Article Per Year (5 Year)
p-Index From 2021 - 2026
0.882
P-Index
This Author published in this journals
All Journal IAES International Journal of Artificial Intelligence (IJ-AI) Jurnal Teknik Informatika Journal of Sustainable Engineering: Proceedings Series Jurnal Teknik Elektro dan Komputer Riau Jurnal Teknik Informatika
Agustinus Jacobus
Department Of Electrical Engineering, Faculty Of Engineering, Universitas Sam Ratulangi, Jln. Kampus Unsrat, Bahu - Manado 95115, INDONESIA
Chemical Engineering, Chemistry & Bioengineering Civil Engineering, Building, Construction & Architecture Computer Science & IT Control & Systems Engineering Electrical & Electronics Engineering Energy Engineering Library & Information Science Mechanical Engineering

Published : 8 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 8 Documents
Search

 1 
Effects of kernels and the proportion of training data on the accuracy of SVM sentiment analysis in lecturer evaluation Daniel Febrian Sengkey; Agustinus Jacobus; Fabian Johanes Manoppo
IAES International Journal of Artificial Intelligence (IJ-AI) Vol 9, No 4: December 2020
Publisher : Institute of Advanced Engineering and Science

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.11591/ijai.v9.i4.pp734-743

Abstract

Support vector machine (SVM) is a known method for supervised learning in sentiment analysis and there are many studies about the use of SVM in classifying the sentiments in lecturer evaluation. SVM has various parameters that can be tuned and kernels that can be chosen to improve the classifier accuracy. However, not all options have been explored. Therefore, in this study we compared the four SVM kernels: radial, linear, polynomial, and sigmoid, to discover how each kernel influences the accuracy of the classifier. To make a proper assessment, we used our labeled dataset of students’ evaluations toward the lecturer. The dataset was split, one for training the classifier, and another one for testing the model. As an addition, we also used several different ratios of the training:testing dataset. The split ratios are 0.5 to 0.95, with the increment factor of 0.05. The dataset was split randomly, hence the splitting-training-testing processes were repeated 1,000 times for each kernel and splitting ratio. Therefore, at the end of the experiment, we got 40,000 accuracy data. Later, we applied statistical methods to see whether the differences are significant. Based on the statistical test, we found that in this particular case, the linear kernel significantly has higher accuracy compared to the other kernels. However, there is a tradeoff, where the results are getting more varied with a higher proportion of data used for training.
Ship-to-Shore Wireless Communication for Asynchronous Data Delivery to the Remote Islands Alwin M. Sambul; Sherwin R.U.A. Sompie; Daniel Febrian Sengkey; Agustinus Jacobus; Alicia A.E. Sinsuw
Journal of Sustainable Engineering: Proceedings Series Vol 1 No 1 (2019)
Publisher : Fakultas Teknik Universitas Sam Ratulangi

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35793/joseps.v1i1.13

Abstract

Nowadays, many people who live in remote islands of Indonesia are still facing difficulties in terms of access to information. In the locations where end-to-end communication is not available, the asynchronous approach can be utilized to send information in the form of digital data. In some areas, we could utilize passenger ships or ferries as physical carriers to deliver digital data to the people in the remote islands which are located at a particular range of distance from the ship’s passing routes. This paper reports the channel performance of long-range WiFi connection oversea at 5 GHz using the real ship’s route at the North Sulawesi province‘s water in Indonesia as a sample scenario. The measurement results showed that the most stable ship-to-shore communication can be achieved in ±15 minutes at the maximum distance between the ship and shore of about 4 km. The maximum channel capacity was 120 Mbps for upload (from ship to shore) and 53 Mbps for download (from shore to ship), which is enough to deliver gigabytes of information to the people at the islands every time the ship passes by.
Implementing Support Vector Machine Sentiment Analysis to Students' Opinion toward Lecturer in an Indonesian Public University Daniel Febrian Sengkey; Agustinus Jacobus; Fabian Johanes Manoppo
Journal of Sustainable Engineering: Proceedings Series Vol 1 No 2 (2019)
Publisher : Fakultas Teknik Universitas Sam Ratulangi

