Incorporating student opinion into opinion mining: A student-sourced sentiment analysis classifier

Garron Hillaire*, Bart Rienties, Mark Fenton-O’Creevy, Zdenek Zdrahal, Dirk Tempelaar

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingChapterAcademic


To build a better understanding of the online student experience, there is promise in exploring emotional measures. The first step in gaining clarity on the role of emotions in online learning is understanding the accuracy of emotional measures. As text is ubiquitous in online learning, in Chapter 13 we reviewed an emotional measure of sentiment analysis which can interpret text for emotional expression. In order to better understand the experience of students through text we take a clear theoretical stance that emotion is socially constructed and that text can be categorized as positive, negative, neutral, or mixed. We trained a student-sourced sentiment analysis classifier by using crowd-sourcing methods with students, and benchmarked it with traditional crowd-sourcing techniques. Our results show that our student-sourced classifier did a better job of predicting future student labels as compared to our benchmarks. In addition, interviews with students demonstrated that five out of six students found our student-sourced classifier useful. The results of this study raised questions about how ground truth is established for sentiment analysis and advocates the utility of anchoring ground truth to the student experience.
Original languageEnglish
Title of host publicationOpen World Learning
Subtitle of host publicationResearch, Innovation and the Challenges of High-Quality Education
Place of PublicationLondon & New York
PublisherRoutledge/Taylor & Francis Group
Number of pages15
ISBN (Electronic)978-1-003-17709-8
ISBN (Print)978-1-032-01091-5
Publication statusPublished - 2022

Publication series

SeriesRoutledge Research in Digital Education and Educational Technology

Cite this