Claim Missing Document
Check
Articles

Benchmarking Deepseek-LLM-7B-Chat and Qwen1.5-7B-Chat for Indonesian Product Review Emotion Classification Nurohim, Galih Setiawan; Setyadi, Heribertus Ary; Fauzi, Ahmad
Journal of Applied Informatics and Computing Vol. 9 No. 6 (2025): December 2025
Publisher : Politeknik Negeri Batam

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30871/jaic.v9i6.11369

Abstract

Upon completing their shopping experience on an e-commerce platform, users have the opportunity to leave a review. By analyzing reviews, businesses can gain insight into customer emotions, while researchers and policymakers can monitor social dynamics. Large Language Models (LLMs) utilization is identified as a promising methodology for emotion analysis. LLMs have revolutionized natural language processing capabilities, yet their performance in non-English languages, such as Indonesian, necessitates a comprehensive evaluation. This research objective is to perform a comprehensive analysis and comparison of Deepseek-LLM-7B-Chat and Qwen1.5-7B-Chat, two prominent open-source Large Language Models, for the emotion classification of Indonesian product reviews. By leveraging the PRDECT-ID dataset, this study evaluates the performance of both models in a few-shot learning scenario through prompt engineering. The methodology outlines the data preprocessing pipeline, a detailed few-shot prompt engineering strategy tailored to each model's characteristics, model inference execution, and performance assessment using the accuracy, precision, recall, and F1-score metrics. Analytical results reveal DeepSeek achieved an accuracy of 43.41%, exhibiting a considerably superior ability to comprehend instructions compared to Qwen, which attained a maximum accuracy of only 20.35% and often yielded near-random predictions. An in-depth error analysis indicates that this performance gap is likely attributable to factors such as pre-training data bias and tokenization mismatches with the Indonesian language. This research offers empirical evidence regarding the comparative strengths and weaknesses of DeepSeek and Qwen, providing a diagnostic benchmark that underscores the significance of instruction tuning and robust multilingual representation for Indonesian NLP tasks.
Multi Criteria Decision Making Method For Developing Smart Indonesia Program Scholarship Recipient Candidate System Supriyanta, Supriyanta; Sutanto, Yusuf; Susilo, Dahlan; Setyadi, Heribertus Ary; Syukron, Akhmad
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.9.6.3706

Abstract

The Government of Indonesia is continuously striving to improve its education quality with the provision of scholarship programs, one of which is the Smart Indonesia Program (SIP). Students' interest in obtaining SIPs is increasing, but the selection process still relies on conventional methods. Without adequate IT support, the selection process for SIP scholarship candidates will be complex, less objective, and somewhat unfair. State Vocational High School (SVHS) 5 Surakarta was selected as a case study for this research to establish the selection process and the data collection methods used in previous years. The research aims to develop a Decision Support System (DSS) to assist in nominating students deemed eligible for SIP scholarship recommendations. The applied methods include Analytical Hierarchy Process (AHP) and Multi-Objective Optimization by Ratio Analysis (MOORA). Four criteria have been set in this DSS: card ownership status, total parental income, household income, and number of siblings. Each of which is further broken down into several sub-criteria and assigned a value for use in the AHP process. Upon comparing data from 2021 to 2023, it was found that the accuracy in 2021 was 92.9%, in 2022 it reached 94.7%, and in 2023 it recorded 92.3%. Based on the results of this system accuracy test, it can be concluded that the AHP and MOORA methods can be used to objectively produce recommendations for students eligible for SIP scholarships, based on the input criteria.
Prediksi Harga Perumahan Menggunakan Metode Principal Component Analysis dan Random Forest Regresi Sutanto, Yusuf; Al Amin, Budi; Ary Setyadi, Heribertus; Eka Purnama, Bambang
Jurnal Teknologi Informasi dan Ilmu Komputer Vol 12 No 6: Desember 2025
Publisher : Fakultas Ilmu Komputer, Universitas Brawijaya

