Tsukasa Hirashima
Hiroshima University, Hiroshima, Japan

Published : 6 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 6 Documents
Search

Comparison of Text Representation for Clustering Student Concept Maps Reni Fatrisna Salsabila; Didik Dwi Prasetya; Triyanna Widyaningtyas; Tsukasa Hirashima
MATRIK : Jurnal Manajemen, Teknik Informatika dan Rekayasa Komputer Vol. 24 No. 2 (2025)
Publisher : Universitas Bumigora

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30812/matrik.v24i2.4598

Abstract

This research aims to address the critical challenge of selecting a text representation method that effectively captures students’ conceptual understanding for clustering purposes. Traditional methods, such as Term Frequency-Inverse Document Frequency (TF-IDF), often fail to capture semantic relationships, limiting their effectiveness in clustering complex datasets. This study compares TF-IDF with the advanced Bidirectional Encoder Representations from Transformers (BERT) to determine their suitability in clustering student concept maps for two learning topics: Databases and Cyber Security. The method used applies two clustering algorithms: K-Means and its improved variant, K-Means++, which enhances centroid initialization for better stability and clustering quality. The datasets consist of concept maps from 27 students for each topic, including 1,206 concepts and 616 propositions for Databases, as well as 2,564 concepts and 1,282 propositions for Cyber Security. Evaluation is conducted using two metrics Davies-Bouldin Index (DBI) and Silhouette Score, to assess the compactness and separability of the clusters. The result of this study is that BERT consistently outperforms TF-IDF, producing lower DBI values and higher Silhouette Scores across all clusters (k= 2 - k=10). Combining BERT with K-Means++ yields the most compact and well-separated clusters, while TF-IDF results in overlapping and less-defined clusters. The research concludes that BERT is a superior text representation method for clustering, offering significant advantages in capturing semantic context and enabling educators to identify student misconceptions and improve learning strategies.
Revealing Interaction Patterns in Concept Map Construction Using Deep Learning and Machine Learning Models F.ti Ayyu Sayyidul Laily; Didik Dwi Prasetya; Anik Nur Handayani; Tsukasa Hirashima
MATRIK : Jurnal Manajemen, Teknik Informatika dan Rekayasa Komputer Vol. 24 No. 2 (2025)
Publisher : Universitas Bumigora

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30812/matrik.v24i2.4641

Abstract

Concept maps are educational tools for organizing and representing knowledge, enhancing comprehension, and memory retention. In concept map construction, much knowledge can be utilized. Still, concept map construction is complex, involving actions that reflect a user’s thinking and problemsolving strategies. Traditional methods struggle to analyze large datasets and capture temporal dependencies in these actions. To address this, the study applies deep learning and machine learning techniques. This research aims to evaluate and compare the performance of Long Short-Term Memory (LSTM), K-Nearest Neighbors (K-NN), and Random Forest algorithms in predicting user actions and uncovering user interaction patterns in concept map construction. This research method collects and analyzes interaction logs data from concept map activities, using these three models for evaluation and comparison. The results of this research are that LSTM achieved the highest accuracy (83.91%) due to its capacity to model temporal dependencies. Random Forest accuracy (80.53%), excelling in structured data scenarios. K-NN offered the fastest performance due to its simplicity, though its reliance on distance-based metrics limited accuracy (70.53%). In conclusion, these findings underscore the practical considerations in selecting models for concept map applications; LSTM demonstrates effectiveness in predicting user actions and excels for temporal tasks, while Random Forest and K-NN offer more efficient alternatives in computational.
Enhancing Semantic Similarity in Concept Maps Using LargeLanguage Models Muhammad Zaki Wiryawan; Didik Dwi Prasetya; Anik Nur Handayani; Tsukasa Hirashima; Wahyu Styo Pratama; Lalu Ganda Rady Putra
MATRIK : Jurnal Manajemen, Teknik Informatika dan Rekayasa Komputer Vol. 24 No. 3 (2025)
Publisher : Universitas Bumigora

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30812/matrik.v24i3.4727

Abstract

This research uses advanced models, Generative Pre-trained Transformer-4 and Bidirectional Encoder Representations from Transformers, to generate embeddings that analyze semantic relationships in open-ended concept maps. The problem addressed is the challenge of accurately capturing complex relationships between concepts in concept maps, commonly used in educational settings, especially in relational database learning. These maps, created by students, involve numerous interconnected concepts, making them difficult for traditional models to analyze effectively. In this study, we compare two variants of the Artificial Intelligence model to evaluate their ability to generate semanticembeddings for a dataset consisting of 1,206 student-generated concepts and 616 link nodes (Mean Concept = 4, Standard Deviation = 4.73). These student-generated maps are compared with a reference map created by a teacher containing 50 concepts and 25 link nodes. The goal is to assess the models’ performance in capturing the relationships between concepts in an open-ended learning environment. The results show that demonstrate that Generative Pretrained Transformers outperform other models in generating more accurate semantic embeddings. Specifically, Generative Pre-trained Transformer achieves 92% accuracy, 96% precision, 96% recall, and 96% F1-score. This highlights the Generative Pretrained Transformer’s ability to handle the complexity of large, student-generatedconcept maps while avoiding overfitting, an issue observed with the Bidirectional Encoder Representationsfrom Transformer models. The key contribution of this research is the ability of two complex models and multi-faceted relationships among concepts with high precision. This makes it particularly valuable in educational environments, where precise semantic analysis of open-ended data is crucial, offering potential for enhancing concept map-based learning with scalable and accurate solutions.
Performance Evaluation of Artificial Intelligence Models for Classification in Concept Map Quality Assessment Wahyu Styo Pratama; Didik Dwi Prasetya; Triyanna Widyaningtyas; Muhammad Zaki Wiryawan; Lalu Ganda Rady Putra; Tsukasa Hirashima
MATRIK : Jurnal Manajemen, Teknik Informatika dan Rekayasa Komputer Vol. 24 No. 3 (2025)
Publisher : Universitas Bumigora

