This is a discussion I had recently with my friend Nicolas Bry on how social media impacting the content recommendation space.
- Hi Ben, you’re a specialist in the field of Content Discovery, and you’re currently updating a “recommendation engine” study, tell us more about it: where does the need for recommendation come from?
Together with my colleague Sebastian Becker from thebrainbehind and David Gillies, we are putting together an updated guide for operators, to speed up their RFI/RFP processes. It’s all hush hush until it comes out, but anyone interested can always keep an eye here.
The traditional reason that we have been grappling with for years is the content explosion, but a newer challenge is brought by companion screens, multiscreen or the TV-Anywhere user experience, which is linked to the emerging social TV phenomenon that you are working on. As always, challenge also means opportunity.
But back to the Content Explosion first. It has been with us for well over a decade, at least since the “long tail” became a challenge and an incredible opportunity. It’s no longer newsworthy to note that almost all of iTunes’s millions of items have been sold at least once, or that Amazon’s critical mass of both catalogue and users enable it, unlike brick & mortar shops, to make a significant proportion of revenue from back-catalogue. But when you now look at Netflix’s catalogue on a TV in the US, or for example the Orange VoD store in France, you can only state the obvious: there’s just so much to choose from that the traditional TV navigation paradigms don’t deliver a satisfying user experience anymore (if they ever did). And if you could fix the issue within your walled garden, the OTT offerings will just shift the goal posts again.
For many user situations the goal may simply be to filter out enough of the stuff viewers don’t want, leaving them with a simpler choice.
The biggest remaining challenge for the traditional approach to recommendation is in knowing who is actually in front of the TV. User logon systems just don’t work in a sitting room TV environment. This is where the companion screen may come and save the day for recommendation vendors.
The recommendation challenge changed along with TV usage. Finding what to watch on a tablet in your lap is a different problem to finding something using a gesture based UI on a games console when with friends or on a smartphone in the train. Users are in a different state of mind, with very different interfaces, yet the TV-Everywhere operator has to provide some form of content recommendation consistency. So the recommendation problem is now even more an integral part of the user experience and EPG challenge. It can no longer be treated as a separate feature with a “click here to see our recommendations” button.
But Nicolas, as you like numbers here are seven strong reasons for an operator to introduce a recommendation engine.
1) Breaking through the clutter: Assort endless information to choose from on TV screen,
2) Revenue up: Increase usage by having users watch more content instead of browsing,
3) Churn down: subscribers that get what they want faster will be happier,
4) New customers: nothing beets a demo at the neighbor’s house,
5) New revenue streams: some content that was gathering dust is valuable if it can be targeted to the right users,
6) Serving all screens: Recommendation offers a good opportunity to provide consistency across multiple devices where the user’s context is different,
7) First mover advantage: we are still early enough in the recommendation game and there are some innovations available that will let any operator with the vision and boldness to really differentiate.
- How can recommendation engines best fulfill these needs?
Content recommendation solutions are usually classified with technical criteria depending on how they work under the hood. Without going into the details these include approaches like Semantic Analysis, where the recommendation engine uses the information contained in the content metadata to create clusters of similar or related programs. Collaborative Filtering is the technique Amazon has made commonplace with the “people who like what you like also like this” feature. Other distinguishing features include a declarative approach where the recommendation system uses information that the viewer gave willingly, i.e. “I like action thrillers & sport but I don’t like romantic comedy”. One of the trickier technique to implement in an IPTV setup is Behavior Analysis, where the system learn what you like more from what you actually watch than from what you say you like.
But the most important success or failure criteria are in how the end-solution is integrated so that it:
- is seamlessly and non-obtrusively part of the user experience, without users having to go to special recommendation pages,
- simply exploits the available metadata properly before doing anything too sophisticated,
- fully exploits all available user data (this is usually an OSS/BSS challenge),
- interacts with other systems like the “Social TV Intelligence” you are working on,
- is transparent enough for users to understand why a recommendation is made to them,
- adapts to the screen being used,
- is flexible so operators can influence the system to promote content with better margins, or for which they have already paid a minimum guaranty
Note that interacting with OTT and open information sources seems very attractive, but may not apply in all situations. TV usage scenarios always retain some lean back elements, even if a lean forward experience is less rare. Viewers welcoming technology under the hood to simplify navigation won’t necessarily welcome a detailed Wikipedia page to help them choose what movie to watch. Indeed as we said earlier, one of the main reasons for being of recommendation was information overload, so be weary of presenting more information to viewers.
- Nicolas, you’ve designed a “social TV intelligence” engine, Blended TV; could you introduce us to this market?
Social TV is digital interaction between people about television content or their digital interaction with that content, as defined by Futurescape.TV.
In my opinion, Social TV covers 3 main domains:
- Firstly is the domain of Content discovery where the EPG is enhanced with Internet information web sites, and social recommendations through Twitter feed and Facebook. Social reviews nurture social curation (“Social TV essentially makes everyone a curator”), empowering viewers to filter and voice their opinions, and then to participate.
- Secondly there is Participative TV where viewers interact with the program for voting, betting, polling, playing, converse with characters and TV presenter, Live Tweet or Facebook chat, and buy things related to the program.
- Finally the domain of Device and cloud control is where you enable channel flicking from a smart phone, flinging stored or bookmarked content around the home, or the world, one-click options to bookmark or save shows to cloud storage (universal queue).
Social TV corresponds as well to the emergence of the Companion App on smart phone & tablets, making all the connections, discovery, participation, control and giving access to all sorts of TV viewing as well: “broadcast” TV, VoD, catch up, and streaming media.
