Article Per Year (5 Year)
p-Index From 2021 - 2026
2.024
P-Index
This Author published in this journals
All Journal International Journal of Electrical and Computer Engineering Harmonia: Journal of Research and Education International Journal of Advances in Intelligent Informatics JOIN (Jurnal Online Informatika) JOIV : International Journal on Informatics Visualization Jurnal RESTI (Rekayasa Sistem dan Teknologi Informasi) International Journal of Artificial Intelligence Research Journal of Applied Data Sciences Bulletin of Social Informatics Theory and Application Advance Sustainable Science, Engineering and Technology (ASSET)
Nafalski, Andrew
Unknown Affiliation
Arts Chemistry Computer Science & IT Control & Systems Engineering Decision Sciences, Operations Research & Management Education Electrical & Electronics Engineering Engineering Industrial & Manufacturing Engineering Social Sciences

Published : 13 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 13 Documents
Search

 1   2 
Attention-Enhanced Convolutional Neural Network for Context Extraction in Andersen's Fairy Tales Daniati, Erna; Wibawa, Aji Prasetya; Irianto, Wahyu Sakti Gunawan; Nafalski, Andrew
JOIV : International Journal on Informatics Visualization Vol 9, No 6 (2025)
Publisher : Society of Visual Informatics

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.62527/joiv.9.6.4056

Abstract

Event extraction in classic literature and fairy tales remains highly challenging due to their non-linear plot structures, archaic linguistic expressions, and intricate character interactions, while advances in modern NLP still show limitations in capturing subtle narrative cues in historical texts. This study aims to address these gaps by developing an event extraction model tailored to the narrative characteristics of Hans Christian Andersen’s fairy tales. We propose a BERT-enhanced Context-aware Convolutional Neural Network (CNN) that integrates an attention mechanism to overcome the limited contextual range of traditional CNNs. The model leverages BERT’s contextual embeddings enriched with an attention layer to detect event triggers, character relations, and narrative transitions across nonlinear storylines. A hybrid dataset was constructed through system-generated annotations refined via manual verification and combined with AN/an cartoon-based representations for model training and final testing. Experimental results show that the proposed model surpasses both the CNN-only baseline and a rule-based approach, achieving precision of 0.92, recall of 0.89, F1-score of 0.90, and accuracy of 0.91, outperforming the CNN baseline (0.85/0.82/0.83/0.84) and rule-based system (0.78/0.75/0.76/0.77). These findings highlight the effectiveness of context-aware representations for processing literary narratives and demonstrate interdisciplinary relevance to digital humanities and AI-based storytelling, with future extensions envisioned for multilingual settings and genre-specific adaptations.
Optimizing Transformer Model FlanT5 for Multi-Question Answering with Context-Aware Learning Rate Suryanto, Tri Lathif Mardi; Wibawa, Aji Prasetya; Hariyono, Hariyono; Nafalski, Andrew
Jurnal RESTI (Rekayasa Sistem dan Teknologi Informasi) Vol 9 No 6 (2025): December 2025
Publisher : Ikatan Ahli Informatika Indonesia (IAII)

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.29207/resti.v9i6.6985

Abstract

This study investigates the performance of FlanT5-based transformer models in handling Multiple-Question Answering (M-QA) tasks, in which multiple semantically related questions must be addressed with a single cohesive answer. Unlike traditional QA systems that focus on one-to-one question-answer pairs, the M-QA approach challenges the model to understand contextual relationships across several questions tied to the same topic. A custom dataset was developed with shared context, grouped questions, and a unified answer to train and evaluate the model. The FlanT5 architecture was fine-tuned using different learning rates (0.0001, 0.0002, 0.0003) to explore the effect of training configurations on model performance. The evaluation was conducted using the ROUGE-1, ROUGE-2, ROUGE-L, and ROUGE-Lsum metrics. The results indicate that a learning rate of 0.0003 provides the optimal performance, achieving a ROUGE-Lsum score of 0.7390. This study confirms the capability of instruction-tuned transformers to manage complex summarization scenarios that require contextual coherence. The findings are relevant for real-world applications such as intelligent digital assistants, clinical decision support, and educational chatbots. Furthermore, this study emphasizes the importance of hyperparameter tuning in improving the effectiveness of question-driven summarization systems for scalable and efficient deployment.
Fixed sherwood duel optimization for time series imputation Utama, Agung Bella Putra; Wibawa, Aji Prasetya; Handayani, Anik Nur; Nafalski, Andrew
International Journal of Advances in Intelligent Informatics Vol 12, No 1 (2026): February 2026
Publisher : Universitas Ahmad Dahlan

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.26555/ijain.v12i1.2396

Abstract

Missing values remain a persistent challenge in time-series data, particularly within large-scale monitoring systems where reliable forecasting and evaluation are essential. Incomplete records often arise from irregular reporting, infrastructure limitations, or system failures, leading to biased analyses and inaccurate predictions. Traditional imputation methods, such as mean, median, and mode substitution, provide computational efficiency but oversimplify temporal structures. At the same time, more advanced approaches, including Multiple Imputation by Chained Equations (MICE) and K-Nearest Neighbors (KNN), offer improvements yet remain sensitive to data distribution and model configuration. To address this gap, this study introduces Sherwood Duel Optimization (SDO). This socio-inspired framework reconceptualizes imputation as a deterministic duel-based optimization problem. In its fixed form, SDO generates multiple candidate imputations and selects the most robust replacement value using a composite multi-metric scoring mechanism that integrates forecasting accuracy and explanatory power. The framework was evaluated using multivariate educational time-series data and further validated across heterogeneous SDG-related domains, and compared against classical and advanced baselines across three forecasting models. Experimental results demonstrate that SDO consistently outperforms existing methods, reducing forecasting error (MAPE) by more than 40%, achieving the lowest RMSE, and producing R² values exceeding 0.95. Statistical testing confirms that these improvements are significant across experimental configurations. These findings highlight the potential of SDO as a reliable, interpretable, and computationally efficient optimization-based imputation framework. By strengthening data reliability at the reconstruction stage, SDO enhances the credibility of downstream forecasting and decision-making in institutional and sustainability-oriented monitoring systems.
Co-Authors Aji Prasetya Wibawa Andini, Nurul Fajriah Anik Nur Handayani Anusua Ghosh, Anusua Atmaja, Nimas Hadi Darwis, Herdianti Denis Eka Cahyani Dewandra, Aderyan Reynaldi Fahrezza Dewi, Popy Maulida Didik Dwi Prasetya Dwiyanto, Felix Andika Erna Daniati Hariyono Hariyono Haviluddin Haviluddin Hery Widijanto Izdihar, Zahra Nabila Langlang Gumilar Marida, Tyas Agung Cahyaning Muladi Nedic, Zorica Ningtyas, Yana Nurfadila, Piska Dwi Purnawansyah Purnawansyah Putra Utama, Agung Bella Raja, Roesman Ridwan Saputra, Irzan Tri Triyanna Widiyaningtyas Utama, Agung Bella Putra Utomo Pujianto Wahyu Sakti Gunawan Irianto Zaeni, Ilham Ari Elbaith