Jurnal RESTI (Rekayasa Sistem dan Teknologi Informasi)
Not a member yet
1071 research outputs found
Sort by
DiG-MFV: Dual-integrated Graph for Multilingual Fact Verification
Get PDFThe proliferation of misinformation in political domains, especially across multilingual platforms, presents a major challenge to maintaining public information integrity. Existing models often fail to effectively verify claims when the evidence spans multiple languages and lacks a structured format. To address this issue, this study proposes a novel architecture called Dual-integrated Graph for Multilingual Fact Verification (DiG-MFV), which combines semantic representations from multilingual language models (i.e., mBERT, XLM-R, and LaBSE) with two graph-based components: an evidence graph and a semantic fusion graph. These components are processed through a dual-path architecture that integrates the outputs from a text encoder and a graph encoder, enabling deeper semantic alignment and cross-evidence reasoning. The PolitiFact dataset was used as the source of claims and evidence. The model was evaluated by using a data split of 70% for training, 20% for validation, and 10% for testing. The training process employed the AdamW optimizer, cross-entropy loss, and regularization techniques, including dropout and early stopping based on the F1-score. The evaluation results show that DiG-MFV with LaBSE achieved an accuracy of 85.80% and an F1-score of 85.70%, outperforming the mBERT and XLM-R variants, and proved to be more effective than the DGMFP baseline model (76.1% accuracy). The model also demonstrated stable convergence during training, indicating its robustness in cross-lingual political fact verification tasks. These findings encourage further exploration in graph-based multilingual fact verification systems
Combining the Cellular Automata and Marching Square to Generate Maps
No full textAs computer technology advances, one of the entertainment media that has emerged is video games. The development of a video game is becoming more expensive and labor-intensive as technology itself continues to grow. One of the characteristics of a game as an entertainment medium is its replay value, which refers to the fact that the subject matter can be played more than once. Automating content through the use of procedural content generation is done with the goal of lowering expenses and reducing the amount of labour that is required. This research has two goals: designing and developing a Maze Game using the Procedural Content Generation method with the Cellular Automata and Marching Square algorithms, and determining the level of player satisfaction with the games developed using the Game User Experience Satisfaction Scale (GUESS) method. This research will utilize Cellular Automata and the Marching Square algorithm as a method for generating 3D game shapes through Procedural Content Generation. After the game has been developed, it will be performed by players, and the Game User Experience Satisfaction Scale will be used to measure the user experience. The result for overall satisfaction, based on the responses of 25 respondents, is 83.14%. Cellular Automata was effectively implemented to generate the map, while Marching Square was used to generate the 3D mesh, albeit with isolated rooms and graphical errors
Enhancing Problem-Solving Reliability with Expert Systems and Krulik-Rudnick Indicators
No full textProblem-solving is one of the skills needed in the 21st century, but there is a significant gap between the ideal conditions and the reality of students' problem-solving skills. One method that can improve students' problem-solving skills is Krulik and Rudnick, but implementing this method with an expert system to improve problem-solving skills is still limited. This research aims to build an expert system to determine the level of problem-solving using Krulik and Rudnick's problem-solving indicators processed using the forward chaining and certainty factor algorithms. The study had five stages: data analysis, rule generation, certainty measurement, prediction, and testing. The data was processed by developing 5 Krulik and Rudnick problem-solving indicators into 35 statements. Each statement was categorized using Forward Chaining by producing three rules: low, medium, and high. The problem-solving level obtained is calculated using the Certainty Factor for a confidence value. The system's prediction results were evaluated using a confusion matrix, resulting in an accuracy of 80%, a precision of 92%, and a recall of 85%, indicating the system's reliable performance in measuring the level of problem-solving. This research can be used as a reference to support problem-solving in various more advanced educational and professional environments
Language Processing for Detecting Fake News on Twitter Using a Long Short-Term Memory Architecture
No full textThe rapid spread of misinformation on social media platforms, particularly X (formerly Twitter), poses a significant challenge to public trust and democratic integrity. Fake news is often crafted to deceive readers and manipulate public opinion, especially in political contexts such as the 2024 Regional Head Elections (Pilkada 2024). Although various measures have been proposed to mitigate this issue, achieving an effective balance between controlling misinformation and preserving free speech remains a challenge. This study aims to address this problem by developing a fake news detection model based on Natural Language Processing (NLP) and Long Short-Term Memory (LSTM). The dataset used in this study was collected from public tweets related to Pilkada, with Kompas.com serving as the validation source to verify content authenticity. Experimental results show that the proposed LSTM model outperformed traditional classification methods, achieving a precision, recall, and F1-score of 0.95, along with an overall accuracy of 94.90%. Confusion matrix analysis further confirmed the reliability of the model by demonstrating low misclassification rates. This study contributes to the advancement of AI-driven hoax detection systems, offering an automated and scalable solution for combating misinformation in political discourse