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35793/joseps.v1i2.27

Abstract

Student feedback is an important evaluation tool for quality improvement. Moreover, in Indonesian higher education system there is an assessment regulation that puts special attention to the availability of the student feedback system. However, parts of the questionnaire are in the form of descriptive text that requires more effort for analysis. This situation leads to a very tiresome work in case of the number of documents reaches several hundred or even thousands. There were some efforts to apply computer-assisted classification by utilizing machine learning, however, most of them only analyzed English documents. Only a handful that studied the classification of documents in Bahasa Indonesia. In reality, we found some cases where the students used mixed languages while filling the evaluation forms. Therefore, in this study, we expand the application of text classification by using Support Vector Machne (SVM) to cases of student feedback in mixed languages. The model was built computationally and from the test, we get 74% accuracy and 0.46 Kappa value.
Analisis Sentimen Berbasis Aspek Ulasan Produk Menggunakan CNN dan Bidirectional LSTM: Aspect-Based Sentiment Analysis Product Review Using CNN and Bidirectional LSTM Obedient Putro; Agustinus Jacobus; Feisy Kambey
Jurnal Teknik Informatika Vol. 20 No. 2 (2025): Jurnal Teknik Informatika
Publisher : Universitas Sam Ratulangi

Show Abstract | Download Original | Original Source | Check in Google Scholar

Abstract

 Abstract — The COVID-19 pandemic has transformed consumer lifestyles in Indonesia, notably increasing the use of e-commerce platforms due to social restrictions. This shift has influenced how consumers evaluate product quality, making consumer reviews a crucial element in purchasing decisions. Traditional sentiment analysis falls short in providing detailed insights into product aspects, making Aspect-Based Sentiment Analysis (ABSA) a promising solution. Deep learning models like Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) offer high accuracy in sentiment analysis. This study analyzes consumer sentiment towards e-commerce product aspects in Indonesia by applying ABSA, addressing the challenges of implementation in the Indonesian language, and measuring the accuracy and effectiveness of the hybrid CNN Bi-LSTM model. The methodology includes dataset preprocessing, aspect extraction and sentiment classification, data training and prediction, and model evaluation. The results show that the CNN Bi-LSTM model achieves an average accuracy of 90% for aspect extraction and 92% for sentiment classification. In conclusion, despite dataset limitations, optimal data preparation and the hybrid model effectively facilitate ABSA.   Key Word — ABSA; CNN; Bi-LSTM; Accuracy; Hybrid Model   Abstrak — Pandemi COVID-19 telah mengubah gaya hidup konsumen di Indonesia, terutama melalui peningkatan penggunaan platform e-commerce akibat pembatasan sosial. Perubahan ini mempengaruhi cara konsumen menilai kualitas produk, menjadikan ulasan konsumen elemen kunci dalam keputusan pembelian. Analisis sentimen tradisional tidak memadai untuk memberikan pemahaman mendetail tentang aspek produk, sehingga Aspect-Based Sentiment Analysis (ABSA) menjadi solusi yang menjanjikan. Model deep learning seperti Convolutional Neural Network (CNN) dan Bidirectional Long Short-Term Memory (Bi-LSTM) menawarkan akurasi tinggi dalam analisis sentimen. Penelitian ini menganalisis sentimen konsumen terhadap aspek produk e-commerce di Indonesia dengan menerapkan ABSA, menghadapi tantangan implementasi dalam bahasa Indonesia, dan mengukur akurasi serta efektivitas model hybrid CNN Bi-LSTM. Metodologi meliputi preprocessing dataset, aspect extraction dan sentiment classification, data training dan prediction, serta evaluasi model. Hasil menunjukkan model CNN Bi-LSTM memiliki akurasi rata-rata 90% untuk aspect extraction dan 92% untuk sentiment classification. Kesimpulannya, data preparation optimal meskipun ada keterbatasan dataset, dan model hybrid efektif untuk ABSA.   Kata kunci — ABSA; CNN; Bi-LSTM; Akurasi; Hybrid Model
Comparison of The Performance of Fasttext and Word2Vec Methods in Detecting Fake News: Perbandingan Kinerja Metode FastText dan Word2Vec dalam Mendeteksi Berita Palsu Try Iksan; Agustinus Jacobus; Fransisca Pontoh
Jurnal Teknik Elektro dan Komputer Vol. 15 No. 1 (2026): Jurnal Teknik Elektro dan Komputer
Publisher : Universitas Sam Ratulangi

