In this episode of Destination Cloud, our hosts Simon McGrath and Alex Borland are joined by Geert Vos, CTO and founder at Media Distillery, and Aidan van Latestein, Principal Senior Architect at Merapar, to discuss all things metadata.
From the origination of metadata to the content management systems along the supply and value chain, this episode takes a deep dive into the benefits metadata is already providing, and the potential it has to improve the overall video consumption experience in a new, freeform and elastic world.
Inside the episode
The episode starts with the difference between technical metadata, which describes the video, audio and image assets, and descriptive metadata, from title and synopsis to moods and cast. Geert Vos explains how Media Distillery began: detecting programme start times for an EPG from the video itself. Aidan van Latestein describes the other end, a CMS aggregating linear and VOD metadata from many sources, in several languages.
From there it moves from fixed structure to generated metadata, produced when it is needed, and to back ends where the UI drives the data model. It ends on chapters and clips, and on Simon McGrath’s point that every clip cut from a programme still has to carry its rights and regional restrictions back into the CMS.
Highlights
- 08:13 “You need really, really good metadata to be able to drive that user experience.” Geert Vos
- 09:50 “You always end up in this mapping nightmare.” Aidan van Latestein
- 17:03 “By the time we have the standard, it might already be outdated.”
- 21:16 On one recent project, 90% of the customer’s hits came from Google, so the metadata work went into tagging pages for search engines.
- 22:15 Large language models as the new cloud: from elastic compute to “elastic intelligence”.
- 27:42 Chapters as time-based rather than asset-based metadata, so a ten-minute segment of a one-hour episode can be served as a clip of its own. Geert Vos
Key questions answered
Why does metadata matter so much to the viewer?
Because platforms are fighting for eyeballs. Geert Vos says the user experience is what keeps viewers on a platform, and it takes very good metadata: viewers can spend half an hour looking for the right content. Knowing the viewer and knowing what the content is about, and bringing the two together, is what delivers the experience.
Why is aggregating metadata from many sources so hard?
Aidan van Latestein’s project met four or five data models. Titles and descriptions map easily; genre mappings, multiple languages and a stream of updates do not, as with a dubbed Disney film whose poster is in another language. Deciding that two catalogue entries are the same film needed people: the system flagged a conflict with an educated guess, and an operator gave the final verdict. For genres, themes and moods, Aidan finds that everybody invents their own standard.
Should metadata be structured, or generated when needed?
Geert’s view is that taxonomies existed so that machines could categorise content, but machine learning now understands content more generically and can map it to any structure, such as IAB categories for advertising. So metadata may become less structured, with a label generated when the interface needs one. In one proof of concept, viewers subscribed to any topic, ice skating say, and were told when a new video covered it; adding ice hockey changed almost nothing in the system.
How should content management systems change?
In Aidan’s experience the UI drives the metadata requirements, and only in the last couple of years has UX really taken the lead. On Aidan’s latest project the customer wanted to change its front end at will, so metadata was stored as large, loosely structured documents with tags in DynamoDB and served in whatever shape the apps asked for. There is also the story of a back end and apps built in two months, where the first thing the customer’s CTO saw in the demo was a competitor’s game show.
What can chapters do for a service?
Geert describes chaptering long programmes such as talk shows, news and documentaries automatically, each chapter with its own title, description and time codes, so a platform can serve it as a clip for social media or a mobile app. The same use case comes up in advertising, where short clips draw viewers to the full programme, though there it had been done by hand.
Speakers
- Geert Vos, CTO and founder, Media Distillery
- Aidan van Latestein, Principal Senior Architect, Merapar
- Simon McGrath, Merapar (host)
- Alex Borland, VP Strategy & Business Development, Merapar (host)