Handling Imbalance in Javanese Manuscript Character Dataset using Skeleton-based Balancing Generative Adversarial Networks
No full textJavanese script is an important part of Indonesia’s cultural heritage, representing cultural values from the past. However, recognizing and classifying Javanese characters within manuscripts is challenging due to the limited availability of data and uneven distribution of character classes. The decline in formal use of Javanese script has drastically reduced the pool of manuscript samples, causing certain characters to appear rarely and skewing class frequencies. Existing methods that utilize Generative Adversarial Networks (GANs) attempt to address this problem. However, they often struggle to generate characters that are both consistent and visually accurate in terms of structural details. To address these issues, this study introduces a skeleton-based balancing GAN (SkelBAGAN), which improves the structural details of the previous method for generating characters. The proposed method introduces three main enhancements: (i) a layer for extracting the character skeleton structure, (ii) an optimized pretrained network using an autoencoder for learning the skeleton distribution, and (iii) refinement of the evaluation function, preserving both the distribution and structural fidelity in the adversarial process. The performance of the proposed model is evaluated against previous methods using the Fréchet Inception Distance (FID) to assess distribution quality and the Structural Similarity Index Measure (SSIM) to evaluate structural fidelity. The results indicate that the proposed methods outperform previous methods in balancing the FID and SSIM metrics. The integration of all enhancements in SkelBAGAN achieves the lowest FID, indicating improved generative quality while maintaining competitive SSIM values. The qualitative study indicates that SkelBAGAN outperforms previous methods in character generation. These results highlight how the skeleton-based improvement of the quality of generated characters enhances the recognition performance for underrepresented Javanese characters in imbalanced datasets. Ultimately, this work contributes to the broader effort to preserve the Javanese script as a vital element of Indonesia’s cultural identity
From Serial to Parallel: Enhancing Needleman-Wunsch Performance through GPU-Based Computing
No full textThe increasing demand for faster bioinformatics analysis calls for more efficient approaches for sequence alignment. In this study, we demonstrate that a GPU-based implementation of the Needleman-Wunsch algorithm can achieve up to 14.8× speedup compared to its traditional CPU-based serial counterpart, without compromising alignment accuracy. By leveraging the parallel processing capabilities and shared memory of an NVIDIA GeForce RTX 3060 Laptop GPU, we significantly accelerated global sequence alignment tasks. Using clinically relevant genes such as NRAS, BRCA1, BRCA2, and Saccharomyces cerevisiae from NCBI ensures realistic alignment challenges and biological significance. Performance evaluation across a wide range of sequence lengths demonstrates the scalability and efficiency of the parallel approach. More importantly, this study provides a unique contribution by showing that a commodity GPU, such as the NVIDIA GeForce RTX 3060 Laptop, can serve as a practical alternative when high-performance computing clusters are unavailable or prohibitively expensive, thereby offering an accessible and cost-effective pathway to high-throughput bioinformatics workflows
The Impact of Squeeze-and-Excitation Blocks on CNN Models and Transfer Learning for Pneumonia Classification Using Chest X-ray Images
No full textPneumonia is one of the leading causes of death due to respiratory tract infections, especially in children and the elderly. Early detection using chest X-ray images is crucial to accelerate diagnosis and treatment, but manual interpretation is often subjective and error-prone. This study evaluates the effect of Squeeze-and-Excitation (SE) Block integration on the performance of a custom Convolutional Neural Network (CNN) model and three popular transfer learning architectures: MobileNetV2, VGG16, and InceptionV3 in X-ray image-based pneumonia classification. A dataset of 5,856 images, taken from Chest X-ray Images (Pneumonia) on Kaggle, was processed through preprocessing, undersampling, and augmentation. Each model was tested in two configurations: without and with SE Block. Evaluation was performed using accuracy, precision, recall, F1-score, and test loss metrics. The results show that SE Block integration improves the performance of most models. The accuracy of the custom CNN increased from 95.17% to 95.88%, MobileNetV2 from 97.18% to 97.59%, and VGG16 from 96.88% to 97.69%. InceptionV3 also saw an accuracy increase from 94.06% to 94.16%, although accompanied by an increase in test loss. SE Block proved effective in strengthening the model's emphasis on important features through an inter-channel recalibration mechanism, especially on efficient architectures like MobileNetV2 and complex models like VGG16. These findings support the development of a more accurate, efficient, and adaptive deep learning-based pneumonia diagnosis system, especially for implementation in healthcare facilities with limited resources.
