CLEVER - Making Complex Learning processes Visible for Enabling Regulation: Change human behavior for learning success

CLEVER dataset consists of video, audio, physiological data (including heart rate, electrodermal activity (EDA), and accelerometer data), self-reports (survey), and qualitative coding (labels) of 30-second segments of video data (students’ participation, type of interaction, group level regulation of learning). The data collection involves secondary school 7th grade students (13 years old, n = 94, male = 36) working collaboratively in groups of three or four during four physics lessons and a collaborative exam. The data collection was implemented in the students’ own classroom. The students were divided into 30 groups heterogeneously based on previous science grades. The groups remained the same throughout the five sessions, but some group compositions were affected by occasional student absences. In addition, nine of the groups had their 90 minute-lessons in two parts due to a lunch break in the middle. During the lessons, the students performed various collaborative learning tasks related to the topic of light and sound. Students and their processes of learning and collaboration were monitored and documented with video and audio recordings and through individual-level physiological measures and self-reports. In total, 225 hours of video were recorded by using Insta360 Pro 360-degree cameras (90 min per session/week=225h). A qualitative video coding protocol with 30-second segmentation was developed and applied to the video data enabling the identification of students’ participation, type of interaction and subsequently instances of co- and socially shared regulation of learning during collaborative group work. Shimmer 3 GSR+ sensors (Realtime Technologies Ltd, Dublin, Ireland) were used to record students’ electrodermal activity, heart rate and accelerometer data. The sampling rate of the recording was set to be 128hz. The sensor data were synchronized and downsampled to 16hz to fasten the further analysis. EDA recordings indicating a missing contact of the electrode were removed from the data set. Butterworth low pass filter with frequency 1 and order 5 was used to remove the small movement artifacts from the signal. Non-specific skin conductance responses with minimum amplitude of 0.05μS were detected from the signal with trough-to-peak method (Dawson et al., 2017). Signal was also decomposed into phasic and tonic components with the Ledalab toolbox (Benedek & Kaernbach, 2010). Self-reports include those captured before and during (situated) the data collection. Self-reports obtained before data collection are for measuring students’ SRL strategies (Cleary, 2006), metacognitive awareness (Schraw & Dennison, 1994), task interest (Cleary, 2006), self-efficacy (Usher & Pajares, 2008), and appraisals of collaborative working (Volet, 2001) were administered. Self-reports during the data collection were collected using the 6Q-tool (Järvenoja et al., 2020), which the students used at the beginning and at the end of each collaborative learning session. 6Q-tool measured students’ appraisals of task understanding, task difficulty, task interest, emotional state, and collaboration as well as their goals, planned strategies, and reflections about the goal achievement. The data collection was conducted as a part of the CLEVER project funded by Eudaimonia, University of Oulu; and EmReg and StrategicLearn projects funded by Academy of Finland and by the support of Multidisciplinary Research on Learning and Interaction (LeaF) infrastructure at the University of Oulu (https://www.oulu.fi/leaf-eng/).