Broadband TV and Recommendation: Improving the Customer Experience Ian Kegel Future Consumer Applications and Services Practice BT Research & Technology
© British Telecommunications plc How can we make TV Recommendation work in a connected, multi-device world?
© British Telecommunications plc Answers? 1.Use a flexible, high performance Recommendation System 2.Make better use of Feedback 3.Incorporate Social Media
© British Telecommunications plc Future TV: Multi-platform, Multi-device Unified access to live broadcast, VOD, Catch-up TV, OTT and local content 20M PS3, Xbox360 & Wii consoles in the UK can deliver on-demand content to the TV today 69% of UK households forecast to have Internet-enabled TVs by 2014 The Netflix effect: over 400 Netflix- ready devices in the US, and now coming to the UK… OTT-exclusive Internet TV players emerging (such as Apple, Boxee and Google TV) Warner OTT content now integrated with Facebook Internet-enabled TVs, STBs, Media Players, Games Consoles Non Internet-enabled TVs and STBs 100M 400M
© British Telecommunications plc Why Recommendation? Huge amounts of content are available – but customers face rapidly increasing difficulty in finding what they want: the crisis of choice Recommendation enhances the customer experience by anticipating customers preferences and enabling new forms of interaction. It also enables the provider to promote specific content, and can reduce the cost of delivery by driving pre-emptive delivery techniques.
© British Telecommunications plc Why Recommendation? November 2011 Major European TV Content Providers are integrating recommendation within their proposition.
© British Telecommunications plc Recommendation Challenges Satisfying new customers who have yet to purchase anything Suggesting new content when few people have already purchased it Integrating different catalogues from multiple providers Making the customer experience clear and simple –Addressing individuals as well as groups –Knowing who is watching at a given time –Using implicit and explicit feedback appropriately
© British Telecommunications plc The MyMedia Project EU-funded Collaborative Project, Goal: To make it easier for Content Providers to take advantage of state- of-the-art recommendation systems. Flexible and modular Core Software Framework allows algorithms, content catalogues, feedback sources and UI components to be plugged in. Library of recommender algorithms can be hybridised in the most appropriate way for the application. Four real-world trials: IPTV, catch-up TV, e-commerce portal, user-generated content.
© British Telecommunications plc MyMedia Today More information:
© British Telecommunications plc Answers? 1.Use a flexible, high performance Recommendation System 2.Make better use of Feedback 3.Incorporate Social Media
© British Telecommunications plc Future TV: Broadcast TV is still king… for now The UK video rental market is small: less than 3 videos per household per year OTT VOD stores have achieved low take- up (eg. Lovefilm is streamed by < 1% of UK households) But there are disruptors on the horizon: YouView, Netflix, Apple TV 83% of viewing is still live broadcast linear TV 64% of non-linear viewing is Catch- up TV (eg. PVR, VOD) UK more linear than US, with focus on fewer higher quality channels
© British Telecommunications plc Types of Feedback Explicit High quality when given Usually based on ratings Both positive and negative End-of-scale bias Not always available Implicit Abundant, theoretically Based on observations No negative feedback Inherently noisy Need to model both preference and confidence Hu, Koren and Volinsky (AT&T Labs): Collaborative Filtering for Implicit Feedback Data Sets, ICDM2008
© British Telecommunications plc An Example 04:0008:0012:0016:0020:00 1Tuner Open/Closed 2Started watching channel or recording 3Stop/Start recording program 4Play, Pause, Rewind, Fast-forward, etc 5Delete Recording 6Schedule/Cancel Series Recording
© British Telecommunications plc Dynamics of TV viewing behaviour Taxonomy of the TV viewing process (Bilandzic, 2004) Scanning: Deliberate, heuristic evaluation of a channel Flipping: Scanning all available channels Grazing: Systematic, slow evaluation of a channel Zapping: Switching to avoid certain content (eg. commercial break) then returning Hopping: Continuous switching back and forth between two or more programmes Wonneberger, Schoenbach and van Meurs (Univ. of Amsterdam): Dynamics of Individual Television Viewing Behavior: Models, Empirical Evidence, and a Research Program, AEJMC2008
© British Telecommunications plc Managing Implicit Feedback Customers ordered by VOD viewing frequency 0O(1M) VOD views per month Implicit Feedback from VOD viewing alone is insufficient for the majority of the population.
© British Telecommunications plc Managing Implicit Feedback Customers ordered by VOD viewing frequency Customers ordered by PVR recording frequency 0 VOD views per month 100 PVR recordings per month Implicit Feedback from PVR recordings improves prediction for more customers. 10 O(1M)
© British Telecommunications plc Managing Implicit Feedback Customers ordered by VOD viewing frequency 0 VOD views per month 100 Additional events captured per month 500 Could we be smarter about collecting implicit feedback? 10 O(1M)
© British Telecommunications plc Answers? 1.Use a flexible, high performance Recommendation System 2.Make better use of Feedback 3.Incorporate Social Media
© British Telecommunications plc Future TV: a Social Experience Second screen interaction during TV viewing is becoming the norm The TV remote is being replaced by a keyboard, tablet or smartphone enabling better interaction Second screens provide access to EPG and PVR functionality Increasingly used for enhanced programme interaction and participation 80% of mobile Internet users under 25 regularly use their device to comment or chat while watching TV: –72% use Twitter 56% use Facebook 34% use other mobile applications –30% said it was "fun" and made TV "more interesting".
© British Telecommunications plc Explicit High quality when given Usually based on ratings Both positive and negative End-of-scale bias Not always available Implicit Abundant, theoretically Based on observations No negative feedback Inherently noisy Need to model both preference and confidence A New Type of Feedback? Social Can be high quality Both positive and negative Can be explicit or implicit (and difficult to interpret) Potentially very noisy Not always available
© British Telecommunications plc Some Challenges What techniques can be used to manage implicit feedback in real-world systems? How should social feedback be balanced with the traditional explicit and implicit forms? What impact will new forms of second screen interaction have on content personalisation?
Broadband TV and Recommendation: Improving the Customer Experience