The emotions that students experience when engaging in tasks critically influence their performance and many models of learning and competence include assumptions about affective variables and respective emotions. However, while researchers agree about the importance of emotions for learning, it remains challenging to connect momentary affect, i.e., emotions, to learning processes. Advances in automated speech recognition and natural language processing (NLP) allow real time detection of emotions in recorded language. We use NLP and machine learning techniques to automatically extract information about students’ motivational states while engaging in the construction of explanations and investigate how this information can help more accurately predict students’ learning over the course of a 10-week energy unit. Our results show how NLP and ML techniques allow the use of different modalities of the same data in order to better understand individual differences in students’ performances. However, in realistic settings, this task remains far from trivial and requires extensive preprocessing of the data and the results need to be interpreted with care and caution. Thus, future research is needed before these methods can be deployed at scale.
Original languageEnglish
Title of host publicationThe multimodal learning analytics handbook
EditorsMichael Giannakos, Daniel Spikol, Daniele Di Mitri, Kshitij Sharma, Xavier Ochoa, Rawad Hammad
PublisherSpringer
Publication date10.2022
Pages261–285
ISBN (Print)978-3-031-08075-3
ISBN (Electronic)978-3-031-08076-0
DOIs
Publication statusPublished - 10.2022
No renderer: handleNetPortal,dk.atira.pure.api.shared.model.researchoutput.ContributionToBookAnthology

    Research areas

  • Domain-specific learning in kindergarten and school - Affect, Emotions, learning analytics

ID: 1844958