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35793/jtek.v15i1.62468

Abstract

Abstract — The spread of fake news (hoaxes) on social media has a significant negative impact on society, such as a decline in public trust and increased uncertainty about information. This study aims to develop and compare accurate and reliable Indonesian-language fake news detection systems, with the hope of improving media literacy among the public. The methods used include collecting several datasets of fake and authentic news, data preprocessing (cleaning, tokenisation, lemmatisation, stopword removal), and applying two word embedding algorithms, FastText and Word2Vec, with two architectures (CBOW and Skipgram). The classification model used is Bi-LSTM, and evaluation is conducted using accuracy, precision, recall, and F1-score metrics. The results show that both algorithms can produce high-accuracy fake news detection models on large datasets (FastText >85%, Word2Vec >87%), but performance decreases on small datasets due to overfitting. This study provides theoretical and practical contributions to the evaluation of word embedding algorithm performance for detecting Indonesian-language fake news based on NLP. In conclusion, the comparison results show that the evaluated word embedding approach is effective in identifying Indonesian-language fake news and can serve as a reference for algorithm selection in the development of future fake news detection technology. Key words — Bi-LSTM; FastText; hoax; NLP; Word2Vec   Abstrak — Penyebaran berita palsu (hoaks) di media sosial menimbulkan dampak negatif yang signifikan bagi masyarakat, seperti menurunnya kepercayaan publik dan meningkatnya ketidakpastian informasi. Penelitian ini bertujuan untuk mengembangkan dan membandingkan sistem deteksi berita palsu berbahasa Indonesia yang akurat dan andal, dengan harapan dapat meningkatkan literasi media masyarakat. Metode yang digunakan meliputi pengumpulan beberapa dataset berita palsu dan asli, praproses data (cleaning, tokenisasi, lemmatisasi, penghapusan stopwords), serta penerapan dua algoritma word embedding FastText dan Word2Vec dengan dua arsitektur (CBOW dan Skipgram). Model klasifikasi yang digunakan adalah Bi-LSTM, dan evaluasi dilakukan menggunakan metrik akurasi, precision, recall, dan F1-score. Hasil penelitian menunjukkan bahwa kedua algoritma mampu menghasilkan model deteksi berita palsu dengan akurasi tinggi pada dataset besar (FastText >85%, Word2Vec >87%), namun performa menurun pada dataset kecil akibat overfitting. Penelitian ini memberikan kontribusi teoretis dan praktis dalam evaluasi performa algoritma word embedding untuk deteksi berita palsu berbahasa Indonesia berbasis NLP. Kesimpulannya, hasil perbandingan menunjukkan bahwa pendekatan word embedding yang dievaluasi efektif dalam mengidentifikasi berita palsu berbahasa Indonesia dan dapat menjadi acuan pemilihan algoritma untuk pengembangan teknologi deteksi berita palsu di masa depan. Kata kunci — Bi-LSTM; FastText; hoaks; NLP; Word2Vec
Gated Recurrent Unit for Clickbait and Non-Clickbait News Headlines Classification: Gated Recurrent Unit Untuk Klasifikasi Judul Berita Clickbait dan Non-Clickbait Desriyanti Dea; Feisy Diane Kambey; Agustinus Jacobus
Jurnal Teknik Elektro dan Komputer Vol. 15 No. 2 (2026): Jurnal Teknik Elektro dan Komputer
Publisher : Universitas Sam Ratulangi

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35793/jtek.v15i2.64652

Abstract

Abstract — Clickbait refers to the practice of crafting sensational headlines to entice readers to click on links and read articles, often at the expense of accurately representing the underlying content. Amid fierce competition among online news portals and readers’ tendency to focus only on headlines, this phenomenon can mislead audiences, thereby necessitating an automated system capable of distinguishing clickbait headlines from non-clickbait ones. This study classifies Indonesian news headlines using a Gated Recurrent Unit (GRU) architecture and compares two pre-trained word embedding models, FastText and Word2Vec. The data used in this study are taken from the CLICK-ID dataset on Kaggle and comprise 15,000 news headlines. The results show that GRU combined with Word2Vec provides the best performance, achieving 78.96% accuracy, 78.80% precision, 78.96% recall, and a 78.86% F1-score, while GRU with FastText achieves 77.04% accuracy, 77.06% precision, 77.04% recall, and a 77.05% F1-score. In the task of classifying Indonesian clickbait news headlines, the use of Word2Vec word embeddings in a GRU-based model is superior to FastText, as it yields higher classification accuracy as well as better computational efficiency. Key words— Clickbait; FastText; GRU; Text Classification; Word2Vec.   Abstrak — Clickbait adalah praktik penyusunan judul yang sengaja dibuat sensasional agar pembaca tertarik untuk mengklik tautan dan membaca artikel, namun kerap tidak mewakili isi berita secara utuh. Di tengah persaingan portal berita dan rendahnya kebiasaan membaca isi secara lengkap, fenomena ini berpotensi menyesatkan pembaca sehingga diperlukan sistem otomatis untuk membedakan judul clickbait dan non-clickbait. Penelitian ini mengklasifikasikan judul berita berbahasa Indonesia menggunakan arsitektur Gated Recurrent Unit dan membandingkan dua pretrained word embedding , FastText dan Word2Vec. Data yang digunakan berasal dari kaggle sebanyak 15.000 judul. Hasil menunjukkan kombinasi GRU dengan Word2Vec memberikan kinerja terbaik dengan akurasi 78,96 %, presisi 78,8%, recall 78,96 %, dan F1-Score 78,86 %. Sementara itu, GRU dengan FastText mencapai akurasi 77,04 %, presisi 77,06 %, recall 77,04 %, dan F1-Score 77,05 %. Pada tugas klasifikasi judul berita clickbait berbahasa Indonesia, penggunaan word embedding Word2Vec pada model GRU lebih unggul dibandingkan FastText karena mampu memberikan akurasi klasifikasi yang lebih tinggi sekaligus efisiensi komputasi yang lebih baik. Kata kunci — Clickbait; FastText; GRU; Klasifikasi Teks; Word2Vec
Klasifikasi Multilabel Artikel Ilmiah Bidang Komputer Menggunakan SciBERT dan TextCNN Regina Nathania Agussalim; Agustinus Jacobus; Sherwin R. U. A. Sompie
Riau Jurnal Teknik Informatika Vol. 5 No. 2 (2026): Juli 2026
Publisher : Universitas Pasir Pengaraian

