Jurnal RESTI (Rekayasa Sistem dan Teknologi Informasi)

DiG-MFV: Dual-integrated Graph for Multilingual Fact Verification

2025-07-28T10:20:00+00:00

The 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.

Face Dermatological Disorder Identification with YoloV5 Algorithm

2025-07-26T10:19:52+00:00

Dermatological disorders are common in humans. The accurate identification of skin diseases is paramount for determining the most efficacious treatment. This system can screen images of skin diseases on the face and provide analysis results in the form of object detection. Dermatological disorders of the face are classified into six categories: acne nodules, melasma, filiform warts, milia, papules, and pustules. The YoloV5 algorithm was selected because of its effectiveness in live-detection tasks. The image-enhancement process involves the implementation of two methodologies: sharpening and histogram equalization. The former adjusts the brightness values whereas the latter adjusts the contrast values. The dataset comprised 1,223 images of skin diseases, with 947 images allocated for training and 276 for validation. The optimal mAP of the filiform wart class was determined to be 87.6%, with values of 76.7% for pustules, 72% for papules, 71% for milia, 68% for nodules, and 38.2% for melasma, representing the lowest value. The low mAP of melasma was attributed to the abstract image data type and complexity of localization. The congruence of object features and disparity in data variance has the potential to influence outcomes.

Real-time Emotion Recognition Using the MobileNetV2 Architecture

2025-07-17T10:19:17+00:00

Facial recognition technology is now advancing quickly and is being used extensively in a number of industries, including banking, business, security systems, and human-computer interface. However, existing facial recognition models face significant challenges in real-time emotion classification, particularly in terms of computational efficiency and adaptability to varying environmental conditions such as lighting and occlusion. Addressing these challenges, this research proposes a lightweight, yet effective deep learning model based on MobileNetV2 to predict human facial emotions using a camera in real time. The model is trained on the FER-2013 dataset, which consists of seven emotion classes: anger, disgust, fear, joy, sadness, surprise, and neutral. The methodology includes deep learning-based feature extraction, convolutional neural networks (CNN), and optimization techniques to enhance real-time performance on resource-constrained devices. Experimental results demonstrate that the proposed model achieves a high accuracy of 94.23%, ensuring robust real-time emotion classification with a significantly reduced computational cost. Additionally, the model is validated using real-world camera data, confirming its effectiveness beyond static datasets and its applicability in practical real-time scenarios. The findings of this study contribute to advancing efficient emotion recognition systems, enabling their deployment in interactive AI applications, mental health monitoring, and smart environments. Real-world camera data is also used to evaluate the model, demonstrating its usefulness in real-time applications and its efficacy beyond static datasets. The results of this work advance effective emotion identification systems, making it possible to use them in smart settings, interactive AI applications, and mental health monitoring.

Optimizing Sentiment Analysis for Lombok Tourism Using SMOTE and Chi-Square with Machine Learning

2025-07-13T10:18:44+00:00

Tourism is a vital economic sector for Lombok Island, which is renowned for its natural beauty and cultural richness as a top destination. The rapid growth of tourism in Lombok requires a deep understanding of tourists' perceptions and sentiments to ensure an optimal service quality. The sentiment analysis of online reviews is valuable for identifying service strengths and weaknesses and addressing tourists' needs more effectively. This not only enhances tourist satisfaction, but also aids in the design of more effective marketing strategies. However, text data analysis from online reviews presents unique challenges such as noise, class imbalance, and numerous features that may affect classification results. Therefore, this study aims to classify tourist sentiment toward Lombok tourism using machine learning methods combined with feature selection and oversampling techniques. This study focuses on optimizing sentiment analysis of tourism-related tweets using a combination of SMOTE oversampling and Chi-Square feature selection on improving classification performance without hyperparameter tuning. The study applies machine learning methods, such as SVM and Naïve Bayes, with feature selection and oversampling using Chi-Square and SMOTE. The dataset used was sentiment data regarding Lombok tourism obtained from Twitter in 2023, consisting of 940 instances divided into three classes: Negative, Neutral, and Positive. The research findings show that the use of SMOTE and Chi-Square can improve the accuracy of the SVM and Naive Bayes methods. Without optimization, the SVM method achieved an accuracy of 73.93% and a Naive Bayes of 67.02%. After optimization with SMOTE and Chi-Square, the accuracy increased for SVM by 90% and Naive Bayes by 84% to classify tourist sentiment towards Lombok tourism. The implications indicate that combining data balancing using SMOTE with feature selection via Chi-Square effectively improves the performance of sentiment classification models for tourist opinions on Lombok's tourism.

Enhancing Areca Nut Detection and Classification Using Faster R-CNN: Addressing Dataset Limitations with Haar-like Features, Integral Image, and Anchor Box Optimization

2025-06-30T16:27:14+00:00

The classification and detection of areca nuts are essential for agriculture and food processing to ensure product quality and efficiency. The manual classification of areca nuts is time-consuming and prone to human error. For a more accurate and efficient automated approach, a deep learning-based framework was proposed to address these challenges. This study optimizes the Faster R-CNN by integrating Haar-like features and integral images to enhance object detection. However, dataset limitations, including low image quality, inconsistent lighting, cluttered backgrounds, and annotation inaccuracies, affect the model performance. In addition, the small dataset size and class imbalance hindered generalization. The Faster R-CNN model was trained with and without Haar-like Features and Integral Image enhancement. Performance was evaluated based on training loss, accuracy, precision, recall, F1-score, and mean average precision (mAP). The effects of the dataset limitations on detection performance were also analyzed. The optimized model achieved better stability, with a final training loss of 0.2201, compared to 0.1101 in the baseline model. Accuracy improved from 62.60% to 73.60%, precision from 0.6161 to 0.7261, recall from 0.3094 to 0.4194, F1-score from 0.2307 to 0.3407, and mAP from 0.1168 to 0.2268. Despite these improvements, dataset constraints remain a limiting factor. While the integration of Haar-like features and integral images into faster R-CNN contributes to detection accuracy, the study also reveals that high-resolution images, precise annotations, and dataset scale significantly amplify model performance.