Building a solution using NLP
“This process was quite time consuming,” explains Karen, “as the team had to trawl through each and every response. Using NLP, we were able to build a solution, which automatically coded comments against topics and correctly identified the sentiment of what was being said. This included making allowance for counter-intuitive sentiment tagging, for example when the answer is ‘nothing’, but the questions is ‘what could we improve?’, then that’s a positive sentiment! Filtering out these basic responses also meant that the team could spend more time focusing on analysis of the more substantive comments.”
With the University of the Highlands and Islands being spread over such a large geographical area, the team were well set up for remote collaboration, so when the covid-19 pandemic put paid to plans to have Karen join the team on-site for part of her placement, the project was unaffected.
Harnessing the power of data
The team were able to apply Karen’s model to a fresh dataset for the first time when the annual NSS results were released towards the end of the placement.
“Previously, Heather’s team would release headline data on the day NSS results were published,” Karen explains. “They would then have to spend weeks preparing bespoke analyses for various different departments and individuals. With our new model, full results were available to everyone within hours on the day of publication.”
While Karen used specialist data programmes such as Microsoft Power BI to design her solutions, the model was also made available in Microsoft Excel, ensuring that colleagues across the university could access and customise that data that mattered to them.