Language：English   Publisher：Scientific Reports  

In this study, we introduce MIAS-427, one of the largest and most comprehensive inertial datasets for in-air signature recognition, comprising 4270 multivariate signals. This dataset addresses a critical gap in the field by providing a robust foundation for advancing research in cognitive computation and biometric authentication. Leveraging embodied cognition theory, we propose a novel feature selection approach using dimension-wise Shapley Value analysis, which uncovers the intrinsic relationship between human motoric preferences and device-specific sensor data. Our methodology includes a thorough statistical analysis with domain descriptors and DTW algorithms, alongside a comparative evaluation of seven deep-learning models on both the MIAS-427 and smartwatch datasets. The FCN and InceptionTime models achieved remarkable accuracies of 98% and 97.73% on MIAS-427 and smartwatch data, respectively. Notably, our analysis revealed that and contributed the most (12.82%) and least (8.71%) for the smartwatch, while and contributed the most (15.63%) and least (7.26%) for MIAS-427, highlighting significant dimension compatibility variations across devices. This research not only provides a valuable dataset for the community but also offers novel insights into human motoric behavior, paving the way for the development of more effective cognitive computation models.

DOI： 10.1038/s41598-025-03917-5

Web of Science

Scopus

PubMed

Publisher：Lecture Notes in Computer Science  

In learning analytics, early risk prediction plays a critical role in identifying students who are at risk of academic failure, enabling timely interventions to improve student outcomes. While effective in predictive accuracy, traditional machine learning models often require the training of multiple models for different time frames and suffer from a lack of explainability, limiting their practical application in educational settings. In this research, we propose ensemble dendritic neuron models (EDNMs), a novel approach for early risk prediction that addresses the challenges of existing models. EDNMs take advantage of a neural pruning mechanism inspired by biological neurons, allowing visual feature selection and explainability. The proposed EDNMs inherent visual feature selection improves transparency, making it easier for educators to interpret which factors contribute to the risk of a student. The performance of EDNMs is evaluated against RNN-based methods, demonstrating superior efficiency in prediction tasks and improved explainability. Crucially, the performance of early predictions over different timeframes is also provided by dynamically adjusting the dendritic state. Unlike conventional models that require multiple versions to accommodate predictions at different timeframes, EDNMs can adjust to varying weeks of early prediction with a single model, significantly reducing computational overhead. This study contributes to the growing field of explainable AI in education by offering a practical solution that enhances both the efficiency and transparency of early risk prediction systems.

DOI： 10.1007/978-3-031-98417-4_13

Scopus

Publisher：Lecture Notes in Computer Science  

Analyzing open-ended learner reflections can provide deep insights into students’ knowledge construction processes, yet these unstructured texts remain challenging to process at scale. In this work, we propose a context-aware graph-based approach to reveal knowledge construction patterns in learner reflections. By transforming reflections into Personal Knowledge Graphs (PKGs) with the assistance of large language models (LLMs), we extract motifs, regularly appearing substructures in graphs, to capture common patterns in how learners organize and connect knowledge. The experiments demonstrate that our approach effectively transforms learner reflections into interpretable motifs while preserving contextual relationships. Through clustering and regression analysis, we confirm correlations between motif structures and learning outcomes. Moreover, motif-based representations enable superior performance in both grade prediction and at-risk identification tasks compared to baseline approaches. This work emphasizes the potential of motif mining and analysis for understanding and supporting learning processes through reflection analysis.

DOI： 10.1007/978-3-031-98465-5_38

Scopus

Publisher：Lecture Notes in Computer Science  

We explore the use of Large Language Models (LLMs) for automated assessment of open-text student reflections and prediction of academic performance. Traditional methods for evaluating reflections are time-consuming and may not scale effectively in educational settings. In this work, we employ LLMs to transform student reflections into quantitative scores using two assessment strategies (single-agent and multi-agent) and two prompting techniques (zero-shot and few-shot). Our experiments, conducted on a dataset of 5,278 reflections from 377 students over three academic terms, demonstrate that the single-agent with few-shot strategy achieves the highest match rate with human evaluations. Furthermore, models utilizing LLM-assessed reflection scores outperform baselines in both at-risk student identification and grade prediction tasks. These findings suggest that LLMs can effectively automate reflection assessment, reduce educators’ workload, and enable timely support for students who may need additional assistance. Our work emphasizes the potential of integrating advanced generative AI technologies into educational practices to enhance student engagement and academic success.

DOI： 10.1007/978-981-96-8186-0_24

Scopus

Publisher：Proceedings of the 21st International Conference on Cognition and Exploratory Learning in the Digital Age Celda 2024  

This paper proposes a three-step approach to develop knowledge graphs that integrate textbook-based target knowledge graph with student dialogue-based knowledge graphs. The study was conducted in seventh-grade STEM classes, following a collaborative problem solving process. First, the proposed approach generates a comprehensive target knowledge graph from learning material contents, establishing a reference framework that represents the target knowledge structure of the course. Second, customized knowledge graphs were generated by analyzing the scientific concepts and knowledge based on the discussion dialogues, showing students' activated knowledge structures. Finally, the dialogue-based knowledge graphs were integrated into textbook-based target knowledge, to identify the activated and non-activated knowledge nodes and connections, as well as the related activated knowledge nodes and connections from other previous lectures or experiences. This three-step approach visualizes students' knowledge activation, and the learning gaps remain. This paper presented three examples of integrated knowledge graphs based on the different group formations. The findings of three different groups were discussed, and some educational implications were provided.

Scopus