Eye tracking provides a marker of attention. In the educational context, such behavior can be harnessed to understand learning behaviors. However, a technology framework that captures and utilizes such multimodal indicators in educational activities is lacking. This paper presents LA-ReflecT, a platform integrating multimodal data for micro-learning activities. Teachers can author learning tasks and enable tracking eye fixation behaviors. A web camera-based eye-tracking function captures the gaze data while attempting the learning task. Learners can control the settings to stop or pause recording. We present data-driven services such as visualizing gaze attention heatmap and genetic algorithm-based group formation. A classroom study with 41 students illustrates using the proposed framework in an authentic context. Data collected is analyzed to answer an initial research question regarding the correlation between the heterogeneity of the click and gaze patterns in a learning task. The work is open for a demo.

Enhancing peer interactions is a crucial topic in collaborative learning, where both the quantity and quality of interaction are emphasized. Despite its importance, the social and emotional aspects of collaborative learning, such as peer interactions, remain underexplored. This study applied learning analytics to examine how learner similarity in learning behavior, engagement, and knowledge influences the quantity and quality of peer interaction within the peer help system. Three aspects were analyzed using logistic regression and ordinal logistic regression: whether learners reply to questions, whether they evaluate the help received, and the evaluation scores given by askers to helpers. The results reveal that knowledge similarity has a significant negative impact on whether learners reply to questions, while no factors were found to significantly impact whether learners evaluate the help received. However, knowledge similarity showed a significant positive effect on evaluation scores. These findings provide insights into the complexity of learner similarity in shaping peer interactions and offer implications for designing methods that balance homogeneity and heterogeneity to enhance peer interactions in computer-supported collaborative learning environments.

This paper introduces a system designed to enhance extracurricular learning by recommending suitable peer helpers. The system integrates educational big data of knowledge and learner models to optimize peer learning opportunities with learning analytics. Utilizing a graph-based recommendation algorithm, it incorporates three distinct indicators to dynamically match learners with suitable peers based on specific learning contexts and needs. A significant feature of the system is its capability to visualize the knowledge and proficiency levels of potential helpers, thereby empowering learners to make well-informed decisions with less bias. The system's adaptability also permits educators to tailor it to meet diverse educational goals. We are conducting an experiment with this system during a university's paper reading course to test its usability. The system aims to reduce the burden on teachers and minimize the time students spend seeking help outside of class.

Digital learning platforms equipped with behavior sensors have provided abundant educational data. Utilizing this data, learner modeling can identify assorted learner characteristics from their behavior logs for learning analytics and dynamically update them in real time. There is a growing demand for knowledge-level modeling, moving beyond behavioral logs to assess knowledge proficiency. Based on item response theories and cognitive statistical models, existing studies estimate learning rates across various knowledge elements in each learning step in intelligent tutoring systems. However, these models, tailored to specific knowledge domains, offer limited flexibility across different knowledge units and scenarios. This paper introduces an adaptation of the basic additive factor model underpinned by logistic regression, focusing on behavior indicators. Drawing upon authentic learning data from an e-book learning infrastructure for junior high math, we examine the feasibility of our adapted models and demonstrate their potential for flexibility across knowledge units and learning phases.

This study proposes indicators from daily handwritten math learning logs of junior high school students to model knowledge proficiency, and analyzes the extent of their actual correlation with proficiency using the LEAF system. Our analysis reveals that specific pen stroke behaviors, such as writing speed and task engagement time, show significant, though weak, correlations with proficiency levels. These findings suggest that handwritten logs can serve as effective indicators of student proficiency, offering valuable insights for enhancing educational outcomes.

This study proposes the Open Knowledge and Learner Model (OKLM), a novel framework that integrates Learning Analytics (LA) with Digital Twin (DT) technology to model learners' knowledge, internal states, and environments. The OKLM DT framework addresses the limitations of traditional LA systems by enabling accurate estimation of knowledge states and personalized learning strategies. We developed a conceptual framework for the learner DT using LA, verified the accuracy of the OKLM-based DT model, and applied it to a learning support system. Initial experiments in an English literature recommendation system showed that, while the recommendations did not significantly enhance learners' motivation, they were well-received and positively correlated with increased engagement among highly motivated learners. A subsequent study involving Intensive Reading (IR) support for EFL learners further validated the model's effectiveness. Additionally, experiments targeting educators demonstrated that the OKLM's visualization tools were valuable for understanding learner characteristics and tailoring teaching materials. These findings suggest that OKLM can enhance the versatility and accuracy of learner models across various educational contexts, offering a significant advancement in the field of LA.

Peer evaluation of individual or group work is often adopted in team-based learning design. However, some raters may not take the evaluation process seriously and exhibit behaviors such as using the same score, rushing through evaluations, or not evaluating during the presentation. This study investigates the issue of unserious peer evaluation in group presentations, focusing on their behavior patterns. Using evaluation behavior analysis indicators, we identified unserious raters who exhibited low reliability in the peer evaluation process. Further, we conducted a preliminary analysis to detect unserious raters based on learner model data available before the peer evaluation process. This information can assist teachers in providing personalized prompts and interventions prior to the peer evaluation process, thus enhancing the evaluation quality of these students with timely prompts to them.

In recent years, machine learning of increasing complexity is being applied to problems in education. However, there is an increasing call for transparency and understanding into how the results of complex models are derived, leading to explainable AI gaining attention. The application of machine learning to automated group formation for collaborative work from learning system logs and other data has been progressing. Building on previous research in this field, we propose a group formation method that is based on a combination of course knowledge structures, reading behavior, and assessment analysis to create optimal heterogenous and homogeneous working groups using a genetic algorithm. The characteristics of each group are presented for explanation as a visualized knowledge map showing the strengths and weaknesses of each group, and are in the structure form of curriculum. We also present a case study of applying the method to junior high school mathematics log data, and provide explanation in a visualized form of standardized curriculum of group characteristics that are often referenced for learning design by teachers.

Jigsaw learning is one of the popular cooperative learning methods which has been utilized in many English as Foreign Language (EFL) classrooms. In recent years, utilization of technology in education has accelerated. In this study, Jigsaw+, a new jigsaw learning approach with an e-book reader, was proposed to explore how the learning design can be applied for English vocabulary acquisition, reading comprehension, and students' engagement in group activities. A quasi-experimental study was conducted at a high school in Japan for advanced and standard class students. Participants read and comprehend a story in English through BookRoll, an e-book platform with an analysis tool embedded. Two research questions were investigated; 1) to what extent did Jigsaw+ affect students' English vocabulary knowledge and reading comprehension? and 2) to what extent did Jigsaw+ promote students' reading engagement in the jigsaw group activities? The results revealed that Jigsaw+ learning tasks assisted to improve students' vocabulary and reading comprehension for both the advanced and standard groups. Moreover, it facilitated students' engagements in activities. Some limitations of the study are also discussed.

This paper introduces a system for collaborative learning which is designed to assist teachers in forming and grading groups for in-class group activities. The system is implemented as an extension of a learning analytics dashboard system and uses log data from a learning management system for operation. It consists of a group formation parameter console and the results console where formed groups are visualized and can be graded. The system supports teachers by using algorithms based on reliable learning evidence thereby simplifying the group formation process. All the group formation and grading data is logged thereby cyclically providing an infrastructure for subsequent collaborative learning activities.