Tech-Forward - Achieving Business Readiness For Copilot in Microsoft 365
HyperTED: exploring video lectures at the fragment levels for enhancing learning
1. HyperTED: Exploring Video
Lectures at the Fragment Levels
for Enhancing Learning
Raphaël Troncy <raphael.troncy@eurecom.fr>
Data Science, EURECOM
@rtroncy
4. Once upon a time …
21/10/2019 - Workshop on Search as Learning with Multimedia Information - 4
5. … leading to sharing Media Fragments
Publishing status message containing
a Media Fragment URI
Use a ‘#’ !
Highlight a
video
sequence
Highlight a
region
to pay
attention to
21/10/2019 - Workshop on Search as Learning with Multimedia Information - 5
6. Making video a "first class citizen"
21/10/2019 - Workshop on Search as Learning with Multimedia Information - 6
7. t0 20 35
temporal media fragment
spatial media fragment
track media fragment
named media fragment“Scared Scene”
What are Media Fragments?
21/10/2019 - Workshop on Search as Learning with Multimedia Information - 7
8. Media Fragments (temporal)
Fragment beginning Fragment endPlayback progress
Original resource
length
21/10/2019 - Workshop on Search as Learning with Multimedia Information - 8
10. Media Fragments URIs
Bookmark / Share parts (fragments) of audio/video
content
Annotate media fragments
Search for media fragments
Develop Mash-ups/Collage
Conserve bandwidth
http://www.w3.org/TR/media-frags-reqs/
http://www.w3.org/TR/media-frags/
21/10/2019 - Workshop on Search as Learning with Multimedia Information - 10
12. New Consuming Paradigm
Users
overwhelmed
with audio-visual
content
What are the
potentially
relevant
fragments ?
How can users easily find
related documents which
complement the video
Can the video be
divided into
meaningful
fragments?
How can those
fragments be
properly
described?
13. Media Fragment support
Chapters
Hot Spots
Media Fragment annotations
Named Entity Extraction
Topic Detection
Hyperlinking
With TED talks chapters
With other educational online resources
HyperTED
14. http://www.w3.org/TR/media-frags/
A Media Fragment is a portion of a multimedia resource
Temporal Fragments
sections along the time dimension of the media
resource with a start and an end point
Media Fragments
15. TED Talks have paragraphs:
a human-made subdivision of subtitles
MF: Chapters
17. “This is Nikita, a security guard from one of the bars in St. Petersburg.”
“This is Nikita, a security guard from one of the bars in St. Petersburg.”
NER
Example taken from the transcript of
https://www.ted.com/talks/2089
PERSON
FUNCTION
LOCATION
Category:
type in the NER task
Natural Language Processing (NPL) Task
disambiguating URL in a knowledge base
e.g. https://www.wikidata.org/wiki/Q656 or
http://dbpedia.org/resource/Saint_Petersburg
Annotations: Named Entities
20. Mobile computers
Annotations: Topics
“I'm wearing a camera, just a simple webcam, a portable, battery-powered
projection system with a little mirror. These components communicate to my
cell phone in my pocket which acts as the communication and computation
device. And in the video here we see my student Pranav Mistry, who's really the
genius who's been implementing and designing this whole system...”
Battery (electricity)
Consumer electronics
Example taken from the transcript of
https://www.ted.com/talks/pattie_maes_demos_the_sixth_sense
Chapter 3
21. 1. Clustering of consecutive chapters which talk
about similar topics and entities
2. Ordering of those fragments based on
annotation relevance (TF-IDF)
3. Filtering: Hot Spots are fragments whose
relative relevance falls under the first quarter of
the final score distribution
MF: Hot Spots
Hot Spot 1
Chapters
Hot Spot 2
Hot Spots
22. 1. Clustering of consecutive chapters which talk about similar topics and entities
2. Ordering of those fragments based on annotation relevance (TF-IDF)
3. Filtering: fragments whose relative relevance falls under the first quarter
MF: Hot Spots
24. • Topics
• Entities
• time code references (startNPT and endNPT)
• extractor confidence
• Resource identifier
• Full text transcript
Granularity level
• Chapter
Features indexed
Hyperlink: Indexing TED Talks
25. Datasets
• Open Courseware
• Open University
Anchors used in search
• Entities Too specific
• Topics Courses about the same thematic
Attributes used in search
• Title
• Description
• Subject, thematic …
Hyperlink: Finding related courses
30. MediaMixer Demonstrator (CERTH-ITI)
Video lectures shot segmentation
Concept detection adapted to
video lectures (37 concepts)
http://multimedia.iti.gr/mediamixer
/demonstrator.html
21/10/2019 - Workshop on Search as Learning with Multimedia Information - 30
32. Let's go even back in time [AIED, 1999]
21/10/2019 - Workshop on Search as Learning with Multimedia Information - 32
Sassine Abou-Jaoude, Claude Frasson, Olivier Charra and Raphaël Troncy.
On the Application of a Believable Layer in ITS. In (AIED'99) Workshop on Synthetic
Agents, Le Mans, France, July 19, 1999
34. Summary / Take Away
Trove of learning content buried in videos
need tools to segmentate / annotate / discover this content
We developed HyperTED in 2014!
the concept is still original
Natural Language Processing (information extraction)
deep learning-based named entity extractors
word and entity embeddings (multilingual, multimodal)
watch ADEL: https://github.com/jplu/ADEL
watch entity2vec: https://github.com/D2KLab/entity2vec/
Recommender systems … at the fragment level
deep learning architecture, using KG embeddings
21/10/2019 - Workshop on Search as Learning with Multimedia Information - 34