Posted 23 Jun
MOD. 
Words by Walter Marsh.
“Traditionally you might go to a fortune teller who says, ‘hey, I see your future’,” Ansah says. “But we use computer science techniques and mathematics.”
Specifically, Ansah’s PhD research has focused on the particularly timely subject of protests, riots and civil unrest, using openly available data from social media to map and project future events.
“We use social media data as the main source of data, and then we build machine learning models that can basically study and understand historical patterns of how people behave prior to protest events. Then we deploy that model to monitor social media tweets in real time, and based on historical patterns they’re able to tell us if something is going to happen or not.”
Ansah’s models use a text-based approach to create a ‘propagation graph’ — a web of users and groups linked by shared causes and engagement. “That shows us how users on social media connect to each other and how they form communities. The way the structure of those connections evolves over time can be an indication of protest events or something happening.
“A typical example: you have a WhatsApp group or Twitter with a group of friends. If you wake up one morning with 200 messages from one particular group chat, what does that tell you? It probably tells you that there was something interesting being discussed in that group within that time.
“We have those sorts of dynamics happening in social media; if there’s a protest event of interest, usually within a short time you’d have a sudden change in the structure of the network. We mine the properties of the network, how fast the network it’s growing, the connected components, how the links between users change over time, and we use those features in a machine learning model to predict future events.”
Ansah looked to overseas movements such as the Arab Spring uprisings, unrest in Venezuela and US Black Lives Matter protests from 2012-2015, all of which featured prominent use of social media. The broad online footprint created by those events and movements offered a variety of variables and indicators that in turn informed Ansah’s modelling, from changes in volume of activity to the escalating tone of communication.
“We have another model that’s basically looking at sentiment analysis — how happy or angry people are in their tweets. That’s an indication of satisfaction, [and] if it hits a certain threshold it can lead to protest events.”
With global unrest, civil rights protests and far right unrest growing more frequent and impassioned in the past year, such research raises broad questions for how authorities and corporate interests might use such innovations (“We can’t prevent people protesting because they have that right,” Ansah notes).
For Ansah, being able to predict highly charged public protests is an opportunity to better calibrate official responses and public warnings for the safety of protestors and bystanders. Importantly, this kind of modelling is also an opportunity to read and respond to public sentiment before a crisis point is reached.
“There are definitely a lot of applications. The good thing is that most of these models are transferable in terms of understanding user perception, user reviews and sentiment about a particular product.”
Having completed his PhD, Ansah now works in an entirely different field as a data science SME (subject matter expert) at a major mining organisation, where he applies data mining and machine learning techniques to support data driven decision making for equipment maintenance strategies.
It’s a completely different lane and subject to his research, but for Ansah such work is proof of the broad potential that big data and machine learning presents. It’s exactly the kind of problem solving that captured his imagination as a high school student in Ghana fascinated by computers.
“Data is the new oil,” he says. “Over time we’re going to get more complex systems that solve problems we used to think were impossible — I’d like to think it will make our lives better.”