Learning analytics applications consider not only the cognitive dimension, but also the physiological dimension of the learner. This paper describes a learning analytics approach that focuses on alerting the critical stress level of the learner using a pedagogical agent. For that purpose, an existing pedagogical agent was expanded by a software component, which analyses heart rate variability data to determine the cognitive load of a user and to offer support with stress reduction. The evaluation study with the physiologyaware pedagogical agent showed an improvement of learning and a reduction of stress.
In order to investigate the specified research question, the functionality of the web-based
pedagogical agent LIZA
To use HRV parameters to determine stress level, the generation of data has to be ensured. A
technical solution was provided by the wristband E4 of Empatica. Such a device was chosen to
minimize the complexity of the handling and to provide a most comfortable position and positioning
of the sensors
Since there are no generally accepted threshold values to determine the degree of a mental load of
a person, a series of individual measurements has to be done
Figure 3 Arithmetic task in stress test 1 In two arithmetic stress tests (see Figure 3), with different levels of difficulty, the user had to subtract a random value from a certain number consecutively for five minutes. The result of the previous equation provides the minuend of the following. The level of difficulty is altered through the time limit for solving the equation, the number of digits of the random value and the value of the start minuend. The second and third factor determine the number of shifts during mental arithmetic, which increase the cognitive load with a growing number. Furthermore, a competition situation is created by requesting to beat LIZA in the number of correct answers under certain Copyright © 2020 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 Internationa3l (CC BY 4.0). conditions in the second test. With that, the first test provides the RMSSD value for a moderate load, the second, which is designed with a higher difficulty level, for an overload.
Certainly, a range of cognitive load could be defined in that way, but a specific threshold, is still missing. Considering that an excessive load can result in a decrease of motivation and an abort of the learning in the long term, the learning situation has to be adapted before such a scenario materialize. Another factor, which has to be taken into account is that an adaption of the learning situation through stress reduction strategies will interrupt the process itself. So, it should be carried out as little as possible but also as much as necessary. A preliminary empirical test, where the task solutions were known and therefore low stress were induced, showed, that a threshold at 50% of the range triggers an intervention nearly every time LIZA was used. This would lead to massive interruptions of the learning process. Based on that, the threshold was increased to 2/3 of the individually defined stress range, where the intervention could be reduced to 40% of the cases. If the current RMSSD falls below the threshold after a specific time, LIZA offers assistance in reducing the stress level through stress reduction strategies. Among different strategies (e.g., mindfulnessbased stress reduction, autogenic training), two methods are proposed that are appropriate for the learning environment of a pedagogical agent. The first one distracts the user by telling jokes, the second one shows a video with relaxing content. The user decides whether it is necessary to start the offered coping process and how long the strategies are used. If the stress level is significantly reduced below the threshold, LIZA proposes the continuation of the learning process.
The goal of the evaluation study is to determine the effectiveness and benefits of the pedagogical agent that was extended with the capability of measuring HRV and detecting critical stress level of learners. Amongst others, following hypotheses were examined: 1) The stress reduction strategies lead to the relaxation of learners; 2) The RMSSD is a suitable indicator for cognitive load; 3) The Copyright © 2020 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 Internationa4l (CC BY 4.0). adaption of the learning process affects the learning performance. To examine these hypotheses a pre- and posttest was performed. For the study, 34 participants (10 males, 24 females) aged between 21 and 59 (mean 31 ± 11 years) were acquired and assigned to test- or control group by random. The test was conducted in a quiet environment under supervision.
Every participant was asked to use the pedagogical agent to perform two stress tests, each with a different level of difficulty, to determine the stress limit range and calculate the threshold. Then 4 tasks were given by the pedagogical agent to be solved, where every task covered a different problem of reasoning. After that, the RMSSD was calculated for the time frame of the first task block and squared with the previously determined stress limits. Only for the test group followed a stress reduction phase, if the current RMSSD fell below the threshold. Both groups continued with the posttest that required all participants to solve again 4 tasks. The reasoning problems of the pretest and posttest were the same, but the tasks were different. In the end, the participant got an evaluation of how successful the tasks were solved.
For every part of the process, the RMSSD was calculated so that the development of the indicator
could be retraced. In addition, the participant had to self-asses its currents state of mental load with
six adjective pairs of opposite meaning after each measurement cycle. A short questionnaire for the
current cognitive load (KAB)
This paper has demonstrated the integration of physiological factors in learning analytics using wearable sensors. It showed, that the analysis of the RMSSD to determine the cognitive load of a learner can be used to improve the learning situation. But to determine the adequacy of the RMSSD as suitable indicator further test has to be conducted. Not only shows the integrating of wearable sensors a potential to improve the learning situation, but adds also the possibility to use cognitive data beyond the use in learning analytics.
Copyright © 2020 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 Internationa5l (CC BY 4.0).