 
A Data-Driven Comparison of Linear Mixed Model and Mixed Effects Regression Tree Approaches for Dairy Productivity Analysis
No full textDairy productivity studies often involve hierarchical and longitudinal data structures that violate the assumptions of linear regression. This study compares two modeling approaches, Linear Mixed Model (LMM) and Mixed Effects Regression Tree (MERT), in predicting dairy productivity based on the 2024 National Dairy Productivity Survey data. Predictive performance was evaluated using MSEP, RMSEP, MAD, and MAPE, with MERT consistently outperforming LMM in accuracy and robustness. Permutational Multivariate Analysis of Variance (PERMANOVA) test results reinforced this finding, yielding a pseudo-F value of 224.7 and a p-value of 0.001, indicating statistically significant differences in model performance. Key predictors of MERT model included farm altitude, the previous week’s milk production, and the amounts of concentrate feed given, which are part of significant predictor variables in LMM. This finding underscores MERT’s superiority in modeling complex agricultural datasets while providing interpretable insights through data-driven segmentation. The study advocates policy focus in sustainable milk production as well as the availability of high quality of feed and altitude-based dairy farms location to improve milk productivity. Should these focuses implemented by the industry, combined with the MBG Program, Indonesia would be progressing better towards achievement of SDGs Goal 2 and 3
Correlation Analysis of ISO 25010 Modularity, CK Metrics, and Architecture Smells
No full textOpen-source software projects face increasing challenges in maintaining design quality as they evolve, often resulting in technical debt accumulation and reduced maintainability. This study explores the relationship between software modularity, measured using ISO/IEC 25010 quality attributes, Chidamber and Kemerer (CK) object-oriented metrics, and architectural smells (AS) in Java-based open-source software. Six Java-based open-source projects were strategically selected based on varying complexity levels (ranging from 6-994 classes) and different application domains to ensure comprehensive analysis coverage using DesigniteJava to extract AS, CK metrics, and modularity indicators. Correlation analyses showed that architectural smells such as Cyclic Dependency, Ambiguous Interface, and God Component are strongly correlated with CK metrics like Weighted Methods per Class, Depth of Inheritance Tree, and Number of Children. These CK metrics also exhibited strong positive correlations with Cyclomatic Complexity, indicating that structurally complex components also tend to have more complex control logic. Dense Structure was found to negatively correlate with Coupling of Components Conformance, suggesting its effect on modularity compliance. On the other hand, smells like Feature Concentration and Scattered Functionality showed weak or inconsistent correlations with these metrics. The findings highlight the importance of addressing specific architectural smells to improve modularity and software quality
Interdisciplinary Analysis of Machine Learning Applications: Focus on Intent Classification
No full textGiven the rapid growth of machine learning publications on platforms such as arXiv, there is a need for systematic approaches to understand their objectives and contributions. This study aimed to analyze scientific intentions across domains, identify research trends, and evaluate the impact of external contextual enrichment on automatic intent classification. We perform a cross-domain comparison of research objectives, methodological designs, and application scenarios in machine learning publications, focusing on computer science and biology. We propose IntentBERT-Wiki, an enhanced BERT model enriched with contextual knowledge from Wikipedia, designed for intent classification in scientific documents. Our dataset comprises annotated sentences extracted from arXiv articles, categorized according to established rhetorical role taxonomies. The model’s performance is evaluated using standard classification metrics and compared to a baseline BERT model. Experimental results show that IntentBERT-Wiki achieves F1-scores of 95.9% in computer science and 87.4% in biology, with corresponding accuracies of 96.5% and 91.4%, outperforming the baseline. These findings demonstrate that Wikipedia-based contextual enrichment can significantly improve intent classification accuracy, enhance the organization of academic discourse, and facilitate cross-domain knowledge transfers. This study contributes to the understanding of how machine learning research is framed across disciplines and provides a scalable framework for scientific content analysis