AIM-HI: Dataset on Motivation and Emotions in Human-AI Interaction during Collaborative Learning

Abstract

The data were collected as part of the Hybrid intelligence research programme at the University of Oulu and carried out with the support of LeaF research infrastructure, University of Oulu. The dataset was collected to study university students’ motivation, self- and socially shared regulation of learning, and collaborative learning processes in a technology-supported learning environment. In particular, the dataset enables the examination of how students regulate their learning while completing an authentic collaborative academic task with access to generative artificial intelligence tools. The research design was structured around a course assignment in which students collaborated in small groups throughout seven seminar sessions. The collaborative task was completed at the University of Oulu in the LeaF research infrastructure (http://www.oulu.fi/leaf-eng), which is a classroom-like learning and research space. Before the collaborative task period, students answered two questionnaires concerning motivation and self-regulated learning. The expectancy-value-cost questionnaire, based on Gaspard and colleagues (2017), consisted of 39 items measuring expectancies, task values, and costs. The self-regulated learning questionnaire, adapted from Hadwin et al. (2025), consisted of 51 items measuring students’ perceived importance, self-efficacy and practices related to self-regulated learning. From the original questionnaire, the 31 question-instrument measuring self-regulated learning challenges was excluded in this adapted version. In addition, background information, such as name, age, and gender were asked. Altogether, 20 responses were collected for both questionnaires, meaning there was no missing data. In addition to the trait questionnaires, repeated situational self-reports were collected at the beginning and end of each seminar session. The self-report at the beginning of the sessions consisted of four items measuring students’ emotional valence, feelings towards the group task, self-efficacy beliefs, and intrinsic value related to the ongoing collaborative task. These items were adapted from the instrument by Törmänen et al., 2026. The self-reports at the end of the sessions consisted of the same four items together with three additional items addressing value of the group task, regulation of emotions and suitable working strategies during task completion. The three additional items were adapted from the self-regulated learning questionnaire developed by Hadwin et al. (2025). Altogether, 124 beginning-of-session self-reports and 123 end-of-session self-reports were collected across the seven seminar sessions. Limited missing data occurred because two three-member groups were absent from one seminar session each. The students’ collaborative work was recorded using multiple data sources. Three groups were present in the learning and research space at a time, each group in their own separate room. Video and audio recordings were collected throughout the seminar sessions using two cameras, one group microphone located in the ceiling, and with additional personal microphones for each participant. Altogether, the observational data included 59 hours of video, with the learning sessions’ mean duration being approximately 90 minutes. 60 minutes of the collected video data were analyzed by an annotation tool, based on AffectGPT (Zheng et al. 2025), developed by Zheng and Wang in the Center for Machine Vision and Signal Analysis (CMVS), Faculty of Information Technology and Electrical Engineering in the University of Oulu. The annotation tool automatically segmented the video into approximately six second segments. In sum, 28 minutes and 41 seconds of these segments were self-annotated by the participants of this study. The self-annotation included correcting the description of one’s behavior, assessing situational motivation, and emotional state as well as the valence and intensity of that emotion. In addition, screen recordings were captured from the groups’ shared computers, which were used for collaborative writing, accessing course materials, and interacting with Microsoft Copilot. The shared computers provided access to Microsoft Word documents used for collaborative tasks, the Moodle learning environment used to facilitate the course, and the Microsoft Copilot generative artificial intelligence tool. Individual log data were collected from students’ actions in the Moodle learning environment. The use of Microsoft Copilot by groups was recorded, including both the prompts and outputs of the discussions. Furthermore, the written assignments produced during the collaborative process were collected in the form of Microsoft Word documents. The participants of the study were 20 second-year university students (17 female, mean age 25,2) enrolled in the Learning Sciences program at the University of Oulu. Students were assigned to six small groups consisting of three to four members, resulting in four groups of three students and two groups of four students. Participation in the study was voluntary and did not influence the way students completed the collaborative task. The collaborative task was integrated into the course requirements and involved conducting a literature review on a topic selected by each group. The task progressed through phases distributed across the seminar sessions. During the first session, students engaged in brainstorming and idea generation. The second session focused on developing a conceptual understanding of the selected topic, followed by literature searching in the third session. The fourth session involved synthesizing information from the identified literature. During the fifth session, students produced written text with particular emphasis on the introduction and discussion sections. The sixth session focused on preparing a presentation, and the final session was dedicated to reviewing and finalizing the written work. The resulting dataset combines trait questionnaire data, repeated situational self-reports, video and audio recordings, screen recordings, learning management system log data, AI and self-annotation of emotions, CoPilot inputs and outputs, groups’ goal setting, reflection and collaborative learning products, providing a multimodal perspective on motivation, emotions, regulation of learning, collaborative learning, and the use of generative artificial intelligence in higher education. References Hadwin, A. F., Rostampour, R., & Winne, P. H. (2025). Advancing self-reports of self-regulated learning: Validating new measures to assess students’ beliefs, practices, and challenges. Educational Psychology Review, 37, Article 8. Törmänen, T., Turunen, M., Suoraniemi, J., Korhonen, E., Heikkala, N., & Järvenoja, H. (2026). MotoR: Dataset on a Motivation Intervention Supporting Students’ Situational Motivation (versio 1). Oulun yliopisto, Kasvatustieteiden ja psykologian tiedekunta. https://doi.org/10.23729/fd-d8361c6a-9aeb-34b9-9acd-a39fd28c4c3a Gaspard, H., Häfner, I., Parrisius, C., Trautwein, U., & Nagengast, B. (2017). Assessing task values in five subjects during secondary school: Measurement structure and mean level differences across grade level, gender, and academic subject. Contemporary Educational Psychology, 48, 67-84. Zheng, L., Chen, H., Chen, L., Sun, H., Sun, L., Ren, Y., Cheng, Z., Liu, B., Liu, R., Peng, X., Yi, J., & Tao, J. (2025). AffectGPT: A new dataset, model, and benchmark for emotion understanding with multimodal large language models [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2501.16566