Machine Learning is Staking a Claim in the Media Industry – Quality Data Creates Opportunities

CDO Mikko Rusama of the Finnish Broadcasting Company Yle invited key media players to discuss a common interest: practical measures for improving the utilization of machine learning in the Finnish media. The key operators in the Finnish media were well represented in the workshop with participants from Yle, Sanoma, Alma Media, KSF Media, MTV and Elisa.

Finnish media houses have already begun making use of machine learning each in their own way. For example, artificial intelligence has been used for image and text analysis and the automatic moderation of online discussions. In the media industry, the greatest needs and most concrete benefits related to machine learning currently lie in increasing the efficiency of content creation and using metadata to understand old content.

The topics of the discussion facilitated by Qvik’s Head of Design, Matias Pietilä, revolved around the current challenges of the media industry: which problems or challenging aspects of the media industry could machine learning have a real impact on and which areas are most in need of new solutions. The discussion also featured concrete international implementations, such as the machine learning solutions recently presented at the Google I/O event and Spotify’s machine learning system.

When considering the potential of artificial intelligence, you should always look at things from more than one perspective. For example, you can look for cost savings through improving process efficiency or seek to identify entirely new operating models, sometimes spanning multiple fields of business.

FROM DESIGN TO PRODUCTION AND LAUNCH TO CONSUMPTION

Media professionals can view content generation as a fairly linear process, with the initial idea followed by design, production, launch and consumption phases and, finally, learning from the content.

Yle’s concept designer Kim Viljanen shared that the public broadcaster has already validated the benefits of machine learning in both proof-of-concept projects and actual services in production. Yle Areena’s recommendation system, Yle’s Voitto robot, the Uutisvahti service and the automatic moderation of online discussions are examples of concrete first steps in the path to machine learning, but the potential is still far greater.

One intriguing field of application for machine learning could be the assistance and automation of media design work. Would it be possible to predict the popularity of content and the best channels or format for it, or even short-term trends in the development of media, already in the design stage?

When you choose the right technology, you can focus on innovation instead of infrastructure and use your time and energy on developing new products and concepts. Netflix and Amazon are pioneers of machine learning, as is the Google Cloud Platform user Spotify. Spotify’s My Daily Mix is a great example of a new concept, using machine learning to predict the listener’s preferences and create playlists.

THE RIGHT KIND OF DATA CAN MAKE ALL THE DIFFERENCE

Everything depends on quality content from which you can gather high-quality metadata for building machine learning models. It is already possible to identify objects such as text, entities or faces from content. These can be used to create predictive models for improving recommendations and media personalization. Later, it may be possible to create entirely new types of content.

Machine learning also enables automatic content analysis, making the use of archives and other content more efficient. The discoverability problem can also be solved with machine learning: if your metadata is in order, you can invite users for your content and recommend interesting content or content that reflects the emotional state of the user.

From the perspective of the consumer, the most interesting thing about machine learning is how it can improve the reading, listening or viewing experience. Media companies should be able to meet the changing needs of people through personalization. This requires a numerical understanding of audience preferences.

Knowing your audience is no simple task for Finland’s media houses. Facebook, for example, requires its users to log in, which provides them with a lot of valuable information on consumers. Many news services do not require a login and thus cannot collect as much user data, or at least the quality of the data is inferior.

The best business potential could be tapped through common rules or even a shared metadatabank. Through cooperation, the possibilities of machine learning and artificial intelligence could be harnessed to serve our small market. It might be wiser to join forces against bigger international competitors in this area instead of fighting amongst ourselves for a small market.