November 8, 2018 - How might artificial intelligence help to categorize the good, the bad and the ugly when it comes to opinions online?
That was the thrust of the question put forward by Thales Canada in its latest Student Innovation Championship. The annual case competition challenges students across Canada to present an innovative solution that solves a theoretical challenge, and then pitch the idea to industry representatives.
With no prior background in computer science, three Western University graduate trainees at Schulich Medicine & Dentistry decided to take a leap and banked on the fact that their critical thinking, thirst for innovation, and ability to think big could lead to success.
Their risk paid off. Of 50 teams across Canada, they walked away with the top prize on November 3 in Montreal, which included $20,000 in prize money.
“I think what we did differently that gave us the edge is that we didn’t think of opinions as a dichotomy,” said Megha Verma, MSc candidate, who entered the competition along with fellow masters students Kartik Pradeepan and Fan Liu. “We looked at opinions on a spectrum which can each be right in different contexts.”
The team used their background in scientific research to narrow down their pitch to specifically look at scientific literature. Their project, titled “Opinion Galaxies: A Machine Learning Network Approach to Big Data in Medical Research,” used an algorithm that examined semantics.
“We wanted to be able to demonstrate that not all scientific papers are created equal so we treated each paper as an opinion. Then we used an algorithm called Doc2Vec that examines language to link the papers together,” said Verma.
The algorithm learns and understands language in the same way children do – by examining the context of words and phrases, rather than through understanding definitions.
The algorithm helped the team to categorize the scientific papers into what they called ‘opinion galaxies.’ The size of the galaxy and networks between the stars within those galaxies helped to determine how much weight to give to a particular opinion or hypothesis.
“We used network visualization to see how papers are linked together and to see the weight that they hold,” said Liu. “In terms of application, this could be used in a variety of fields, specifically for researchers who want to see how paradigms for certain diseases can be linked together, or for a clinician to see which network is the most feasible for that clinician’s patients.”
The team says the fact that they have strong scientific backgrounds but only a narrow understanding of computer science actually played to their advantage.
“Because we had only limited knowledge of some of the identified challenges to this idea, we were allowed to just think big and imagine where we could take it, and that’s where our strength was,” said Pradeepan.
Article by Crystal Mackay, MA'05