EXPLORING THE AI FRONTIER IN THE MEDIA DOMAIN
Nowadays Artificial Intelligence (AI) is a hot topic, not only in IT but across all sectors of our society. The applications of AI range from tiny apps like tech support robots (chatbots), to paradigm-shifting cutting-edge tech like self-driving cars. The reach of AI is not just limited to a few niche areas. AI techniques and software solutions based on AI algorithms are being used in various domains including the media.
Who we are
Before I go into detail, let me give you a short introduction to our team. We are AI Platform Services, developing software solutions powered by AI, in one of the largest privately-owned media groups in Germany, ProSiebenSat.1 Media SE. We are a small team of 6 AI enthusiasts comprising of product owners, AI engineers and architects. We would like to shed some light on the topics we work upon in a series of articles aimed at folks who are curious to understand or get into the field of AI and those who are already experts in the field and want to collaborate with us in some form.
In this article, I will try to provide you with a comprehensive yet brief overview of the nature of the projects that we work on. The scale of our projects ranges from simple proofs of concepts (PoC) to distributed and scalable software applications running on a hybrid cloud infrastructure. We build real innovative products with real world implications (The German way of doing things ;)) and not just experiment to fulfil our AI fantasies.
Our role in the company
Our primary goal is to empower and support all the different business units within ProSiebenSat.1, so that content creators, editors, media documentalist and sales representatives, just to name a few, can perform more efficiently and successfully. We try to support them by automating certain tedious and repetitive tasks so that they can focus on things that actually matter. We achieve this by developing customised creative solutions, catered to the needs of our clients, and by employing the latest tools and technologies. Now, let’s look at some of our on-going projects.
What we do
Improving the Ad-Experience
Ads have become an integral part of everything online and offline. As a private media group, ProSiebenSat.1 heavily relies on ads (just like Google!). But you might ask, can ads be more relevant to you? Is it necessary for the ads to pop up in the middle of your favourite star’s punch line? Well… with some effort and some AI magic, we can fulfil all your wishes and, apparently, we are already working on it. We are building models that make ads more relevant and less intrusive. Cloud services like AWS Transcribe and AI techniques such as natural language processing help to understand what is happening within the content you are watching and enable us to transform/extract features that we could act upon. This generated knowledge is used to propose suitable ads related to the content of a video as well as suitable timestamps to place an ad, so that you are no longer disturbed while e.g. Matthias Opdenhövel is announcing the next “Masked Singer”. More importantly, these methods do not require us to use your personal data.
Content documentation and search
Humans are always looking for something. There are always many options to choose from. This is also true for digital content. Suppose content creators/editors are creating a new video clip and they want to re-use scenes from already existing videos. For that they need to look through thousands of videos to find one with (let’s say) a cat running around or one with a popular celebrity dancing. The simplest approach would be to watch as many videos as possible until they find something that could be used. I needn’t point out that this approach is not only extremely inefficient but also unfulfilling. So, are there no alternatives? Are there no efficient approaches? Of course, there are, and this is where we come in. On the one hand, we are automating the documentation of content and on the other, we provide a solution to find the content easily. This is done by generating relevant metadata for video-content in the form of text, allowing editors to search the textual data to find the corresponding video. We achieve this by extracting pieces of information such as objects (cat, car, cake etc.), actions (dancing, kissing etc.) and celebrity information from the videos. In other words, by using AI video mining techniques, we generate descriptive metadata (visual and audio tags) for videos at frame level and aggregate them to provide tags at scene/video level. Using associated information and search options, the editors can search for relevant content in few seconds.
Process Automation and Consulting
Although the media industry improved a lot over the last 10 years in terms of digitalization, there are still several tasks that could be automated. Picking the right thumbnail for an episode, generating subtitles for short video clips or translating the subtitles into other languages, just to name a few examples. By applying AI to these problems, we not only accelerate the processes but also save valuable time for our editorial departments.
Apart from implementing proofs of concept, developing products and automating processes, we also support our customers to find the right solutions for their issues, challenges or business ideas. Based on our expertise and experience, we provide insights into the latest approaches, technologies and best practices. Together with our customers we turn these insights into suitable solutions that sooner or later evolve into a project or a concrete product.
What’s Next?
I hope you enjoyed this article and are excited to know more about AI in the media. In the next article, we will dive deep into one of our use cases and give you a detailed account of both the business and technological aspects.
