Providing feedback to move learners forward (i.e., feed-forward feedback) is one of formative assessment's key strategies (Black & Wiliam, 2009; Hattie & Timperley, 2007). Yet, within the context of science experiments, teachers lack adequate information on students' experimental conduct to be able to provide feed-forward feedback that is timely, structured and individualized. In this regard, we can leverage computer vision to detect student actions during science experiments for feedback, thereby creating an additional source of information to support teachers in implementing formative assessment. This study aims to investigate the construction of action recognition models using data collected from 140 ninth grade students as they conduct a science experiment in school laboratories. In two separate model investigations, we 1) compared model performance of various action recognition architectures (C2D, I3D, and SlowFast), and 2) inspected confusion matrix of the best fine-tuned action recognition model. Our results indicate that 1) SlowFast architecture is most suited for recognizing student actions in school laboratories and 2) similar and consecutive student actions, as well as uneven training data distribution, generate confusion for action recognition models. Further review of these findings allowed us to 1) explore suitability of using computer vision as an added information source for formative assessment and 2) consider details of implementing computer vision for formative assessment support. Overall, this study represents a preliminary investigation that uncovered intricacies of building action recognition models in a real-world setting and unveiled the ways in which computer vision can support teachers in implementing formative assessment in science experiments.
In collaborative problem-solving (CPS), group cohesion is widely regarded as an important contributing factor, yet its underlying processes remain insufficiently understood. Traditionally, cohesion has been treated as a static construct, assessed primarily through self-report measures. In contrast, theory conceptualizes cohesion as a dynamic and emergent state, pointing to the need for temporally sensitive approaches that consider self-report evidence alongside process data. This exploratory study examined how post-task survey-based cohesion, selected socially shared regulation (SSR) discourse indicators, and synchronized leaning forward as a behavioral cue can be interpreted in relation to one another across CPS phases. Specifically, we investigated how selected SSR discourse indicators and synchronized leaning forward were configured in relation to survey-based dimensions of cohesion, including task, social, perceived, and emotional cohesion. We also conducted an in-depth qualitative analysis of selected groups differing in knowledge-construction quality to examine how SSR discourse and synchronized leaning forward appeared in interaction. The analysis is intended as an exploratory and interpretive account: synchronized leaning forward is treated as a context-dependent behavioral cue, and the observed patterns are interpreted as descriptive configurations rather than validated, causal, predictive, or generalizable relationships. The findings suggest that post-task survey-based cohesion can be descriptively interpreted alongside phase-specific discourse and behavioral configurations. In particular, early synchronized leaning forward appeared more meaningful when situated within elaborative and regulatory discourse in the selected cases. Overall, the study presents an exploratory and phase-sensitive interpretation of how survey-based cohesion, selected SSR discourse indicators, and context-dependent behavioral cues can be jointly interpreted to understand cohesion-related processes in CPS.
As the conversational AI agents become integrated into digital learning environments, students' interaction with these systems can show conversational disengagement, including mocking, misleading, or otherwise diverting the agents. Although task-based disengagement such as gaming the system and off-task actions are well documented, the nature of the consequences of this conversational disengagement in student-AI dialogue remains largely unexplored. This study examined how different types of conversational disengagement relate to middle-school students’ mathematical learning with an AI-powered teachable agent embedded in a large-scale platform. We coded 5522 utterances from 246 students into six conversational disengagement categories (False Answering, Mocking, Ghosting, Mirroring, Withholding, and Random/Noisy) versus engagement. Regression analyses showed that some disengagements (False Answering and Mocking) were positively associated with post-test scores after controlling for pre-test performance, although these associations did not replicate in a sensitivity analysis restricted to human-annotated utterances. To capture process-level dynamics, we modeled how conversational disengagement ends using a competing-risks approach and separately clustered students based on their turn-by-turn transition probabilities. Most conversational disengagement tended to end within the first few turns, whereas Withholding showed a strong tendency to continue. Students in the highest-performing cluster combined a moderate level of conversational disengagement with strong recovery pathways, whereas the lowest-performing cluster exhibited persistent self-loops without recovery. Together, these findings suggest that some apparent disengagement in conversation with AI may occur within interaction trajectories that subsequently return to productive engagement.
