TV screen streaming movies
Video giants are harnessing an array of powerful data to understand online viewing trends and personal customer differences. PHOTO: Jens Kreuter

In the competitive online video race for eyeballs, big data is a big deal.

Both Netflix and Hulu report 75 percent of of their audience viewing is based on recommendations. These players are constantly tweaking their algorithms, casting the net further to gain valuable insights to encourage audiences to stay tuned.

Just last month, Hulu acquired Video Genome Project, one of the largest video content databases. Hulu will use Genome’s data to advance its personalization capabilities ahead of its expansion into live over-the-top (OTT) TV early next year.

Video Giants: Harnessing Powerful Data

As online video giants team with data scientists and analytics platforms, they are harnessing an array of powerful data to understand general trends and personal differences.

This could be gathered from a user’s device, to their most recent tweets. The competition is fierce and with evolving artificial intelligence (AI) technologies, they’re rapidly uncovering new insights. So, where is the industry headed — and what can big data do for the smaller players?

Netflix

Netflix engineers are focused on fine-tuning the service’s recommendation algorithms. The original proprietary system, which was called CineMatch used basic defined categories, individual customer ratings and saved lists along with combined data from all subscribers.

The company hires freelance video experts to hand-tag video metadata using human judgement. As one analyst explained, “It covers everything from big picture stuff like storyline, scene and tone, to details of whether there is a lot of smoking in the movie."

The OTT video giant also employs computer vision, or more specifically, “extracting image metadata”, for example in deciding images and text placement when creating recommendations.

Hulu

Using similar techniques, Hulu’s Video Genome Project (VGP) dynamically aggregates metadata around video content. VGP has over 8 million records, and automatically tags video component data, separating genres on a granular level from basic sci-fi into alien or zombie.

Layering the catalogue with huge amount of data enables OTT video providers like Netflix and Hulu to dynamically categorize these. And as ratings become less important, “user actions” — what a user played, searched for, rated, browsed and scrolled past, combined with time of day, device and geo-location and larger audience trends — tell algorithms which of these categories are successful, and what to recommend to a viewer when they next log on.

Adding Context to Recommendations

Netflix recently discovered a series binge is typically followed by a period of series abstention. According to exclusive USA Today reports, 59 percent of viewers take a break for around three days before launching into a new series.

Netflix studies also reveal 61 percent of viewers went on to select a film. After "Narcos," a crime web television series, a popular option was "Pulp Fiction," switching Colombian cartels for fast-talking Los Angeles mobsters. Netflix also found viewers switching pace, turning over from "Stranger Things" and "American Horror Story" for some light relief with "Zootopia" or "Mean Girls."

Recognizing overarching trends like this helps Netflix to refine its recommendations, suggesting video content that is statistically more likely to match user tastes, at that point in time. Netflix even looks at what is trending on pirate movie sites to help inform these algorithms.

Advanced 3rd Party Analytics

In a similar fashion, Telefonica’s Movistar+ video on-demand (VOD) service uses collective intelligence to understand the user, along with content intelligence from Yomvi and Imagenio to power its ‘Para Mi’ feature.

As the industry advances, powered by third-party analytics and data-driven solutions, digital leaders are casting the net further to gather more and more information.

ThinkAnalytics, the world’s most deployed real-time content recommendation engine, used by international cable companies Liberty Global, BSkyB and Virgin Media, leverages big data to help increase pay per view purchases. The provider incorporates social media learnings, from sources such as Facebook "likes" and Twitter trending topics to provide personalized recommendations for over 150 million subscribers.

Including a social aspect gathers more information on a user; their tastes, their friends tastes and opens the door to real user sentiment. Pairing computer vision with natural language processing means that when a user highlights a certain video clip, machine intelligence can quickly identify those engaging moments for promotions.

Lessons for New Entrants to the OTT Scene

Established television providers like HBO, in moves to embrace the power of big data, are now realizing the major headache of reconfiguring internal dataflows and structures.

Small businesses have the luxury of being in a position to lay the foundations of a centralized platform for data capture. This means from the start, they can create the data channels to help organize users into cohorts, analyze churn and effectively market to new members, while also providing smart recommendations to keep existing users engaged.

One central CMS is vital when you have various components all working together, for instance when creating new apps or integrations. It starts at the user acquisition, through the sign-up, for example signing in through Facebook provides platforms with profile information, tracking all the various stages and decisions.

We’re seeing a rush of data analytics providers and off-the-shelf content discovery tools, like Taboola or Rovi, though many don’t work well together.

A number of OTT video providers, like the BBC, are also creating their own in-house proprietary systems. Though building your own technology to handle this will easily cost hundreds of thousands of dollars.

A user’s online footprint of data can be a treasure trove of insights helping to predict future tastes and suggesting new content to pique a viewer’s appetite. This could be using real-time trending topics on Twitter to suggest relevant shows, or using computer vision to analyze video metadata, or even social media conversations.

As OTT steps into Live TV, this will all be accelerated.

It’s not just categories or time of day anymore, industry leaders have set the bar high. As big data gets even bigger, this means more recommendations, better recommendations and new discoveries.