Here comes a new big-data approach trying to crack the age-old problem of understanding what a TV show or movie is really about.
Entertainment-analytics startup Vody is coming out of stealth after more than two years of development and testing. Co-founders and co-CEOs Stephanie Horbaczewski and Jeremy Houghton, who both were previously top execs at YouTube network StyleHaul, claim they’ve built a better mousetrap. The company’s proprietary system, they say, uses machine-learning tech to trawl the internet and compile a comprehensive database of entertainment titles — designed to plug into streaming services for more accurate content recommendations.
The L.A.-based company was formed by Horbaczewski, previously founder/CEO of StyleHaul, and Houghton, who was StyleHaul CTO. They both left the RTL Group-owned fashion and beauty digital network before RTL shuttered StyleHaul last year.
“We want to give platforms the understanding of content that human beings have,” Horbaczewski told Variety. “That’s really what we set out to do.”
Horbaczewski’s goal is to sell Vody to one of the big streaming players, pitching the company’s tech as offering a key differentiator in the streaming wars.
Vody (vody.com) recently completed trials of its system with Comcast and WarnerMedia. In addition, it closed a $10 million round of Series A funding, from backers including Horbaczewski; friends-and-family investors; family office Clemens Management; and San Francisco-based Cacker Capital, led by principals Matthew Bloodgood and Christina Clark.
According to Vody’s founders, its system processes billions of data points daily, culled from hundreds of thousands of websites and social services (such as Facebook and Twitter) searching out commentary on films and TV shows, similar to the way Google crawls the web. The Vody system then applies natural-language processing on the data to create profiles for a TV or movie title (which the company calls “embeddings”). Houghton said Vody’s standard embeddings comprise about 5,000 attributes with emotion-based and descriptive tags.
Vody’s approach lets its determine nuances that metadata systems are unable to provide, according to Horbaczewski. “We found that ‘Dirty Grandpa’ and ‘The Exorcist’ are both described as ‘creepy’ – but they’re creepy in different ways,” she explained.
The company’s natural-language processing system also is able to interpret a query like “What’s the movie where Buzz Lightyear is reset to speak in Spanish?”, according to Houghton. (For the record, it’s “Toy Story 3.”)
Historically, services like Netflix have relied on collaborative-filtering models to inform their recommendations (i.e., they predict titles individual viewers are likely to be interested in by matching up their tastes and viewing history with that of other users). Vody, by contrast, collects and crunches input from the vast universe of internet users — yielding a richer set of data to refine recommendations, the founders claim.
Newer streaming arrivals like Disney Plus don’t have “Netflix and Amazon’s 10 years of data. But people talk about Disney online all the time,” Houghton said.
Vody (which rhymes with “body”) has found some surprising correlations among entertainment titles. For example, the company’s database shows a match between superhero flick “Iron Man 3” and “Deal of the Century,” a 1983 comedy starring Chevy Chase, Gregory Hines and Sigourney Weaver. Both movies are based weapons and technology, and deal with the ethical dilemma of selling weapons. Vody’s system determined that the movies appeal to the same group of viewers because of their use of satire and the consistent theme of modern technology, and because the emotional response from the viewers of each film is similar.
Houghton said the startup’s tests with media companies have shown Vody can improve recommendation engines’ precision (how likely it is that someone watches a recommended title) and recall (the percentage of recommendations a viewer actually watches), which are the two standard metrics for evaluating such tech. (He declined to quantify the lift that Vody and its test partners found.)
Meanwhile, applications for Vody’s database can extend beyond recommendations, Houghton said. “Over the last two years, every single month a potential partner comes to us, saying, ‘This is the exact technology we need for X,’” he said, such as voice search, advertising targeting and strategic content planning.
Vody builds on the work Houghton did at StyleHaul, where he oversaw the team that created AI-based technology to optimize viewership on YouTube by figuring out what drove engagement with specific influencers.
In 2017, when Vody was an early skunkworks project, Horbaczewski and Houghton launched a consumer-facing Android app in Southeast Asia to test how well its algorithms recommended content. “Get personalized recommendations and discover new tv shows and movies from all the services (Netflix, iflix, Hooq, Amazon, Hulu, etc) all on Vody,” reads the Google Play description of the app. “As you explore shows and movies, VodyBot gets smarter at recommending things for you to watch. Spend more time watching and less time finding.” The app was eventually phased out, per a Vody rep, after it had “served its purpose.”
Currently, Vody has 15 employees, which includes engineers and developers who hail from media companies including CBS and Disney, as well as IBM Watson, the CIA and NASA. Horbaczewski and Houghton also have tapped two media-industry vets as advisers: Sandy Grushow, former chairman of Fox Television Entertainment Group and CEO of Phase 2 Media, and Jon Miller, former chairman/CEO of News Corp’s digital media group where he oversaw its investments in Roku and Hulu.
For Horbaczewski, who started StyleHaul in 2011 and then sold it to RTL for $150 million in 2014, the new venture would appear to be a dramatic change of scenery — taking her from the world of fashion and beauty influencers to the business of advanced data science.
But Horbaczewski said she’s more passionate about Vody than anything in her career: “This is the coolest work I’ve ever done.”
Source: Read Full Article