As the guardian of the French television, radio, and web media archives, the INA is responsible for the proper conservation of the French audiovisual heritage, its safeguarding, and its promotion to the general public, researchers, and professionals. At the time of its creation, the French audiovisual landscape only included a handful of channels that were exclusively public. Since then, the INA’s scope of action has considerably expanded, as has the volume of broadcasting flows to be processed (the INA captures 184 TV and radio channels every day, and collects content from more than 15,000 social network accounts, more than 10,000 video platform accounts, and more than 5,000 podcast collections). However, in order to carry out its mission, the Institute has to deal with a technological legacy made up of different solutions that were implemented successively, in doing so gradually locking up data in “sealed” systems and thus requiring incessant manual synchronization from one database to another. This situation led, in 2014, to the implementation of a data governance strategy, based on the overhaul of its information system. This approach, aimed at centralizing and harmonizing systems, has enabled the company to initiate various data production automation projects.
The documentary processing of media in the legal deposit currently relies on data acquired from partners, such as broadcasting companies or some agencies specialized in media monitoring.Yet these data, provided by channels, are sometimes incomplete and above all forecasted, which implies numerous verifications. In this context, the INA has been particularly interested in automating the cutting and cataloguing of TV and radio listings. The objective was to get rid of the existing technological “millefeuille” and its inherent problems (non-harmonized data, duplicates, etc.) to rely on an intelligent solution that would combine different tools. This solution had to identify the programs, but also reconstitute the daily program listings to facilitate the cataloguing and description of the collections.
The automation aims to gain autonomy in the reconstitution of program listings, to better promote the Institute’s data, and to participate in the implementation of a unified and detailed access to broadcasting archives.
The INA chose to use Deepomatic’s platform to build and train its own image classification models, in order to perform a first breakdown of the programs per day. The Institute was convinced by Deepomatic’s promise: business teams could quickly become totally autonomous thanks to its solution, without having to systematically use the skills of a data scientist.
This solution first allowed for a segmentation of newscasts, automatically distinguishing between on-set images and images of reports, credits, or weather forecasts. These first results made it possible to consider the analysis broadcasting days of 24H rolling news channels. In this context, the image classification offered by Deepomatic allows not only to distinguish “universal” elements such as the sets, but also “specific” elements, such as the set of a given program. Currently, the INA is planning to use Deepomatic on music variety shows and talk shows, which have more complex structures, alternating on-set passages, artistic performances, reports, and interviews.
The benefits of Deepomatic for the INA can now be seen mainly on 3 levels:
Founded in 1975, the National Audiovisual Institute (INA) is a public broadcasting and digital company whose main missions consist in collecting, archiving, and promoting the French televisual heritage. In order to constantly innovate their users’ experience, the INA promotes its contents and shares them with as wide an audience as possible. Thus, since 1995, its two-fold mission is divided between managing, preserving, and marketing public television archives on one side and, for the sake of cultural preservation, collecting and consulting film, TV, and radio legal deposits on the other.