Teresa Phillips, Spherex | AI-AI-AI-OH! Machine Learning in Content Recommendation & Cultural Compliance

Teresa Phillips, Spherex | AI-AI-AI-OH! Machine Learning in Content Recommendation & Cultural Compliance

Photo credit: techdaily.ca / unsplash

 

May 24, 2021 | Not a day goes by without some company, somewhere in the world, talking about how it’s using artificial intelligence (AI) to improve how consumers locate content to consume. Whether it’s Netflix, Amazon or Hotstar, companies are constantly searching for better ways to make appropriate recommendations to their customers.

Teresa Phillips | Spherex

In digital entertainment, this problem is particularly critical because with an average of 356,000 titles being released each year across dozens of platforms, competition for consumer attention is fierce. Subscription churn and fatigue are big obstacles, and AI machine learning is the tool nearly every platform uses to keep their content and their brand at the top of consumers’ very short attention spans.

But when most folks think about AI, they think of recommendation engines and data mining, only two of the hundreds of AI applications being employed across digital entertainment. The future of our industry depends on being innovative in our use of AI to address many of the challenges we face today. One that I have spent the past 10 years studying, is the ability to AI-ify cultural markers that allow content distributors to safely and easily adapt their content for local markets around the world. Read on for more.

Because you watched…

Let’s presume for a moment that everyone reading this has at least one subscription to a video-on-demand or OTT provider. Let’s further suppose that at some point in time, each and every one of you has gone to that provider’s site, clicked through the recommended or new content lists; and after about 10 minutes, said to yourself, “Nothing’s on” and moved on to the next provider and repeated that same process until you found something to watch.

If you’re like me, there are times when you’ve looked at Netflix, Apple, Amazon, HBO Max and maybe one or two others and finally settled on something you’re really not all that excited to watch. This may not necessarily be because these content providers don’t have something in their catalog that would interest you – they probably do. It’s more likely that their search and recommendation algorithms aren’t really tuned to what does interest you.

Say you just finished watching Black Panther and after the credits roll, you go back to the home screen and find many recommendations of what to watch next. In a real-life search, which we encourage you to try yourself, you might get a listing of other superhero films like Wonder Woman, Iron Man, Batman, Justice League and Guardians of the Galaxy. None of these movies include T’Challa (the Black Panther character). Since you just finished watching Black Panther, you might be more interested in Avengers: End Game, Avengers: Infinity War and Captain America: Civil War, all of which do have the Black Panther character.

Why aren’t they recommended? Blame the algorithm.

All AI engines contain the biases of their developers. It’s unavoidable. Some focus on genre, cast, theme, producer (e.g., weighted toward platform-produced content), recent versus older production dates, and so on. There are myriad ways to spin it and all of those make sense. But teaching machines to generate coherent recommendations isn’t solving the right problem.

We’re reminded constantly in our work that machines don’t subscribe to video-on-demand and OTT services, people do. People tend not to make the same generalizations as machines because, well, they aren’t machines. Looking at different cues within a title – and there are many – helps train AI to make recommendations that make sense to people.

Culture matters

One overlooked aspect of AI as it relates to content is understanding and assessing cultural and religious norms in various countries around the world. With the global expansion of streaming, it is increasingly important that content meet local cultural guidelines if it is to be exhibited overseas. The typical “subs and dubs” work done to prepare films for international distribution is simply not enough.

Localization, as we call it, is the process of adapting to one’s local area with regards to language, technology, or legal requirements. On the other hand, culturalization is adapting to one’s cultural environment – their beliefs, values, and customs. Localization is quite a mature industry, whereas culturalization is just emerging.

The issues arise in evaluating content for cultural fit. In other words, will the film or TV show resonate with its local audience and avoid offending minorities or classes of society?

Most of this work must be done by humans who manually screen content by watching it. But advances in AI, particularly in multi-modal architecture, make it possible for much of the higher-level review to be done by machines. Not only does this reduce the human workload to that of a reviewer, but it also gives content creators the opportunity to get a head start on reviewing their work for cultural issues that may require edits, reshoots or cuts.

There is no question that AI has become a critical tool in helping consumers find something to watch. The effectiveness of search results is based largely upon how the system is trained and how results are weighted. While it is understandable that streaming platforms want to promote what’s new, popular and related, or something they created, the fact remains humans are the ones looking for what to watch, not machines.

By focusing on the human aspect, platforms and content creators can present options that keep people on their service and do not force them elsewhere.

 

Teresa Phillips is the CEO and Co-Founder of Spherex, a  global technology and data company that rejoined DEG in April 2021.  “Spherex is thrilled to rejoin DEG in its mission to advocate for innovation across the entertainment ecosystem and foster a community of businesses supporting that ecosystem,” said Phillips.

By matching TV and film content with local audiences worldwide, Spherex’s expertise and technology offers digital entertainment companies the ability to adapt all of their content for every market and culture around the world. Learn more at www.spherex.com.