The perfect recommendation from a consumer perspective should be relevant not just for the person, but to the moment. It should also be somewhat unpredictable. Television channel controllers strike a balance between giving the audience what they know they want and what they don’t know they want
Content Discovery is an essential part of the TV experience and alongside television viewing in general is undergoing a major evolution. Steve Plunkett, Director of Technology and Innovation at Red Bee Media believes we can expect to see more innovation, experimentation and ultimately a change in how we engage with television content because of it.
We live in an age of television plenty. As consumers, we have never before had access to so much content, from so many sources, available on so many different devices. And yet, while we are watching increasing amounts of TV, consumers regularly state that they can’t find anything interesting to watch. Red Bee Media recently conducted primary research with 1,000 UK consumers and 74% said they regularly couldn’t find anything to watch on live TV. Given that we have over 1,000 channels available, plus rapidly increasing volumes of content on-demand, this seems like an absurd statement to make.
Conversely, broadcasters have never had so much competition for viewer attention. Both from other content sources and the increasing time that consumers are spending on platforms such as Facebook.
This proliferation of content and diversity of viewing devices has been driven by technology, so it is appropriate that technology be employed to help consumers navigate and discover relevant content. A growing category of technology and solutions broadly called Content Discovery are emerging. This article explores the current state of the art and what we can expect to see in the future.
While the latest technologies might be new, the process of content discovery has been around since the earliest days of television. We have used TV listings, electronic programme guides, recommendations from friends, trailers of upcoming shows and channels themselves to navigate and discover the linear schedule. All of these methods still matter but they do not scale very well and don’t easily accommodate the on-demand world that is increasingly important.
The next generation of content discovery tools and technologies.
When the Internet emerged from academia in the mid-nineties to become a mass-market consumer platform, the volume of content (non-video in this case) began to grow exponentially. The original forms of discovery and navigation, based around category listings and online directories, could not scale very well. Search developed as a more flexible alternative and there are obvious similarities between the early, categorized, Internet and fixed TV listings today.
So will search play an important role in TV content discovery? Today’s television viewers are very familiar with the search paradigm and as they migrate some of their viewing to more connected devices they will expect to find search capabilities.
But search is different when applied to TV content. The algorithms that return results for TV need to consider the availability as well as location of content. TV search needs to provide more relevance as well as semantic accuracy. A search for a TV show might need to include the linear schedule, free catch-up, paid VOD, repeats versus new shows, my preferences, my subscriptions and so on.
So search matters and it’s a different problem requiring a different solution when applied to television viewing. From the consumer perspective, relevance as well as accuracy is key. From the broadcaster perspective, search should allow business rules to be combined with technical efficiency to produce positive search results.
While search tries to provide an answer to a consumer who knows what they want to watch, recommendations try to signpost a viewer to something they might like but have not gone looking for. Recommendations come in many forms and are not a new concept. We have always talked about and recommended our favourite TV shows to each other. Broadcasters build their channel schedules in an attempt to keep viewers engaged and use trailers and in programme pointers to recommend upcoming shows to their audience. Publishers and critics recommend shows to their readers and ratings are used to score a recommendation within listings and EPGs.
Recommendation engines and algorithms take many of these same sources, in digital form, and provide individual recommendations to consumers on connected devices. The goal of a recommendation engine is to provide a highly relevant recommendation based on a particular context or consumer insight. The quality of their recommendations is built upon the depth and accuracy of their datasets and the sophistication of the algorithms that sort results.
The data that drives a recommendation could include usage information, relationships between TV shows, expressed and implied user preferences, social data and other data sets that highlight the relationships between programmes. Recommendation data should improve in quality over time and as the volume of data increases.
Algorithms process the data and produce results. They should be flexible enough to accommodate variable data types, fast enough to provide immediate results and efficient enough to scale and not over burden the underlying infrastructure.
The perfect recommendation from a consumer perspective should be relevant not just for the person, but to the moment. It should also be somewhat unpredictable. Television channel controllers strike a balance between giving the audience what they know they want and what they don’t know they want. There is a joy in flicking channels and finding something we did not expect but like, or tuning in to watch a specific show and staying on channel because the next show looks interesting. Recommendations should aim to reproduce an element of serendipity.
Just as search for TV requires a different approach than general purpose internet search, so too does recommendation. Today, many people will have experienced recommendations on ecommerce websites – people who bought the book you are looking at tended to buy these other books etc. This form of collaborative filtering does not transfer directly to television. The most popular item on Amazon might represent a very small percentage point of sales because the inventory is huge. The most popular show on TV might represent 30% of viewing. The algorithms must account for these differences if they are to produce relevant results.
Recommendation technologies are likely to be one of the most important areas of development in the future TV experience and will be a core part of the content discovery process. Netflix recently awarded a $1m prize to any group or individual who could improve the relevance of their recommendation engine by 10% or more. We can expect to see significant research and experimentation in this area as vendors and platforms strive to provide the strongest recommendations and so retain viewers for longer.
How recommendations are presented to the viewer depends upon the device (or devices) they are using to consume content. Today they tend to appear as suggested items in menus, categories or after video playback has ended. In the future we can expect to see a more linear like viewing experience where on-demand content is streamed consecutively and seamlessly to the viewer where the content choices are made using recommendation engines – much the same way that online music streaming has been evolving recently.
Finally, recommendations need to accommodate the business model of the broadcasters or platform operator. A recommendation engine must allow business rules and objectives to be blended with the ‘pure’ technical recommendations if they are to support commercial business models.
The electronic programme guide has served as an essential tool in content navigation on digital television platforms. It has been a largely static, textual experience. The EPG is beginning to evolve and will do so for the next few years at least. Firstly, it is becoming much more visually rich. The latest television platforms (and connected televisions) are including images, extended programme descriptions, video clips and links to other content and information within the EPG.
The EPG is also finding a new home on second screen devices such as PCs, tablets and smartphones. This is an area of rapid innovation at the moment. As the EPG (or TV guide application) moves to a personal device, it becomes a candidate for personalization and customization. The channel line-up can be based on individual preference rather than EPG channel numbers.
Social overlays can be added to provide information on what TV shows your friends have ‘liked’ on Facebook or even what they are watching right now. Twitter conversations around a TV show can be presented and participated in from within the EPG. Trending data from Twitter can be overlaid on the EPG to show where the current buzz is in relation to TV viewing. Red Bee has built some interesting examples of socially enabled EPGs for some major clients in recent months that show how the EPG is becoming a live socially active and personal interface.
Finally, as TVs become connected, they can be controlled by local second screen devices. The EPG applications are developing remote control capabilities so that you not only use them to discover content but also to activate it on the big screen.
Second Screen Applications
Another area of significant industry innovation at the moment relates to so called second or dual screen applications. These are applications that recognize the TV show that is playing on the big screen and extend the viewing experience in some way. This is an area that goes well beyond content discovery but such applications can use the context of the TV show to provide complimentary content and pointers to related content and will undoubtedly form part of the content discovery experience going forward.