Where is the space for emotion in Learning Analytics
Handbook of Learning Analytics | 2nd edition, Chapter 12 Reflection
“Learning is not a cold intellectual activity, it is nuanced with emotion.” This quote from chapter 12 is often where debates around learning technologies land. Learning analytics focused solely on academic performance gives us one side of a multi-dimensional story. The authors point out that there is still too much to learn about emotion and the emotional states learners enter when learning but point out that what can be observed and quantified for analysis is a learner’s actions and interactions where their affective state can be inferred, which, while not the same thing, ties closely to the emotions of the learner.
One way to measure affective state is through the Student Activity Data. This technique uses trained observers to log various affective states like boredom, frustration, engaged concentration, and confusion. This type of data is less useful for immediate adjustments to student learning. It can be used to make an overall product better by pinpointing trends in low affect states of users and it can be used to compare to machine interpretations of affective states in order to assess the precision of affect detection.
Another way to measure affect is through the use of bodily signals detected by machines. Our text describes a study where students were observed playing a physics educational game. Videos of students faces were analyzed using the FACET computer vision program which predicts the users state based off facial actions like raising eyebrows or tightening lips. The video data alone was only partially usable due to interruptions like excessive movement or poor lighting, but the text points out that when coupled with interaction-based detectors, the combination of the two yielded 98% applicability.
What I found the most impressive and interesting in this chapter is the section where the authors dive into the applications for this type of affect data. Affect-aware technologies can start to scratch the surface of accommodating what many learners miss when learning with technology- a sensitivity to learner needs that go beyond the content proficiency data. In this chapter we learn about Affective AutoTutor, a physics, computer literacy, and scientific reasoning learning system. Affective AutoTutor monitors students' affective states and adjusts with prompts to respond to counterproductive states. For instance, if the system detects boredom, the computer-generated tutor will prompt the learner with encouraging statements and opportunities for review.
This layer of feedback to students has the power to deliver data-informed interventions developed from a wider spectrum of learning analytics to understand the whole learner. We have the ability to formatively assess the state of the learner and respond in real-time to procure content data that reflects the learner’s abilities when at their best state of mind for learning. I think when we think about learning analytics, we often assume because we are referencing data and numbers, we inherently exclude the humanness and emotion of learning. However, research in the field of affective states, when applied to the process and plans for the learning trajectory, not only includes the human state of the learner but also finds ways to steer it into its most powerful version.
