MediaLoep combines documents readily available within the broadcasting company (subtitles, news preparation, ...) with semantic web technology to create a powerfull media search application.
Presented at EBU Production Technology Seminar 2011
2. VRT is the
Flemish Public Broadcaster
3 TV-channels, 5 radio channels
VRT-medialab is the
research department
creation, distribution and
management of media content
Vrt-Medialab 2
3. Lots of audio and video material illustrating our
cultural heritage.
Also includes new material (news clips, …)
Used by programme-researchers & journalists
VRT-medialab 3
4. The problem of media search
MediaLoep project
Re-using production metadata
Linking to the semantic web
Vrt-Medialab 4
5. The problem of media search
MediaLoep project
Re-using production metadata
Linking to the semantic web
Vrt-Medialab 5
6. Not self-descriptive → we need metadata
Series: Flikken
Keywords: violence, robbery
Description: Robbery on
shop. Attacker hits shop
owner with gun.
Video / Audio are continuous media
with a time-dimension
VRT-medialab 6
7. Not self-descriptive
Video / Audio are continuous media with a
time-dimension
→ we prefer time-coded metadata
00’00”>01’43” 01’43”>04’20” 35’00”>36’33”
Robbery on shop Police agent Observation by
looks worried police
VRT-medialab 7
11. Not enough detailed annotations available
◦ “X spits on the ground after Y makes a goal”
◦ The entire dialogue so we can search for quotes
◦ Labels, locations, links, maps, photographs, …
as the creation of these annotations is very
time consuming.
Vrt-Medialab 11
12. The problem of media search
MediaLoep project
Re-using production metadata
Linking to the semantic web
Vrt-Medialab 12
14. The problem of media search
MediaLoep project
Re-using production metadata
Linking to the semantic web
Vrt-Medialab 14
15. News Rundown with auto-cue texts, overlay
labels, …
EPG data contains a summary of the
programme, the broadcast dates, …
A drama script contains dialogues and
actions, …
Subtitles ~ transcript of spoken text
Vrt-Medialab 15
16. Information added by an archivist
keywords
textual description
other fields
Vrt-Medialab 16
17. Information added by the news preparation:
overlay captions
autocue text
links to other items in
this news broadcast
Vrt-Medialab 17
18. Information added by the subtitles:
time-coded transcript
of the dialogue
Vrt-Medialab 18
20. The problem of media search
MediaLoep project
Re-using production metadata
Linking to the semantic web
Vrt-Medialab 20
21. Archivists add thesaurus keywords to clips
Geneva
Obama, Barack
Europe
…
By linking these keywords to a thesaurus, we
can make the search system smarter
Geneva → coordinates on a map?
Obama, Barack → a picture?
…
Vrt-Medialab 21
22. Public knowledge bases provide information
about resources using ‘triples’.
Geneva country Switzerland
area
15.86 km2
Vrt-Medialab 22
23. Links to the same resource in other
knowledge bases can be created.
Geneva country Switzerland
sameAs
area
15.86 km2
Geneva
latitude 46° 12' 0" N GeoNames
Vrt-Medialab 23
24. A network of linked knowledge is created.
Vrt-Medialab 24
25. We linked MediaLoep to DBpedia, which is in
turn linked to many other knowledge bases.
MediaLoep
Vrt-Medialab 25
26. AALMOEZENIER
AALST
AALTER
DE WEVER, BART
GENEVA
…
VRT-Thesaurus DBpedia / Wikipedia MediaLoep
Vrt-Medialab 26
29. Improved search by combining existing
information.
Enhanced results visualization and semantic
query suggestions by coupling to the
semantic web.
Vrt-Medialab 29