Structural Topic Modeling of Student Feedback Data

A/Prof Greg Tucker-Kellogg




What Used to Be

Bioinformatics, mathematical modelling, and basic computational analysis of biological and essential educational competencies for Life Sciences graduates.

Problems We Identified

The pedagogical problem is to incorporate true inquiry-based writing in double-blind, laboratory coursework setting, while simultaneously reducing the needs for excessive contact hours that were reducing the time of students spend on problem-solving activities and making it impossible to offer to more students.

What We Did

We tested the idea first for a project I did in one of my own modules (LSM2241), and found that students provided different feedback when we changed one aspect of the module assessment, even though the questions were unchanged. From the success of those initial results, we decided to analyse three years of teaching feedback for our whole department, a total of over 45 thousand student comments. We found topics of strength that students singled out in the teaching they rated the best, and topics for improvement that students associated with teachers rated less effective.

How This Helped

Every teacher wants to improve their teaching, but we do not always recognise what would make the most difference to our students. Using unbiased data mining technologies gives us a tool we can use to evaluate changes to our own modules. But it also has potential to identify new “levers” for effective professional development. I imagine a time near future where we use this technology to provide individualised guidance to a teacher that serves as roadmap to improved teaching. This will help new generations of NUS students and teachers, and help NUS maintain and strengthen its leadership in learning technology.