4. Users
Main group User Target
Researchers &
Educators
Journalists Research
Academic
researchers &
students
Investigate
Academic educators Educate
Public users Citizens Entertainment,
Infotainment
Media Professionals Broadcast
Professionals
Reuse
Media Archivists Annotate
7. 1998 2002 2008
2010 2013 2015
DATA
BIG DATA?
not representative
representative
8. Search & Hyperlinking task
• User oriented: aim to explore the needs of real users
expressed as queries.
– How: UK citizens and crowd sourcing for retrieval
assessment
• Temporal aspect: seek to direct users to the relevant
parts of retrieved video (“jump-in point”).
– How: segmentation, segment overlap, transcripts.
prosodic, visual (low-level, high-level; keyframes)
• Multimodal: want to investigate technologies for
addressing variety in user needs and expectations
– varied visual and audio contributions, intentional gap
between query and multimodal descriptors in content
9. ME Search & Hyperlinking task
in development: 2012 – 2014
Search Hyperlinking
2012 2013 2014 2012 2013 2014
Dataset BlipTv BBC BlipTv BBC
Features released:
Transcripts 2 ASR 3 ASR 2 ASR 3 ASR
Prosodic features no yes no yes
Visual clues for queries yes no no
Concept detection yes yes
Type of the task Known-item Ad-hoc Ad-hoc
Query/Anchors creation PC iPad PC iPad
Number of
queries/anchors
30/30 4/50 50/30 30/30 11/ 98/30
Relevance assessment MTurk users (BBC) MTurk MTurk
Numbers of assessed cases 30 50 9 900 3 517 9 975 13 141
Evaluation metrics MRR, MASP, MASDWP MAP(-
bin/tol),
MAP MAP(-bin/tol), P@5/10
10. Dataset: Video collection
• BBC copyright cleared broadcast material:
– Videos:
• Development set: 6 weeks between 01.04.2008 and 11.05.2008 (1335 hours/2323 videos)
• Test set: 11 weeks between 12.05.2008 and 31.07.2008 (2686 hours, 3528 videos)
– Manually transcribed subtitles
– Metadata
• Additional data:
– ASR: LIMSI/Vocapia, LIUM, NST-Sheffield
– Shot boundaries, keyframes
– Output of visual concept detectors by University of Leuven, and University of
Oxford
11. Dataset: Query
• 28 Users
- Policemen, Hair dresser, Bouncer, Sales manger,
Student, Self-employed
• Two hour session on iPads:
– Search the archive (document level)
– Define clips (segment level)
– Define anchors (anchor level)
Statement of
Information Need
Search
Refine
Relevant Clips
Define
Anchors
12. Data cleaning: Usable Information Need
• Description clearly specifies what is relevant
• A query with a suitable title exists
• Sufficient relevant segments exist (try query)
13. Data cleaning: Process
• For each information need in batch
1. check if usable
2. If in doubt use search to search for relevant data
3. reword & spellcheck description
4. select the first suitable query
5. Save
14. Data cleaning: Usable Anchor
• Longer than 5 seconds
• Destination description clearly identifies the
material the user wants to see when he would
activate the anchor described by label
• It is likely that there are some relevant items
in the collection
15. Data cleaning: Process
• For each information need in assigned batch
– Go through anchors
• check if usable
• reword & spellcheck description
• Assess whether it is like to find links in the collection
(possibly using search)
– Save
18. Ground truth creation
• Queries/Anchors: user studies at BBC:
- 28 users with following profile:
Age: 18-30 years old
Use of search engines and services on iPads on the daily basis
• Relevance assessment: via crowdsourcing on Amazon MTurk platform:
– Top 10 results from 58 search and 62 hyperlinking submissions
– 1 judgment per query or anchor that was accepted/rejected based on
an automated algorithm, special cases of users typos checked
manually
– Number of evaluated HITs:
9 900 for search, and 13 141 for hyperlinking
19. • P@5/10/20
• MAP based:
Evaluation metrics
• MAP: taking into account any overlapping segment:
• MAP-bin: relevant segments are binned for relevance:
• MAP-tol: only start times of the segments are considered:
27. Lessons learned
1. iPad vs PC = different user behaviour and
expectation from the system.
2. Prosodic features broaden the scope of the
search sub-task.
3. Use of shot segmentation based units
achieves the worst scores for both sub-tasks.
4. Use of metadata improves results for both
sub-tasks.
28. Lessons learned
1. iPad vs PC = different user behaviour and
expectation from the system.
2. Prosodic features broaden the scope of the
search sub-task.
3. Use of shot segmentation based units
achieves the worst scores for both sub-tasks.
4. Use of metadata improves results for both
sub-tasks.
29. The Search and Hyperlinking task was supported by
We are grateful to
Jana Eggink and
Andy O'Dwyer
from the BBC for preparing the collection and hosting the user trials.
... and of course Martha for advise & crowdsourcing access.
30. JRS at Search and Hyperlinking of Television
Content Task
Werner Bailer, Harald Stiegler
MediaEval Workshop, Barcelona, Oct. 2014
31. Linking sub-task
• Matching terms from textual resources
• Reranking based on visual similarity (VLAT)
• Using visual concepts (only/in addition)
• Results
– Differences between different text resources
– Context helped only in few of the cases
– Visual reranking provides small improvement
– Visual concepts did not provide improvements
35
32. Zsombor Paróczi, Bálint Fodor, Gábor Szűcs
Solution with concept enrichment
• Concept enrichment: the set of words is
extended with their synonyms or other
conceptually connected words.
• Top 10 vs top 50 conceptually connected words
for each word
• Conclusion: the results show that concept
enrichment with less words give better precision
because at the opposite case the noise is greater.
33. Television Linked To The Web
LinkedTV @ MediaEval 2014
Search and Hyperlinking Task
H.A. Le1, Q.M. Bui1, B. Huet1, B. Cervenková2, J. Bouchner2, E. Apostolidis3,
F. Markatopoulou3, A. Pournaras3, V. Mezaris3, D. Stein4, S. Eickeler4, and M. Stadtschnitzer4
1 - Eurecom, Sophia Antipolis, France.
2 - University of Economics, Prague, Czech Republic.
3 - Information Technologies Institute, CERTH, Thessaloniki, Greece.
4 - Fraunhofer IAIS, Sankt Augustin, Germany.
16-17 Oct 2014
www.linkedtv.eu
34. Reasons to visit the LinkedTV
poster
• Different granularities: video level, scene
level (visual/topic) and sentence level.
• Different features: text (subtitles /
transcripts), visual concepts, keywords, etc…
LinkedTV @ MediaEval 2014 Search and Hyperlinking Task
35. Reasons to visit the LinkedTV
poster
• How to incorporate visual information
to the search?
• Visual concept detection in the search
query:
Mapping between query keywords and visual
concepts (151 semantic concepts from
TRECVID 2012)
– Semantic word distance based on WordNet
– Identification of salient visual concepts from
Google Image search results (query keywords)
LinkedTV @ MediaEval 2014 Search and Hyperlinking Task
36. Reasons to visit the LinkedTV
poster
• How to incorporate visual information
to the search?
• Integration of detected visual concepts
to the search:
– Designing an enriched query, based on textual
(text query) and visual information (range
query)
– Fusion of text score (Solr) and visual concepts
scores
LinkedTV @ MediaEval 2014 Search and Hyperlinking Task