Why do we watch what we watch on TV? It’s a tantalizing question and the person that definitively answers it, should be able to make a fortune. But what about predicting what video and audio programming people will watch across a combination of TVs, handsets, portable players, DVRs, VoD and PCs ‘ that is one step beyond and will surely be one of the most important weapons in any triple or quadruple play.
There have been many efforts at this over time, but no-one has yet become the Amazon of TV recommendation software. We spoke to the CEO of one company this week who thinks it has a head start, operating under the name of Vizimo.
We were quite blunt with Vizimo that we worry about TV recommendation systems, because there is really very little known about why people watch one program rather than another. In the US, viewers are famous for ‘zapping,’ sampling a few seconds of each channel and wasting huge chunks of viewing time just searching for their next viewing pleasure. The arrival of the DVR, allows, with some forethought and planning, the chance to pick favorites ahead of time and record them, and watch them at leisure. CatchUp TV systems make it even easier. But even this is a worry ‘ how many times have you opened up your TiVo or cable supplied DVR and still been unsure of which program to watch?
As video fragments further into broadcast, VoD and stored content each on PC, handset and TV, it makes it even tougher to provide an enjoyable experience. Choice actually gets in the way, and most households watch just 16 channels or so consistently, but have a choice of 100 or more. One way around this is to work from a reliable recommendation service, but the emphasis is on that word ‘reliable.’
And to achieve that reliability, which criteria is most likely to make the difference ‘ should you recommend programs which are on a similar topic, programs with similar actors, programs which your friends have recommended, programs which celebrities say they have watched, or do you rely purely on huge advertising spend which make TV programs popular by highlighting their best few seconds of viewing. In fact can operators carry less programming if they can channel their viewers into programs they have bought into, instead of handing out wide ranging recommendations.
And how do you cope with that sudden swing in mood ‘ let’s say an ardent sports fan and action thriller viewer decides to watch Sleepless in Seattle all the way through ‘ how was that predictable? But it can and does happen.
The worry with a system that just doesn’t have the depth of recommendation to cope with this, is that TV becomes boring again, wading through 300 TV channels once you’ve become bored with your obvious favorites.
Let’s look at Cinema for example, without the huge promotional and advertising budgets spent at the time of theatrical release, virtually none of the top ten films would ever gather high DVD viewing figures. So the Cinema doesn’t work on recommendations it works on marketing cash spent.
Another major issue here is that most TV viewing experiences are shared across family members, so trying to analyze viewing habits is almost impossible since at any given time, depending who is home, successful recommendation would work differently.
Vizimo CEO Simon Steward was familiar with all of this and we could feel him nodding at the other end of the phone, pleased that someone grasped just what a tricky job his company was taking on. But we leveled one more thing at him and that’s the fact that a small UK start up was unlikely to come up with a service which would attract the attention of major pay TV operations, and would need to work through some kind of licensing mechanism or get squashed by some very big players.
‘We are in the business of creating software which understands what programs are about and clusters them by that understanding,’ he said and went on to talk about the ‘aboutness’ of a TV program.
‘We watch TV in three different ways’ said Steward, ‘by accident, such as when someone else in the room has the remote control; by subject ‘ when someone always wants to watch all programs on a particular subject and by arrangement.
‘We partly use a process of collaborative filtering,’ he said and by this he means knowing both what a viewer has liked in the past and linking that to an understanding of what similar viewers have also viewed.
But Vizimo can temper all of that by giving a weighting to the top ten programs viewed by everyone, or to what friends are watching if that detail is available, or what critics says is worthwhile, as well as offering tools so people can see what celebrities have watched, as well as storing how influential this information is over time.
‘Our software offers the ability to assemble virtual channels and it uses bayesian probability’ said Steward, the same techniques used in software such as Blinkx and Autonomy. This starts with a number of conflicting hypotheses which predict viewing habits, and then as more data is collected can work out which is the right rule for predicting viewer likes and dislikes.
All of this starts with a manually initiated ‘topics based hierarchy and then the Bayesian logic begins working after that, creating multiple additional rules automatically and then establishing which of them are the primary movers in recommendation.
The first step is to allocate a topic which the program covers, the next is to cluster videos together which contain the same or similar topics, and then to apply Bayesian logic to create a kind of fingerprint of the show by which to compare them.
Sure this might initially put ER and House in the same categories as programs ‘ both one hour long US dramas about medicine ‘ but then the creation of further Bayesian rules can move them further apart. If we then know whether or not a viewer offered that recommendation chooses not to view or record that program, this changes all the assumptions and eventually this recommendation disappears.
Vizimo also says that it has also put yet another layer of data, from social networking, into the mix, factoring in what your group of friends might be watching or have watched in the past.
This week Vizimo launched an API-based service which enables a service provider to add TV recommendation into its own products, feeding the service output to a set-top box, throughout the web or onto a mobile system. This is something that Vizimo is showing for the first time and this goes some way to show how Vizimo sees its route to market.
The service has three tiers, one that is free where a service provider simply sends the reference for one TV program and the new V-TV Free returns recommendations based on the standard body of TV content that is shown in that particular geography. The second tier (V-TV Plus) adds the ability to create and maintain profiles of recommendations and this includes time-sensitive recommendations or those from one particular content source, such as the operator’s existing broadcast channels. Finally V-TV OD, allows recommendations to include on-demand databases and allows the operator to distinguish between sources such as mobile and TV VoD options.
So far most of Vizimo’s revenues have come from consulting with clients, but it recently bought rival Tioti which stands for Tape It Off The Internet, an application that has sold well on the Apple App Store which provides indexing to online TV services such as Apple, Amazon, AOL and Joost, indexing 75,000 TV shows and more than 500,000 episodes and wrapping them in social tools, allowing users to recommend shows to friends.
Steward adds, ‘We have offered newspaper viewing guides access to our tools and that has brought in some money and we want to now license IPTV middleware businesses, so those businesses can give Telcos especially, recommendations over three services ‘ the internet, mobile and IPTV.’
The next step is for Vizimo to add recommendations which will be pushed to iPhone users of its Tioti application, using the new Apple push notification system announced last week. It also plans to use its technology as a basis for marrying addressable advertising with the right content, something it already has customers for on the web, says Steward.
We can’t help but agree with Vizimo that recommendation, though complex technologically, is only part of the process of how humans decide what to watch, when to watch, and on which device to watch video. This process is going to get more and more complex the more fragmented video becomes, and will be constantly shifting, in the same way that people shift from one social network to another, almost searching for the one that best represents their identity right now. Recommendation software has to be in it for the long haul and therefore very well funded, we feel.