ViSTA-TV: Linked Open Data, Statistics and Recommendations for Live TV
In this post, Libby Miller and Chris Newell introduce the project, which has just started.
ViSTA-TV (Video Stream Analytics for Viewers in the TV Industry) is a two-year collaborative research project about linked open data, statistics and recommendations for live TV, involving online TV viewing data, programme metadata and other external sources of data. We are working with three research institutions (University of Zurich, TU Dortmund University, and the VU University Amsterdam) and two companies (Zattoo and Rapid-I) to create:
- Real-time TV recommendations for viewers
- Highly accurate low-latency audience research
- A high-quality, linked open dataset about TV
- A marketplace for audience metrics
For us, it's an opportunity to find interesting uses for the streams of data that are produced by iPlayer showing (anonymously) what channels people are watching and when they start and finish watching. Part of the project is to analyse - in aggregate - whether there are observable events within the video, audio or text streams that make people change channel, and combine this with external features such as peoples' interests or peoples' actions on social media.
It also gives us the opportunity to learn more about machine learning techniques, particularly data mining over large streams of data. We hope to draw on the expertise of Rapid-I and the Universities of Zurich and Dortmund to get a greater understanding of the contribution of various forms of data to the quality of the information we get back in return. So for example we might find out whether it is worth doing video processing to look for events of interest, or whether subtitles is a better route - or perhaps whether working with both datasets gives us quantifiably better results.
A major goal for some of the partners is to produce a marketplace for audience metrics, to lay the groundwork to be able to offer characterisations of audiences in real time for sale. 91Èȱ¬ R&D is interested primarily in the other end of the spectrum - being able to create and use linked open data sources for TV-related data (such as /programmes), and to this end the work of the VU University Amsterdam in the project will be very interesting, as it builds on work begun in the NoTube project on collating and enriching metadata about TV.
The 91Èȱ¬'s role in the project, like Zattoo's, is a dual one: that of data provider (anonymised streams of behavioural data from iPlayer simulcast streams, subtitles, metadata, video and audio) and application maker. Our goal is not just to find out more about the data but to give audiences something useful back, such as recommendations of what to watch right now, or other applications that make interesting use of the results of the project and enable us to evaluate them. We are very pleased to be able to follow up aspects of the MyMedia and NoTube projects, which in different ways addressed data for recommendations - this project takes those results and pushes them in new directions, towards real-time analytics.
As I mentioned in weeknotes last week, we'll be holding a workshop on possible applications, so if this is something that interests you, let me know and I'll try and get you an invite.
Comments Post your comment