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30606/rjti.v5i2.4705

Abstract

The increasing number of scientific articles makes the process of searching, filtering, and organizing relevant literature more challenging. This problem becomes more complex because one scientific article can be associated with more than one research topic. This study aims to implement and evaluate a SciBERT-TextCNN model for multilabel classification of scientific articles in the Computer Science field based on title and abstract metadata. The dataset used was arXiv article metadata obtained from Kaggle and limited to Computer Science categories. The research stages included data preprocessing, Multi-Label Random Under-Sampling, tokenization using SciBERT tokenizer, model training, threshold tuning, and evaluation. The model was tested using five scenarios: SciBERT frozen and fine-tuning the last 1, 2, 3, and 4 layers. The best performance was obtained by fine-tuning the last 2 SciBERT layers, achieving precision of 0.708, recall of 0.726, F1-score of 0.715, macro F1-score of 0.649, and Hamming Loss of 0.02333. These results show that SciBERT-TextCNN can be used to support automatic multilabel classification of scientific articles.
Analisis Sentimen Berbasis Aspek Ulasan GoFood di X dengan Ekstraksi Aspek Rule-Based dan Klasifikasi IndoBERT Nazlah Ayunda Syahputri Jacub; Agustinus Jacobus; Muhamad Dwisnanto Putro
Riau Jurnal Teknik Informatika Vol. 5 No. 3 (2026): November 2026
Publisher : Universitas Pasir Pengaraian

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30606/rjti.v5i3.4822

Abstract

Social media reviews provide insights into user perceptions of food delivery services, but general sentiment analysis does not identify specific service aspects. This study applies Aspect-Based Sentiment Analysis (ABSA) using a hybrid Rule-Based and IndoBERT approach to GoFood reviews on X. Rule-Based methods extract aspects using keywords and structured rules, while IndoBERT classifies sentiment based on textual context. From 28,763 Indonesian-language tweets obtained from Kaggle, 5,657 aspect–sentiment pairs were obtained after preprocessing, aspect extraction, auditing, and manual verification. The dataset covers four aspects: price, service, food quality, and application. Using a StratifiedGroupKFold split of 70:15:15, IndoBERT achieved 90.31% accuracy and a Macro F1-score of 0.8883 on 846 test samples. Food quality achieved the highest aspect-specific accuracy (92.62%), followed by price (90.40%), service (90.00%), and application (87.88%). Price was the most frequently identified aspect, while positive sentiment dominated food quality (60.4%) and neutral sentiment dominated the other three aspects. These findings provide detailed aspect-level insights into user perceptions and can help identify service aspects to maintain or examine further. However, the impact on service quality improvement has not been directly evaluated.
Co-Authors Alicia A.E. Sinsuw Alwin M. Sambul Daniel Febrian Sengkey Desriyanti Dea Fabian Johanes Manoppo Feisy Diane Kambey Feisy Kambey Fransisca Pontoh Nazlah Ayunda Syahputri Jacub Obedient Putro Putro, Muhamad Dwisnanto Regina Nathania Agussalim Sherwin R. U. A. Sompie Sherwin R.U.A. Sompie Try Iksan