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.25126/jtiik.2025126

Abstract

Penentuan harga merupakan salah satu aspek krusial dalam kegiatan pengembangan properti mengingat hal tersebut akan mempengaruhi margin keuntungan yang diperoleh pengembang dan pilihan pembelian properti.  Selama bertahun-tahun, prediksi harga rumah telah menjadi topik penelitian utama, karena permintaan rumah terus meroket. Sangat penting untuk mengembangkan kerangka kerja yang sesuai yang memungkinkan pembeli dan penjual untuk membuat keputusan cepat dalam hal membeli atau menjual properti. Dalam penelitian ini menggunakan metode Principal Component Analysis (PCA) dan Random Forest (RF), dengan tujuan untuk melakukan analisis akurasi penggunaan kedua metode dalam prediksi harga rumah dan untuk mengetahui pengaruh penggunaan PCA dalam mengoptimalkan metode random forest. Data yang digunakan adalah harga rumah di kota Surakarta berdasarkan hasil scraping data di situs propertygurugroup.com. Hasil analisis menunjukkan bahwa jumlah penjualan rumah tertinggi adalah daerah Plesungan, dan penjualan rumah yang memiliki sertifikat hak milik juga paling tinggi. Dari sepuluh variabel yang ada, luas tanah dan bangunan paling berpengaruh terhadap harga jual. Hasil pelatihan model menunjukkan bahwa peggabungan metode RF dan PCA memiliki nilai yang lebih optimal dibanding hanya menggunakan metode RF saja. Tingkat kesalahan dalam metode  PCA lebih kecil, dengan rerata 0,0257 maka nilainya lebih konsisten dibanding hanya menggunakan metode RF yang nilai kesalahannya lebih besar dengan rerata 0,0332. Waktu pelatihan model menggunakan PCA lebih cepat (5005,75) dibanding hanya menggunakan metode RF (6099,25).   Abstract Determining prices is one of the crucial aspects in property development activities considering that this will affect the profit margin obtained by developers and property purchase choices. Over the years, home price prediction has been a major research topic, as demand for homes continues to increase. It is important to develop a suitable framework that allows buyers and sellers to make quick decisions when it comes to buying or selling a property. This research uses the Principal Component Analysis (PCA) and Random Forest (RF) methods, with the aim of accuracy analyzing using both methods in predicting housing prices and to determine the effect of using PCA in optimizing the random forest method. The data used is house prices in Surakarta city based on data scraping results on propertygurugroup.com site. The analysis results show that the highest house sales is in Plesungan area, and houses sale with ownership certificates is also the highest. Of the ten variables, land area and building have the most influence on selling price. Model training results show that combination of RF and PCA methods has a more optimal value than using only RF method. Error rate in the PCA method is smaller, with an average of 0.0257, so the value is more consistent than just using the RF method, which has a larger error value with an average of 0.0332. The model training time using PCA is faster (5005.75) than just using the RF method (6099.25).
Capacity Building on Waste Management and TrashGo Application Utilization for the Community in Joglo, Surakarta Susanti, Nani Irma; Sutanto, Yusuf; Suseno, Adnan Terry; Setyadi, Heribertus Ary
DIKDIMAS : Jurnal Pengabdian Kepada Masyarakat Vol. 4 No. 3 (2025): DIKDIMAS : JURNAL PENGABDIAN KEPADA MASYARAKAT VOL 4 NO 3 DESEMBER 2025
Publisher : Asosiasi Profesi Multimedia Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.58723/dikdimas.v4i3.556

Abstract

Background: The transaction mechanism for scrap items designates collectors as the purchasing party and community members or scavengers as the selling party. In operating their business, they still rely on manual methods, involving circulating door-to-door and relying on phone calls from regular customers. The current situation requires residents to wait for collectors to circulate their area and leaves them uninformed about the unpredictable and unstable pricing of scrap goods.Aims: This study aims to provide education provision regarding waste and scrap, followed by a workshop detailing the operation of the TrashGo application that was developed during prior activities.Methods: Asset-Based Community Development (ABCD) which consists of five stages: discovery, dream, design, define, and destiny.Result: Participants understood the business side of waste and scrap. Both collectors and the community were able to effectively utilize the TrashGo application for conducting scrap material transactions. A 95% level of understanding regarding the scrap business and application usage was achieved, signifying the training's success and utility.Conclusion: Data from questionnaires filled out by all participants show that this activity's targets were met. Based on the satisfaction survey, it can be concluded that participants were pleased and felt supported by the program.
Utilizing Data Mining Approach For Hypertension Diagnosis Classification Pudji Widodo; Heribertus Ary Setyadi; Hartati Dyah Wahyuningsih; Sundari Sundari
Jurnal Teknologi Informasi dan Terapan Vol 12 No 1 (2025): June
Publisher : Jurusan Teknologi Informasi Politeknik Negeri Jember

