36
ROUND TABLE Content Management
have reached some level of maturity for general purpose indexing, although they’re not ready out-of-the-box for every type of content. To be efficient with this automatic analysis of content, you need to fine-tune and train your AI models to analyse your specific type of content. We’ve recently been working with a radio station archive. We took a couple of hundred hours of their content and catalogued it manually, then ran that through an AI tool that learned from that content. After fine-tuning the content analysis, we got some very good results, including with some very old content in specific languages, with poor quality of audio. GOODY GROUP: McBoolean was obsessed with boxing. Among the collection are many hours of boxing footage, including film and tape of material dating back to 1945. He also had collections around his other great obsessions, like Soviet-era animation. These collections are virtually uncatalogued. Some are simply cargo crates labelled ‘Training – Dempsey, et al’ or ‘Leningrad Fest 1952?’ How can we find out what is in these collections and what are your ideas about how to best monetise them? VIVEK KHEMANI: Once we have tagged the collection with relevant metadata, it can then be annotated with ‘micro-moments’. A micro-moment is a combination of players, locations and audio high point tags that can be best monetised. For instance, identifying Dempsey (player) in footage along with a boxing punch (object), with only one person in the ring (locale) can lead to identification of moments when he is training. This type of tagging can then be utilised to search the catalogue efficiently for creating a documentary or other content around Dempsey. THE STARTING POINT OF THE TAXONOMY SHOULD BE THE USER SEGMENTS THAT YOU WANT TO TARGET
Or, the entire catalogue can be opened for search by general users. The search log can then be used to identify micro-moments that can be monetised. IAN MOTTASHED: I’m a sucker for a good sports documentary. Assuming you have the rights, then pitch the collection of boxing assets to a media and entertainment production company. Look for the story – or let them find one. Give them a good deal on the licencing. If you can get a decent feature film out of the archive and get it distributed, then you’ve struck gold – and that will provide great publicity for the rest of your content. If the boxing doc works out, you can then hire more people to work out what to do with the Russian stuff! KEVIN SAVINA: Again, if you can supply some content-specific machine learning, you can get some good, efficient results in identifying and cataloguing content. But that should be the result of analysis about the commercial benefit of that content. When you have these very specific collections, the challenge will be to identify a market for them. It’s defining who will be interested in that content. GOODY GROUP: The collection is so eclectic and unwieldy, we are having trouble getting our heads around the best ways to use it. How can we
tease out themes, insights and monetisation opportunities? In some ways, it’s too much of a good thing. How can we focus? VIVEK KHEMANI: The place to begin is to create a taxonomy that can form a ‘sitemap’ to this content store. Segmenting the content at a high level and then creating a hierarchy of concepts can help. AI-generated metadata can be used to populate this index and map content to each of the nodes of this taxonomy. To ensure that we are focusing on monetisation, the starting point of the taxonomy should be the ‘user segments’ that you want to target with this content – such as broadcasters, content production houses, sports leagues, et al. IAN MOTTASHED: Do your best to work out what you think is valuable first – personalities, historic moments. Digitise that and make it available online. Meanwhile make your full catalogue available as a data-only search – assuming
feedzine feed.zine feedmagazine.tv
Powered by FlippingBook