The document discusses using AI to support reuse of archival audiovisual content at scale. It describes the ReTV project which aims to automatically adapt broadcaster content for digital platforms through techniques like video summarization. The project also explores personalizing access to archives through a chatbot and making content more discoverable in media asset management systems through generous interfaces that allow retrieval by concepts. The goal is to generate more value from archival collections by bringing content to audiences on digital channels.
Similaire à Content Adaptation, Personalisation and Fine-Grained Retrieval: Applying AI to Support Engagement with and Reuse of Archival Content at Scale
Similaire à Content Adaptation, Personalisation and Fine-Grained Retrieval: Applying AI to Support Engagement with and Reuse of Archival Content at Scale (20)
2. retv-project@ReTVproject@ReTV_EUretv-project.eu
ReTV: Reuse of Audiovisual Archives
Fundamental need:
- Generate value out of archival AV content through reuse; nothing
good comes out of just keeping the content locked in your digital
basement
- Reusing content with impact - bringing content to where your
audiences are
- Archives need to be visible and present on digital channels
2
4. retv-project@ReTVproject@ReTV_EUretv-project.eu
Reuse in Audiovisual Archives
4
This is where AI steps in!
- Understanding AV content for
meaningful curation
- Making AV content discoverable
on a granular level
- Automatically adapting
broadcasted content and
publishing it on digital platforms
7. retv-project@ReTVproject@ReTV_EUretv-project.eu
ReTV Use Cases
Audiovisual Content Reuse
1. Broadcaster Content Adaptation
- Reusing Videos on Social Media
1. Personalised Access to the Archive
- Personalised Content Delivery via a Chatbot
1. AV Content Discovery & Retrieval
- Support Reuse through Generous Media Asset Management Systems
7
8. retv-project@ReTVproject@ReTV_EUretv-project.eu
Broadcaster Content Adaptation for Social Media
Video Summarisation
8
- Fast-paced consumption - no time for full-length
programmes
- Target distribution platforms & devices have varying
requirements (e.g. optimal duration that is not suitable for
full-length broadcaster content)
12. retv-project@ReTVproject@ReTV_EUretv-project.eu
Requirements:
- Different training for different content (e.g. separate trained models can be
used for different TV programmes)
- Action-packed shots (no talking heads shots)
- BUT no spoilers
- Audio track analysis to supplement visual analysis
12
Broadcaster Content Adaptation for Social Media
Video Summarisation
14. retv-project@ReTVproject@ReTV_EUretv-project.eu
Personalisation for Archives
Chatbot
- Exploring what personalisation means for archives
- How to bring content to your audiences and keep them interested?
Possibilities for long-term engagement via personalised content
Archival Chatbot
- Bringing content directly to audiences via channels convenient to them
- Personalisation to increase the reuse of AV content - get that niche
content out there!
14
16. retv-project@ReTVproject@ReTV_EUretv-project.eu
Personalised Access to the Archive
Chatbot
Lessons learned - rethinking personalisation
- great tool for fostering more personal engagement with the past
- expand what people are interested instead of narrowing their interests down -
not only content that you are already familiar/interested in but helping to
discover something new
- make personalisation transparent - let users understand and be in control
(through direct interaction via chatbot)
16
18. retv-project@ReTVproject@ReTV_EUretv-project.eu
AV Content Discovery & Retrieval
Generous Media Asset Management Systems
- boost the reuse of content through MAMs by producers, artists,
documentary makers, amateur creators
- retrieval constrained by high-level descriptions - you need to know
what you are looking for
- retrieval through a generous interface for AV content
18
20. retv-project@ReTVproject@ReTV_EUretv-project.eu
AV Content Discovery & Retrieval
Generous Media Asset Management Systems
Why go generous?
- more intuitive, serendipitous search
- discovery of rarely used content
- overview at a glance on an item or collection level
- granular access to specific segments and concepts
20
24. retv-project@ReTVproject@ReTV_EUretv-project.eu
AV Content Discovery & Retrieval
Generous Media Asset Management Systems
Challenges for being generous with AV content
- meaningful concept annotation
- scalability
- putting all the pieces together - video analysis, speech to text, face
recognition, etc.
24
25. retv-project@ReTVproject@ReTV_EUretv-project.eu 25
This work was supported by the EUs Horizon 2020
research and innovation programme under grant
agreement H2020-780656 ReTV
Rasa Bocyte, NISV
rbocyte@beeldengeluid.nl
@rasa_bocyte
Johan Oomen, NISV
joomen@beeldengeluid.nl
@johanoomen
Notes de l'éditeur
add partners
use cases
REST service available (for integration in applications / CMSs)
Levuro screenshot
Audio - not lose contextual info to contextualise the story
So we wanted to see how this translates to archival collections and the mission that CHIs have.
MAM interface
temporal fragmentation of a video (subshots/shots/scenes)
fragment annotation with concept labels that describe them (>6000 labels), text to video matching.