April 16, 2018
In a keynote address at NAB in Las Vegas, ETC data & analytics project director Yves Bergquist described how the changing economics of media audiences require new measurement methods and metrics. For the first time, he said, the media and entertainment industry can leverage behavioral psychology, computational neuroscience and machine learning to understand the deep cognitive relationship between audiences and content. He pointed to director Alfred Hitchcock’s prescient statement that, “Creation is based on an exact science of audience reactions.”
In a market where demand is flat (no more than 24 hours in a day) and supply of content is exploding, audiences perform complex cognitive risk assessment for units of time dedicated to media products. A viral video is a key media product precisely because of its powerful cognitive cost/benefit profile: low cost (it’s just a few minutes long), low investment risk (it’s been recommended by my friends) and high return.
The second challenge, according to Bergquist, is that media audiences are expert in the product of media, which is a unique (and unfortunate) feature of the industry: audiences have made tens of thousands of recursive decisions about media consumption, so they are highly trained in recognizing (often from faint signals) good from bad products.
He added that, luckily for the industry, media and entertainment audiences are passionate and very vocal about the product, so there’s a lot of data out there to “hack” this cognitive relationship between content and audiences. But that data is unstructured, nuanced, and generally hard to interpret past surface-level insights (volume of conversation, engagement rate, etc.).
The industry needs to dive deeper, said Bergquist, and use the tools provided by machine learning (computer vision and natural language processing) and computational neuroscience to unlock the secrets of audience affinities.
At ETC, Bergquist is pioneering the use of both neuroscience and artificial intelligence to “hack” storytelling, or what he dubbed “audience genomics.” This type of semantic intelligence offers new opportunities for the entire industry to tell stories that have never been told before with more control over development and distribution risk (which right now, Bergquist said, is poorly defined).
“We are on the cusp of a revolution in audience insights,” he said. “This will create an explosion of creativity and studio revenue.”
“With artificial intelligence/machine learning in production and distribution via OTT, you get cheaper content and more opportunities to experiment with new stories in a controlled, data-driven environment,” said Bergquist. He outlined the steps of a “hierarchical de-risking process.” First, audiences will ask themselves if the product they are considering is a brand (a brand is a risk-mitigation vehicle: it predicts whether audience expectations will be met).
The second question, according to Bergquist (and perhaps the most important) is around the dimension of novelty in the content considered: expert audiences first and foremost want novelty, but at the same time that novelty has to be adequately deployed without messing with genre canons. The third question will deal with whether or not the product has good buzz. An answer of “no” to these questions predicts failure, according to Bergquist.
Both at ETC@USC and his AI startup Corto, Bergquist is building applications to address new industry needs: Bergquist and his team mine millions of fan conversations on social media and elsewhere to extract deep semantic insights on that cognitive relationship between audiences and content: what attributes of plot, narrative structure, character and arc drive passion and content performance? What kind of narrative domains are conversations in? What kind of audience needs is expressed? How interesting is the content to audiences? The application uses natural language processing for sentiment detection across 60 attributes.
Bergquist is also looking at content genomics, or which narrative structures drive the highest theatrical performance. “The tool can parse a script for presence or absence of 60 emotional tonalities,” he said. He also pointed to Corto, a semantic knowledge engine for media and entertainment that can represent and analyze all audience and content data together to surface deep semantic insights on what attributes of what content drive affinity and passion in what audience segments, thereby driving content performance.
“The Corto AI application builds and evolves machine learning algorithms in real time to produce the best answer to any audience question,” he said. “It’s an AI that builds better AIs to become more intelligent over time.”
“Today’s content recommendation is based on genre, cast, director, producer, target audience, high-level character type,” he concluded. “Tomorrow’s will be semantic, dynamic and nonlinear.”