What are you going to discuss in your conference session at IBC 2014?
Navneeth Kannan (NK): We are presenting a method for a machine learning framework that simulates how we could develop a more personal viewing experience based on viewer behavior.
Video consumption behavior is evolving at a frenetic pace with the viewer freed from both the confines of the living room and the limits of traditional broadcast TV. While there have been advances in the user interface for over-the-top content, like catch-up and on-demand, interfaces for broadcast television need to catch up.
Consumer devices and network infrastructure have advanced to the point where channel navigation interfaces could, and should, be significantly improved from the traditional Electronic Program Guide (EPG). We believe we can harness the increased processing power of consumer devices to make machine learning techniques feasible.
You mention a machine learning framework. What does that mean?
Dinkar Bhat (DB): The adaptive user-interface takes advantage of content characteristics and viewership behavior based on the premise that though the number of channels available has increased, people are ultimately only interested in a certain number – i.e., we all have our favorite channels.
NK: The framework we’re putting forward allows for a platform that can accommodate multiple viewer profiles that orders channels based on the number of times a viewer opts for a channel and the amount of time spent once there. We’ve added in a few more factors with the ultimate aim to be able to learn viewer channel preference and even predict channel ranking for unwatched channels – something that could be done for alternative movie, sports or news channels for instance.
DB: We believe this adaptability is vital. Even if you have user interfaces that provide filtering features for large number of channels, they are not able to flex to changing user viewing patterns.
Why are you recommending an alternative to the traditional grid-style Electronic Program Guide (EPG)?
DB: Program guides originated at a time when the number of broadcast channels was far fewer, and programming was limited to prime time. Newspapers and magazines published these guides as tabular grids, with channels listed as rows, time-slots listed in columns, and program information filling the cells.
The EPG came to replicate this structure, but as the number of channels vastly increased, this grid interface remained – despite developments in video delivery, the capabilities of set-tops and on-screen graphics, and that’s before we even begin to talk about metadata. Everything is still ordered by channel numbers, and that’s something we believe is far too limiting for today’s consumption habits.
Even recent advances at the headend that organize channels by genre are themselves still too rigid, in that they are unable to offer an experience personal to each viewer.
What benefits does machine learning have for broadcasters and service providers?
DB: This framework modernizes the EPG to bring a level of informed suggestion, flexibility and personalization to TV experience. By enabling better navigation, broadcasters and service providers can potentially benefit from an increased understanding of audience behaviour, which could be extended to create opportunities for targeted advertisements as well as cross-selling VOD or other high-value content.
It is also scalable and agile. Newer components can be added after initial development and deployment, making it possible to refine and optimize the navigation experience – something we think broadcasters and service providers will find valuable.
What are some ways to improve the user interface and make it more attractive to consumers?
NK: We believe that the user interface should be as streamlined as possible to let viewers get to the content they want to watch as quickly and easily as possible. Recent OTT players have made viewers increasingly accustomed to recommendation engines, and the framework we’re proposing speaks to that in a live broadcast sense. Viewers increasingly demand an immediacy with their technology, and having to scroll through over a list of a hundred unsorted channels makes little sense in this day and age.
Navneeth Kannan and Dinkar Bhat co-wrote “Adaptive Television User Interface using Machine Learning Concepts”, a paper that Navneeth will present at IBC 2014: Paper session: New consumer trends – the rise of social media and second screen technologies (11:30-13:00, September 12, 2014)