Smart phones and tablets are called the second screen in this context. A second screen brings many benefits: it is convenient (“big picture on TV, Facebook on second screen”), intuitive, frictionless, personal, and of vital importance to many operators, it is already paid for and can be monetized!
While Social TV meets high usage growth, competition is fierce: more than “50 apps currently socialize your TV”!
Once the dust of Social TV hype settles, content recommendation will be changed forever
IV Nicolas, to design your “social TV intelligence” engine, called Blended TV, where did you look for innovative inspiration?
We laid our design on 3 pillars: belief, metaphor, and model following Prof. Nonaka’s framework.
- Belief = our starting point was the belief that there is great value in social conversations around TV, but that this value is difficult to capture with the tools available to us, especially for non frequent users. Our idea was to filter out the noise so as to enable content discovery in real-time, by providing TV buzz and clean content trends, in-depth conversations related to a program, social TV computed data and metrics.
- Metaphor = Our metaphor was that of a filter, or a funnel.
- Model = from the outset, we based our approach on collaborative design. Rather than completing an end-user application, we focus our innovation endeavor on a social TV component, an underlying enabling technology, which could be embedded in various end-user applications and devices, letting others make value out of our data and build services on top of our platform through an API.
This component, called Blended TV is a semantic engine scanning social conversations, harnessing comments and filtering them. It was developed within an open innovation framework: we partnered with a social media intelligence specialist called Mesagraph and benefited from the precious overview of designer Jean-Louis Frechin from NoDesign. We also cooperated with Social TV consultants (Thibault Celier @kindoftv, Marc-Emmanuel Foucart), harnessed accurate insights (@gip89, @advid, @laouffir) and leveraged on HTML5 and interactive video skills from Djingle.
Our bet starts to win-back: developed in very short time, Blended TV is currently used or in the process of being used by various applications within Orange (Orange Sports web portal, Rendez-Vous TV / Le Mag TV companion app, Roland Garros app, Orange France web portal), and outside Orange (Broadcasters, TV metrics provider, TV guide).
V Nicolas, why is there so much buzz about the rise of Social TV?
Social TV challenges the paradigm of TV ratings:
- Nielsen has analyzed the relationship between social media buzz and TV ratings. It has shown “significant relationship throughout a TV show’s season among all age groups, with the strongest correlation among younger demos (people aged 12-17 and 18-34), and a slightly stronger overall correlation for women compared to men”.
- Social media is a great measure of audience engagement, viewers engaging to become content ambassadors on online media, before, during, and after the show is aired; ”in particular, 27-33 year-old women on Facebook are the most active sharers, and drive the highest conversion rates”!
- Social recommendations encourage interactivity, meaning stickiness to a program, and provide strong user behavior data, that can further processed to target users for advertising purpose and specific offerings.
Some predict an even stronger impact, amending the story telling:
Viewers’ engagement around TV shows will become so massive that it will start undermining the current ways of creating shows and become the main driver for new TV content” claims Anne-Marie Roussel, expert in Social TV at Sharp in Silicon Valley.
It’s a major issue for broadcasters and networks. “The future isn’t either traditional or digital: it’s a feedback loop between the two. Television fans want to get involved and be counted. It’s how creative we are in engaging those fans – and keeping them connected – that will determine how potent and profitable we will be in the future.” says Kevin Reilly, President of Entertainment, Fox Broadcasting.
“Content will then be created with social interaction in mind”, adds Anne-Marie, “the audience will be able to interact with the storyline”. Voting online for some game shows, and affecting the outcome of the show is just a start: welcome to the era of Transmedia!
VI Is to possible to merge recommendations from dedicated engines and social media? What is the challenge to meet success from a customer point of view?
Nicolas: the main challenge and the main objective of this endeavor remain relevancy and simplicity.
Mixing engine-based and social-media recommendation should bring the best of both worlds to end-users. But we’ll have to respect the specific cultures of each world to provide a straightforward and accurate suggestion.
Furthermore, consumers use a variety of sources to discover what’s personally relevant. Richard Edelman distinguishes 4 main spheres in “Media Cloverleaf”:
I believe the user interface has to screen the complexity of the engine, reflected in the various spheres, and the range of data that could be processed by a recommendation tool, such as program metadata and consumer behavior.
It should “combine the different types of recommendations together and come out with a perfect mix“, presenting a very simple proposal of “what’s up/recommended tonight”, learning to know the viewer better everyday (“the system recognizes me!”) and enable him to refine settings if he wants to engage more.
I also see social curation as creating an opportunity for a second loop for recommendation, exposing the suggestion to the social network of the user, and starting viralization of the content service … so lots to explore and I certainly think it’s worth testing and iterating!
Ben: All the recommendation engine vendors already claim to be implementing social recommendation, but much of that is vaporware so I agree with you Nicolas that there is an exciting opportunity for experimentation. Social TV will probably change the TV landscape forever. However, I don’t yet know if it’s just another feature, albeit an important one, or a real paradigm changing disruption. Issues remaining include the fact that I simply don’t want to broadcast all of what I watch to my whole social network, so I’d say that two key challenges and success criteria will include a seamless integration, and a very powerful filtering mechanism.
“Who is in front of the TV?” has proven to be an obstacle that many recommendation solutions couldn’t satisfactorily overcome. Social TV has a great side effect: it brings personal second screens into the living room.
The 50 competing apps you mentioned at the beginning of our discussion Nicolas, are all in the early hype phase. But even if they never truly deliver on their fantastic promises of a new social TV paradigm, they will at least enable plain-vanilla recommendation to at last work fully i.e. personally.