia assets to other datasets, it MUST yield high quality interlinks
Challenge - Privacy & Trust
- linked data consumers (e.g., user agents or syndicators) rely on the ability to determine which information is trustworthy for a certain topic, task, etc.
- current technologies such as OpenID, OAuth or FOAF-based whitelisting are just a first step
- whatever solution we're after for linking multimedia assets to other datasets, it MUST address privacy issues properly and enforce trust in the so created interlinks
Challenge - Partners & Use Cases
- plenty of material is available; two possible streams are:
- professional content (media industry, broadcaster archives, digital cinema, digital libraries, etc.)
- semi-professional and prosumer content (flickr, YouTube, Last.fm, MS Live, etc.)
- idea: start with something rather simple (e.g., flickr Exif metadata + UCI ?) and demo to potential partners such as BT, BBC, RAI, etc.
The End?
Let's start the work!