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30812/matrik.v24i3.4729

Abstract

Open-ended concept maps generated by students give better flexibility and present a complex analysis process for teachers. We investigate the application of classification algorithms in assessing openended concept maps, with the purpose of providing assistance for teachers in evaluating student comprehension. The method used in this study is experimental methods, which consists of data collection, preprocessing, representation generation, and modelling with Feedforward Neural Network, Random Forest, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Decision Tree, and Logistic Regression. Our dataset, derived from concept maps, consists of 3,759 words forming 690 propositions, scored carefully by experts to ensure high accuracy in the evaluation process. Results of this study indicate that K-NN outperformed all other models, achieving the highest accuracy and Receiver Operating Characteristic-Area Under the Curve scores, demonstrating its robustness in distinguishing between classes. Support Vector Machine excelled in precision, effectively minimizing false positives, while Random Forest showcased a balanced performance through its ensemble learning approach. Decision Tree and Linear Regression showed limitations in handling complex data patterns. FeedforwardNeural Network can model intricate relationships, but needs further optimization. This research concluded that Artificial Intelligence classification enables a better assessment for teachers, enables the path for personalized learning strategies in learning.
Machine Learning for Open-ended Concept Map Proposition Assessment: Impact of Length on Accuracy Reo Wicaksono; Didik Dwi Prasetya; Ilham Ari Elbaith Zaeni; Nadindra Dwi Ariyanta; Tsukasa Hirashima
MATRIK : Jurnal Manajemen, Teknik Informatika dan Rekayasa Komputer Vol. 25 No. 1 (2025)
Publisher : Universitas Bumigora

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30812/matrik.v25i1.5044

Abstract

Open-ended concept maps allow learners to freely connect concepts, enriching understanding by linking new and prior knowledge. However, manually assessing proposition quality is time-consuming and subjective. This study proposes an automatic classification model for proposition quality assessment using term frequency–inverse document frequency (TF-IDF), a text representation method based on word frequency, and several machine learning algorithms. Two datasets were used are Relational Database with an average 5 words per proposition and Cybersecurity Authentication with an average 10 words per proposition. Comparative experiments with Support Vector Machine (SVM), a supervised classification algorithm, K-Nearest Neighbor, Random Forest, and Long Short-Term Memory (LSTM), a recurrent neural network for sequence data, revealed that SVM with RBF kernel achieved the highest performance on shorter propositions 87% accuracy, Cohen’s Kappa 0.76, while LSTM showed greater strength in handling longer propositions 85% accuracy, Cohen’s Kappa 0.69. These findings suggest that proposition length influences model effectiveness. The proposed approach can reduce the burden of manual assessment, increase the objectivity of evaluation, and support more efficient implementation of concept maps in education.
Assessing the Semantic Alignment in Multilingual Student-Teacher Concept Maps Using mBERT Nadindra Dwi Ariyanta; Didik Dwi Prasetya; Ilham Ari Elbaith Zaeni; Tsukasa Hirashima; Reo Wicaksono
MATRIK : Jurnal Manajemen, Teknik Informatika dan Rekayasa Komputer Vol. 25 No. 1 (2025)
Publisher : Universitas Bumigora

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30812/matrik.v25i1.5046

Abstract

This study examines the effectiveness of mBERT (Multilingual Bidirectional Encoder Representations from Transformers) in assessing semantic alignment between student and teacher concept maps in multilingual educational contexts, comparing its performance with TF-IDF. Using datasets in both Indonesian and English, the study demonstrates that mBERT outperforms TF-IDF in capturing complexsemantic relationships, achieving 96% accuracy, 96% precision, 100% recall, and a 98% F1 score in the Indonesian dataset. In contrast, TF-IDF achieved higher precision (73%) and accuracy (79%) in the English dataset, where mBERT recorded 54% accuracy, 47% precision, but 90% recall. Semantic alignment was measured using cosine similarity to calculate the cosine of the angle between vectorsrepresenting textual embeddings generated by both models. This method facilitates cross-linguistic semantic comparison, overcoming challenges related to word frequency and syntactic variations. While mBERT’s computational demands and the study’s limited linguistic scope suggest room for improvement, the findings highlight the potential for hybrid models and emphasize the transformative impact of AI-driven tools, such as mBERT, in fostering inclusive and effective multilingual education.