The world of market research is evolving, and it’s vital that we continuously innovate to stay one step ahead for the sake of our clients – it’s something we pride ourselves on.
After a ten-hour day of action-packed coding, and a judging panel consisting of our CEO – in addition to other senior leaders – the inaugural Marketcast Hack 1.0 has concluded, and the winners have been announced.
What Specific Challenge Needed Solving?
Currently, it takes several hours for our operations and research teams to manually link answers from open-ended questions in movie trailer testing surveys to specific scenes in the trailer itself. While the manual effort means we’re able to capture results with incredible accuracy, we believed there was a way to tap into new AI technology to speed up the process without losing any of the quality of the results.
The Judging Criteria
Like any Hackathon, each MarketCast team participating had the choice of using open-source technology and publicly available developer APIs. It was mandatory that teams consisted of no less than three people who worked in the same office / time zone. MarketCasters took part from around the globe, including the U.S., UK, and India.
It’s also worth mentioning this Hackathon wasn’t just meant for coders. Teams consisted of technical, research, operations, and data science teams with varying levels of experience from across the business. And, like with any competition, winning meant prizes, along with bragging rights to the team that achieved the highest level of accuracy compared to the manually coded work.
So, let’s delve into what the teams uncovered and how they got on.
The Winning Approach
With so many smart and creative MarketCasters participating, it could only be expected that the day would bring about various concepts and approaches from the contributing teams that would be worthy of winning. Some of the teams elected to use a fully automated approaches to attack the challenge. A few chose to leverage state-of-the-art “Natural Language Processing” to perform their classifications, while others used very specific parts of sentences (“parts of speech”) as inputs into their model.
“Turnout and participation for our first ever Marketcast Hackathon was amazing,” said Tom Weiss, MarketCast CTO. “The winners stood out for couple of reasons. Firstly, the accuracy of their model was almost a full twenty percentage points higher than the accuracy achieved by the team in second place. And, their machine learning model achieved close to an 85% match to the manually coded truth set.”
It was impressive to see the winning team get such a good result in a short space of time. Our winners, who didn’t have the most experience in the Hackathon, chose an incredibly creative approach that involved the blending of human efforts to manually classify open-ended questions and then machine learning using the manual classifications to automatically classify the rest. In fact, every team that participated, they were the only one to choose an approach that blended people and machine power as an input into the process.
Congratulations to our winners, Jack Denham, Chris Marlow, and Loveesh Tomar “The New Starters”!
Interested in finding out more about MarketCast or joining our team? Have a look at the roles we’re recruiting for.