What radio listeners want
Personalized radio or TV content? In order to achieve this, it is important to find out the particular wishes of the respective user. An innovative tool can combine the advantages of the existing approaches.
If you browse through the offerings of online stores, you will read the following note: “You may also be interested in these products...” With this kind of recommendation, online sellers want to keep customers interested and tailor their offering to their interests and needs. Radio and television stations would also like to respond to the individual wishes of their listeners and viewers and need the right recommendation tools.
Both content and usage analysis are suitable as a basis for such recommendations: Via automatic analysis and metadata, providers know which content is contained within individual broadcast segments – and can thus offer users content similar to that which they have called up so far. “Collaborative filtering,” as it is known, is an alternative to that: Usage data is used to determine relations between users or items, and to draw conclusions about possible wishes of the current user from that. Both methods have their advantages and disadvantages. While content analysis allows recommendations across different domains and media types – i.e. image, text, audio, video, usage analysis enables system to better react dynamically to the actual usage.
Researchers from Fraunhofer IDMT are combining both methods in so-called hybrid recommendation approaches, thus combining the advantages of both approaches. This has already been tested in an initial scenario. “Authors were supposed to be able to write text contributions for a web portal while making optimal use of existing audio and video material that makes the content more interesting,” reports Patrick Aichroth, group manager at Fraunhofer IDMT. “For the authors, however, it was a tedious task to find the appropriate audio and video files. Our hybrid recommendation now allows that to happen automatically.”
For methods like this, data protection plays an important role. This is where a patented protocol developed by Fraunhofer IDMT comes in, which strongly decouples real identities from user pseudonyms: In this way, data analyses and personalization and recommendation services can be realized with a pseudonym that demonstrably corresponds to a real identity. However, the real user can no longer be identified from this pseudonym – even if the data is lost.