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.25047/jtit.v12i1.446

Abstract

Hypertension is one of the factors contributing to the highest death rates from non-communicable diseases in various countries. Every year, the number of hypertension sufferers increases significantly. It is estimated that in 2025, the number of hypertension sufferers will reach 1.5 billion individuals. Data mining aims to identify patterns that can help in decision making, classification, and prediction. One of the well-known algorithms or methods for classification is the Support Vector Machine (SVM). The SVM method aims to find the best hyperplane or decision boundary function that can separate two or more classes of data in the input space. This research purpose is to determine the classification results and accuracy of the diagnosis of hypertension using the SVM method. Eleven attributes used include age, smoking habits, physical activity, sugar consumption, salt consumption, fat consumption, alcohol consumption, lack of fruit and vegetable consumption, systolic and diastolic blood pressure. This research will utilize Jupyter Notebook tools and Python programming language as research tools. The SVM method was trained with various kernel attributes and hyperparameters to produce the best model. From the results it is known that the RBF kernel used with parameters ???? = 100 and ???? = 0.1 produces an accuracy of 97.5% which is the best model in classifying hypertension. From these results it can be concluded that the SVM method is able to produce a very good classification of hypertension diagnosis and can provide a diagnosis to detect hypertension early
UTILIZING END USER DEVELOPMENT METHOD FOR DEVELOPING PENCAK SILAT ORGANIZATION INFORMATION SYSTEMS Setyadi, Heribertus Ary; Wahyuningsih, Hartati Dyah; Nurohim, Galih Setiawan; Sundari, Sundari
Jurnal Pilar Nusa Mandiri Vol. 21 No. 2 (2025): Pilar Nusa Mandiri : Journal of Computing and Information System Publishing Pe
Publisher : LPPM Universitas Nusa Mandiri

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33480/pilar.v21i2.6487

Abstract

Gondang is one of the PSHT sub-branches located in Sragen Regency, Central Java, Indonesia. In managing member data from recruitment to promotion, conventional methods are still used using office applications and information dissemination is still using brochures and social media. This research aims to develop an information system that can help manage data and disseminate information at PSHT Gondang. The system developed can manage the registration of prospective member to become a member and the process of promotion. Delivery of information in the form of organizational structures, announcements, activity schedules, services for member and community, activity galleries containing photos and videos can also be accessed through the system.EUD was chosen as a method in system development because time required is quite short with a relatively small cost allocation. The system is created using Laravel framework and Firebase as a database with a responsive display so that it can be accessed using a smartphone. By using the EUD method, users can modify the appearance and existing information if there is a change in data from the organization which was not available in previous research.
Penguatan Kapasitas Komunitas Gamelan Desa Jarum, Bayat, Jawa Tengah melalui Program Pelatihan Literasi Digital, Bahasa Inggris, dan Manajemen Wisata Sabdo Aji, Ananto; Devi Hari Putri, Emmita; Ary Setyadi, Heribertus; Rejeki, Sri
Jurnal Abdi Masyarakat Indonesia Vol 6 No 2 (2026): JAMSI - Maret 2026
Publisher : CV Firmos

