New technology to solve an old marketing problem
With artificial intelligence being one of the most disruptive technologies these days, researcher Efthymios Constantinides and four master’s students looked at new ways to use AI as a marketing tool: using machine learning for segmenting potential UT students. ‘From thousands of interactions of prospective students on the UT website, we identified search profiles of students most likely to convert – meaning actually starting a study here,’ Constantinides explains. ‘It starts with finding lots of reliable and – for privacy reasons – anonymous data. Using this data and machine learning, we identified the potential students’ search journeys and profiles based on the search behaviour of prospective students.’
‘We were able to identify twelve different customer journey steps and six different visitor profiles. These findings could help the university marketing department to better assess the likelihood of someone becoming a student and better understand the UT website visitors’ behaviour. Still, this is both a new tool and a first step. The next step for the marketers is to adapt their marketing strategy and efforts in line with the knowledge gained from this machine learning method.’
To Constantinides, the research offers a multitude of benefits. ‘For students and marketing professionals, it is a new way to learn and solve a typical marketing problem, namely, to increase conversions. You can apply the same approach to for example underperforming studies and to any other type of business.’
Machines checking open answer exams
‘I do have an enormous passion for integrating new technologies in education,’ says Adina Aldea. That’s why she’s working on a way to automatically grade (digital) open question exams using machine learning, as a ‘pet project’.
‘Checking exams is a very tedious and repetitive task, especially when it comes to open question exams. And most of the time, answers are quite similar, but the length of answers can vary from a sentence to a paragraph or even a page,’ says Aldea, posing the question: ‘What if we could eliminate several problems, by automating the process?’
Without that many tools for open question exams, Aldea knew it came down to pioneering. ‘I approached the subject in two ways. Firstly, by looking at standardising questions so it becomes easier to group answers that are more or less the same. Secondly, is the nuance in answers of open exams teachable to an algorithm? This is quite difficult, and I feel that we’re making progress. In the ideal situation, when the algorithm groups the exams properly, you would only need to grade about five to ten exams, for groups of more than five times that size. It also helps students: if they are assessed in a similar manner, there is less bias in grading.’
For the future, Aldea thinks that an expert system might be a solution. That system could generate model answers, which could be compared to the answers of students. ‘This is worth a try, for the benefit of both teachers and students.’
Never a classroom too full or too empty
Rudy Oude Vrielink did research on forecasting classroom utilization to improve timetabling. ‘I know it sounds lame, but bear with me: it’s the opposite of that. I believe I found a way to reduce costs by using classrooms more efficiently, while at the same time giving true power to teachers.’
He already took an important first step, by using sensors that he placed in 55 lecture rooms all over campus. With at least 90 percent accuracy, these sensors monitor how many students are in a lecture hall, real-time. An example: Waaier 3, with a capacity of 135 seats, had only 38 seats occupied at the time of the interview. ‘Teachers tend to overclaim lecture rooms. They expect all students who follow a course to show up – sometimes even more than that. That’s almost never the reality.’
By keeping track of the occupation and utilization of lecture rooms and rebooking rooms last-minute when needed, Oude Vrielink thinks the university can save at least 300 thousand euros a year. ‘That’s the first step, that can already be implemented – if the university wants,’ he says. ‘The second step is to schedule courses based on appropriateness and enhance a community feeling by having classes of the same study near each other. The third step is to use data to predict the influx of students. The fourth step is to combine all previous steps and have true adaptive scheduling.’
‘Imagine a university where a system learns from the feedback of teachers and the data from the room sensors. Imagine that teachers and students automatically receive a message each morning where their classes are. Imagine that we don’t have to use an entire education building, because everyone fits. Imagine happy teachers that know they get the room they prefer, to give the best education possible. That situation might feel far away, but it is very much in reach.’
