How AI is taking the difficulty out of content discovery

Posted by Miles Weaver


As a species, we’ve always been used to having choices. Whether it’s browsing the shelves of a bookshop or scanning the menu of a restaurant, we’re naturally accustomed to making decisions based on a finite number of options in front of us.

But within the online world, those finite choices have started evolving towards the infinite, and we’re left with more choice than we’ve ever been used to. We shop on websites that sell everything we could possibly dream of buying, and we listen to music on streaming services that gives us access to tens of millions of records — far more than we could ever listen to in our lifetime.

This newfound wealth of opportunities should be liberating us, but instead it’s overwhelming us and making it more difficult for us to actually make choices. It’s a phenomenon that the American psychologist Barry Schwartz describes as ‘The Paradox of Choice’ in his book of the same name, and amongst other things, it’s beginning to hinder the way that we discover and enjoy content on our televisions. Not long ago, enjoying TV was simply a case of picking one channel out of five or six and hoping that we enjoyed what was on-screen. Nowadays, broadcasters, channel owners and online streaming providers are expanding their content catalogues at such a rapid rate that users simply don’t know where to begin when it comes to settling down and enjoying a TV show or movie.

Essentially, we no longer want to spend time searching for something that we might like. Instead, we want something that knows exactly what TV and films we enjoy to make the decisions for us.

In a bid to banish the dreaded paradox of choice and make the content discovery process simpler, companies have been relying on metadata: a set of data which, when attached to content such as TV shows or movies, allows users to search according to pre-determined criteria such as genre, country or actor. However, current technological limitations mean that the search terms can’t get much more specific than that. Plus, with every single piece of content requiring a separate metadata file that has to be inputted manually, metadata organization can become an extremely time-consuming process.

However, thanks to artificial intelligence (AI), we might be able to utilize metadata in much more intuitive and creative ways.

It’s not long ago that AI was an alien concept to anyone who wasn’t a fan of sci-fi films, and some of the first publicly available offerings turned out to be more irritating than intelligent. However, fueled by better algorithms and exponentially growing computational ability, AI has evolved considerably over the years, and it’s now woven into the fabric of our society. Consumers are increasingly comfortable with asking Siri, Cortana or Alexa for information and suggestions on where to eat or what train to get to work, and so it only makes sense that it’s also utilized by companies who want to keep their customers engaged and satisfied. By harnessing an area of AI called deep machine learning, or simply, deep learning, users can enjoy highly personalized content recommendations and a much more granular content discovery experience.

Deep learning has been around for a while, but it’s only recently that we’ve been able to realize its full potential. Before deep learning, AI approaches were largely based on trying to encode an understanding in machines through a set of rules.

For example, if you wanted to train a computer to identify cats, you would have to provide information about what a cat is supposed to look like: four legs, fur, pointed ears, medium-sized animal, etc. But, perhaps unsurprisingly, this proved not to be an efficient path to accurate machine learning. Deep learning is different in that it mimics the way a human learns.  As children, our parents pointed to enough cats and eventually we understood what they were: instead of telling the computer what a cat looks like, a deep learning algorithm would analyze images of cats using several layers of abstraction, with the outputs of each layer becoming the input to the next. It’s all about trying to teach computers to make more human connections to things.

With artificial intelligence, and deep learning in particular, there are numerous tools that can be used to better organize your content and deliver a simpler content discovery process.

Natural language processing is perhaps the most useful of those tools. It automatically analyses all of the dialogue within a given piece of content, and then allows you to categorize the content according to any number of different topics and themes. For example, once your content catalogue is analyzed using natural language processing, you could let users search for films or TV shows that had specific words of catchphrases in them. You could also go deeper than that and use natural language processing to determine the overall mood, emotion or intensity of content. Studies have also found that most of us establish a connection with a movie or TV series when we can relate to the personality traits of the main protagonists, and so by analyzing the vocabulary and sentence structure used by these characters, this would be a search criterion that companies could take advantage of. So in theory, users could search for a high-intensity drama film that features a miserable protagonist who swears a lot…for example.

Deep learning algorithms for facial and location recognition are also valuable AI tools that can massively enhance the way users search for content they want to watch. Not only is this a useful way of users being able to find all the TV shows and movies featuring a certain actor, but it would even allow them to search for actors or characters within specific scenes. For example, if you ever wanted to watch all the episodes of Modern Family where Jay is on a golf course, AI can make that happen.

Aside from improving the content discovery process, AI can also help you track the effectiveness and engagement of your service. It can be used to tell you how many users watched a certain film the whole way through and whether they watched it all in one sitting, or how many people watched numerous episodes of a TV series in a row. By combining this valuable data with your subscriber management information, you can create tailored, trusted content recommendations that will increase engagement and ensure your customers are never stuck for things to watch.

All of this can help to improve content discovery and tailor search, discoverability and recommendations for hungry viewers. Personally, I can’t wait for the day when I can come back from work, ask my TV to show me a movie I’d like to watch, and it’ll know exactly what that is — even if I don’t.

Article originally appears on Access AI


Miles Weaver Piksel

Miles Weaver is Director of Product Marketing at Piksel. Miles is an avid commentator on the digital TV revolution speaking regularly at industry events and being published in The Guardian and Read/Write. Connect with him at @MrMilesWeaver









Topics: Insights, TV, Content Discovery, online video, OTT, Metadata, AI, artificial intelligence, analyze, content recommendation