The rapid adoption of digital technologies has greatly increased the volume of real-world data (RWD) in education. While these data offer significant opportunities for advancing learning analytics (LA), secondary use for research is constrained by privacy concerns. Differentially private synthetic data generation is regarded as the gold-standard approach to sharing sensitive data, yet studies on the private synthesis of educational data remain very scarce and rely predominantly on large, low-dimensional open datasets. Educational RWD, however, are typically high-dimensional and small in sample size, leaving the potential of private synthesis underexplored. Moreover, because educational practice is inherently iterative, data sharing is continual rather than one-off, making a traditional one-shot synthesis approach suboptimal. To address these challenges, we propose the Cyclic Adaptive Private Synthesis (CAPS) framework and evaluate it on authentic RWD. By iteratively sharing RWD, CAPS not only fosters open science, but also offers rich opportunities of design-based research (DBR), thereby amplifying the impact of LA. Our case study using actual RWD demonstrates that CAPS outperforms a one-shot baseline while highlighting challenges that warrant further investigation. Overall, this work offers a crucial first step towards privacy-preserving sharing of educational RWD and expands the possibilities for open science and DBR in LA.

In many schools, teachers ensure that learners acquire academic knowledge and competences and promote their comprehensive psychological and social development through a process known as Guidance and Counseling (G&C). However, this process often involves dealing with difficult tasks and requires considerable effort from teachers. In recent years, as ICT tools have become more common, log data on daily use and learning log data have accumulated. These data have enabled teachers to understand leamer processes. Utilizing data across contexts is expected to deepen learners understanding, which is the basis of G&C. However, despite these needs and potential, the use of data for G&C has not yet been fully explored. Therefore, this study examines how the log data accumulated in the Goal-Oriented Active Learner (GOAL) system can be used by teachers to understand learners. Specifically, we extracted the characteristics of each learner's situation for various contexts from the log data, which can capture learning and daily routines. We then interviewed two teachers to determine how the visualized characteristics might be useful for G&C. The results suggest that the visualized characteristics are valuable for understanding learners' conditions both inside and outside of school, and these characteristics could support teachers' G&C activities, transcending specific subjects and activities.

This study investigates the transferability of Self-Direction behavior across different contexts and phases of learning using the GOAL system. Self-directed learning (SDL) is crucial for lifelong learning. It is significantly influenced by Self- Direction Skills (SDS), a meta-skill that is said to be transferable across different contexts, including the ability to identify learning needs, set goals, select strategies, and evaluate outcomes. Utilizing log data collected from Japanese junior high schools and analyzed using the iSAT system, we explored how Self-Direction behavior acquired in one context can be transferred to another and how these skills vary across the SDL phases. The results indicated that the Self-Direction behavior transferred between different activities and phases. In addition, the way of transfer is suggested to vary from phase and context. This study provides useful insights for the design and guidance of SDL support systems in educational programs. It suggests that it is important for educators to identify factors that facilitate the development and transfer of SDS.

This study looks at learning habits of temporal regularity in learning activities. Building such habits involves learners' regulation of their behaviors and requires learning strategies for time management, which is a cornerstone of self-regulated learning (SRL). Given the importance of habit-building in education, Learning Analytics (LA) techniques have been applied to various long-term supports by monitoring learners' habitual behaviors from the trace data. However, building a learning habit does not always mean the productive use of time. Scant supports attend to recommending learners by building which habit can improve their learning productivity. Hence, this study proposes recommendations for productive learning habit-building from learning logs. We focus on the context of English reading in a Japanese junior high school and design an algorithm to compute a recommended learning time slot. Furthermore, we collect learners' perceptions of their productivity and learning status at different times of the day. The comparison between self-report and log data presents that learners are not aware of their learning as the detection from their learning logs. This implies the potential of the proposed recommendations for facilitating learners to build productive learning habits. Specifically, our study can suggest an optimal time in learning plans and provide learners with a sustainable cue to automate learning behaviors from long-term perspectives. By building productive learning habits, learners can become more engaged in their studies as well as lead more balanced lives.

Self-directed learning (SDL) is considered a crucial skill for 21st-century learners, promoting personalized and responsive educational experiences. This study explores the untapped potential of self-reflection, particularly in e-learning environments. The research focuses on self-reflection notes, which contain strategies past students adopted when facing different situations or challenges. These notes can help current or future students facing similar situations. In this study, students take tests weekly and leave their self-reflection notes after tests. In these notes students recorded their feelings and issues, offering perspectives and insights that experts might overlook or misunderstand in some details, thus failing to provide appropriate assistance. Extracting and categorizing information from self-reflection notes is crucial to further utilize this data. Our research introduces a machine learning-based approach that effectively classifies these self-reflection notes such as cognitive, metacognitive, experiential, and irrelevant text. We compare the performance of BERT, based on the transformer architecture, with traditional machine learning classifiers such as Support Vector Machines (SVM) and Random Forests (RF). Additionally, we enhanced the BERT model by training it on synthetic data generated through GPT-4 and employing a hybrid loss combining Supervised Contrastive Learning (SCL) and Cross-Entropy (CE) to improve classification capabilities. Our results indicate that the BERT model, enhanced with advanced training techniques, outperforms traditional models in classifying learning strategies from self-reflection notes. This study not only advances the understanding of SDL in online learning environments but also demonstrates the potential of tailored machine-learning solutions to foster more effective and adaptive learning strategies.

We extracted learners' characteristic answering behaviors from handwritten process log data and investigated whether learners' situations could be inferred based on these characteristics. The result showed we were able to extract several characteristic answering behaviors, such as stopped pen stroke and late start. Furthermore, we examined the learners' situation for each feature in the actual answering process. The results revealed that several characteristic answering behaviors indicated situations such as learners' stumbling or giving up. These results imply that handwritten process log data can allow teachers to capture learners' situations and support teachers' interventions.

In terms of the learning logs accumulating from real-world educational activities, techniques of learning analytics help understand the characteristics of learners' behaviors. However, extracting learning habits from log data and data-driven support for building learning habits have not yet attracted much attention. Therefore, this research proposes the approach of “Learning Habits Mining,” which aims to extract the types and stages of learning habits from learners' daily learning logs and support learners to build learning habits with data-driven methods. We identify two contributions of this research. First, this research reveals the learning habits of K12 learners and provides an approach to trace the process of building learning habits automatically. Second, this research proposes interventions to the data-driven support for building learning habits so that learners can build learning habits based on evidence derived from learning logs.