Figure Eight & Human Digital

Learn how Figure Eight helps Human Digital create ground truth for algorithms that unlock actionable insights from the messy world of social media.

“Figure Eight is integral when it comes to sorting, enriching, and refining our data. This job would’ve taken [a single researcher] nearly 2 years to complete, but using Figure Eight we were able to complete this job in 4 days.”

– Luke McManus,
Junior Analyst, Human Digital

The Company

Human Digital is based on the principle that data-led insight is only possible when human beings direct technology. They mine social media data for deep insights about what people really think and how people really behave online. With a roster of some of the largest and most innovative companies in the world, Human Digital delivers deeply nuanced social media datasets that lead to better understanding and smarter insights.

The Challenge

There is an unfathomably large amount of social media data created each and every day across the world. Over 600 million tweets, 4 million hours of YouTube videos, and over 4 billion Facebook messages are created each and every day. But people in different countries use different platforms for different purposes. They use unique colloquialisms, slang, and vocabulary. No single model or approach can understand it all and finding actionable insights into that data requires massive amounts of annotations and expertise.

Each of Human Digital’s clients has their own, individual needs. Social media, traditionally, has been mined for insights ranging from the efficacy of a new product feature to market forecasting to understanding a specific population’s wants, beliefs, or needs. Each client needs their own unique dataset, labeled their own specific way, to uncover the pertinent information that client needs.

Human Digital employs a methodology they call S.O.A.R., which involves sourcing data, organizing it, analyzing it, and reporting on it. They have a tremendous in-house team to do it, as well as a network of subject matter experts to leverage for specific insights. What they don’t have is an ability to label data for their NLP models at scale. So they turned to Figure Eight.

Figure Eight & Human Digital

The Solution

Recently, Human Digital was asked to help analyze Russian social media behavior, specifically surrounding religious messages related to Islam and the Russian Orthodox Church. They wanted to understand what messages were about, if the groups interacted, what they have in common, what they didn’t, and where the messages originated, so they could create a map of religious expression in Russia.

But there were a few challenges here. For starters, all the messages were, of course, in Russian. The location data, which was key here, was in different formats, was hyperlocal, and was often misspelled. And to top it off, there were over 31,000 pages to annotate.

Those pages required 200,000 human judgements. Human Digital estimates that for a single researcher, working without lunch breaks or sleep, that amount of work would take 703 days. On Figure Eight, it took four.

No out-of-the-box model or existing in-house algorithm could handle this unrefined data, after all. Human Digital wasn’t looking for broad sentiment but several specific insights their customer needed. That meant specific annotations on each piece of data so they could build a model that, eventually, could analyze raw data coming in from social media. Human Digital was able to target nearly 5,000 Russian speakers on Figure Eight to get the annotations that power this particular model, leveraging their subject matter expertise about hyper-local geography and Russian idioms, ensuring their models pick up on nuance and slang.

Human-led contextualization is incredibly important when you’re trying to truly understand human expression. But once those models are trained, they unlock the ability to do real-time assessments of online extremism, disinformation, new trends or events, or really whatever it is their clients need to.

Figure Eight is proud to help Human Digital analyze human data at scale, building ground truth for algorithms that unlock actionable insights from the messy world of social media.