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.54082/jamsi.2633

Abstract

Komunitas Gamelan Desa Jarum, Bayat, Jawa Tengah memiliki potensi besar sebagai pelaku wisata budaya, namun menghadapi keterbatasan kapasitas dalam literasi digital, kemampuan Bahasa Inggris, dan manajemen pariwisata. Kondisi ini berdampak pada rendahnya promosi digital, keterbatasan komunikasi dengan wisatawan mancanegara, serta belum optimalnya pengelolaan pertunjukan gamelan sebagai produk wisata. Kegiatan pengabdian kepada masyarakat ini bertujuan untuk menguatkan kapasitas komunitas gamelan melalui program pelatihan terintegrasi yang mencakup literasi digital, English for Gamelan Players, dan manajemen pariwisata berbasis kearifan lokal. Pelaksanaan kegiatan dilakukan melalui pelatihan berjenjang dan praktik kontekstual yang disesuaikan dengan karakteristik peserta. Hasil kegiatan menunjukkan peningkatan kemampuan peserta dalam memanfaatkan media digital untuk promosi, peningkatan kepercayaan diri dalam berkomunikasi sederhana menggunakan Bahasa Inggris, serta pemahaman yang lebih baik terhadap pengelolaan kegiatan wisata budaya. Selain itu, evaluasi pelaksanaan kegiatan menunjukkan tingkat kepuasan peserta yang tinggi dan peningkatan kualitas pelaksanaan pada sesi lanjutan. Kegiatan ini berdampak langsung pada meningkatnya kesiapan Komunitas Gamelan Desa Jarum dalam mendukung pengembangan desa wisata berbasis budaya lokal serta memperkuat peran komunitas seni sebagai aktor utama dalam ekosistem pariwisata yang berkelanjutan.
Penggunaan Metode Analytic Hierarchy Process Untuk Pembobotan Perilaku Kerja Dalam Penilaian Prestasi Kerja Dosen Kristianto, Agus; Setyadi, Heribertus Ary
Paradigma - Jurnal Komputer dan Informatika Vol. 24 No. 1 (2022): Periode Maret 2022
Publisher : LPPM Universitas Bina Sarana Informatika

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.31294/paradigma.v24i1.989

Abstract

Assessment of work performance consists of two elements, they are the employee's work goals and work behavior. The assessment of work behavior which consists of service orientation, integrity, commitment, discipline, cooperation and leadership that has been running so far is not considered objective between lecturers who have structural positions and ordinary lecturers. The purpose of this research is to produce a database information system that uses the AHP method to assist leaders in assessing work behavior. Work behavior between lecturers and structural officials will have different weights for some of the behavioral points assessed. AHP method is used to determine the weight of work behavior whose value is generated from the value of the interest ratio that has been entered. Result of this research is the value of work behavior according to the weight of each position and with predetermined criteria.. The system also generates the value of the Tri Dharma Tinggi activities for lecturers which include teaching and education, research, community service and supporting elements. System is made using visual basic programming and has been automated for AHP calculations and the value of lecturer activities. The weights generated by the AHP method are values that are consistent and feasible to use because they have been tested for consistency.
LEVERAGING CONTINUAL FINE-TUNING FOR EMOTION CLASSIFICATION IN PRODUCT REVIEWS ON MSME SUSTAINABILITY SUPPORT Galih Setiawan Nurohim; Heribertus Ary Setyadi; Pudji Widodo; Yusuf Sutanto
JITK (Jurnal Ilmu Pengetahuan dan Teknologi Komputer) Vol. 11 No. 4 (2026): JITK Issue May 2026
Publisher : LPPM Nusa Mandiri

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33480/jitk.v11i4.7729

Abstract

Automatic analysis of consumer product reviews is essential for understanding granular customer perceptions beyond basic sentiment. While transformer-based models are prevalent in Indonesian sentiment analysis, their adaptation for multi-emotion classification shifting from broad polarities to specific affective states remains underexplored. This study addresses this gap by proposing a Continual Fine-Tuning (CFT) approach to adapt a pre-trained IndoBERTweet model from three sentiment categories into five distinct emotion classes: Happiness, Sadness, Fear, Love, and Anger. The novelty lies in the strategic repurposing of sentiment-oriented weights to capture nuanced emotional representations in Indonesian e-commerce discourse. Experimental results on the PRDECT-ID dataset demonstrate that the proposed CFT model achieves an accuracy of 0.8157 and a weighted F1-score of 0.8118, significantly outperforming traditional neural networks and multilingual baselines. The CFT model demonstrates a 2.13% improvement in accuracy compared to the base IndoBERTweet without continual tuning and a substantial 59.54% lead over the multilingual BERT (mBERT) baseline. Despite limitations concerning the dataset scale (5,400 samples) and inherent subjectivity in emotion labeling, this research provides a robust conceptual framework for model adaptation in the Indonesian NLP ecosystem. These findings suggest that CFT is an efficient strategy for enhancing the emotional intelligence of transformer models, especially in domain-specific tasks where high-quality labeled data is constrained.