The BBC and IBM Corp have signed a strategic alliance that will see them working on several joint projects, the first being the use of an IBM Research system code-named ‘Marvel’ that can visually analyze images and video to automatically categorize the content and make it more easily searchable.
The software is initially being used to analyze and tag the BBC’s CBeebies and CBBC children’s programming. Paul Cheesebrough, controller of digital media at the BBC told Computer Business Review that the BBC has just started trialing Marvel in anger, and that the trial is expected to last until the autumn.
We want to see how accurate the software is and the kind of information it generates, said Cheesebrough. But if all goes well we will be looking at using the technology in other areas of the BBC.
The BBC obviously has a vast archive of material — around 1.4 million hours across all of its channels and programs — and according to Cheesebrough the goal is to make much of that programming more accessible to users, for example visitors to the BBC’s website. For that audience we want to atomize much of our content, he said, by breaking it into shorter components, clipping it up, and presenting it in different ways.
The IBM Marvel search and categorization technology that the BBC is trialing came out of IBM’s labs. It is a web-based technology that makes digital photos and video searchable through automated tagging of visual content. It essentially goes through pictures, video and even audio and adds metadata tags that make that content more readily searchable.
IBM’s Marten Symons, UK global business services practice leader for media and entertainment industries, told us that a number of organizations are currently trialing the software, and that it is yet to be turned into a commercially available product.
Symons said that one of the unique things about the Marvel system is that it learns as it goes along, helping users search very large multimedia content repositories faster and more effectively. But that also means that the software may need some training on a customer’s particular multimedia content before it starts to accurately tag the content.
The BBC’s Cheesebrough said that an example of the value of the software might be in helping young web visitors find clips that specifically feature their favorite presenters, characters, or a particular topic they are interested in, rather than having to watch the entire program to find the bits they want to watch.
We could go through our content and do this manually but we would have to throw lots of people at it and it would involve huge time and effort, said Cheesebrough. We want to see how accurate this technology is at enriching our data and content.
Financial terms of the alliance between IBM and the BBC were not disclosed, but IBM noted that it has a well-established relationship with IBM spanning many years in consultancy, training, development support, and results and analysis for major sporting events.