Claim Missing Document
Check
Articles

Systematic Review of Hyperparameter Adjustment and Evaluation Metrics in Bert-Based Sentiment Analysis Bahari, Aris Rifki Setiya; Utomo, Fandy Setyo; Berlilana, Berlilana
Journal La Multiapp Vol. 7 No. 2 (2026): Journal La Multiapp
Publisher : Newinera Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.37899/journallamultiapp.v7i2.3046

Abstract

The development of sentiment analysis towards Aspect-Based Sentiment Analysis (ABSA) has made significant progress thanks to deep learning technology, especially through the Bidirectional Encoder Representations from Transformers (BERT) architecture. Despite its increasing popularity, a comprehensive synthesis of global research patterns and optimal model configurations is still urgently needed. This study presents a Systematic Literature Review (SLR) combined with bibliometric analysis to examine BERT-based ABSA research indexed in Scopus. Using the PRISMA and VOSviewer frameworks for visualization, a total of 62 eligible articles up to mid-2025 were analyzed. The results of the study show a strong upward trend of publications with a peak in 2024, where China, India, and Indonesia emerged as the major contributors in this domain. Further, the review identified a critical technical standard for effective model training: the Adam optimizer was the most dominant choice, typically paired with a learning rate between 1e-5 to 2e-5 and a batch size of 16. Regarding performance evaluation, Accuracy and F1-Score are set as de facto standard metrics. These findings provide strategic guidance for researchers to optimize BERT implementation and identify future directions in more in-depth sentiment analysis tasks.
Comparative Performance Analysis of Random Forest and Logistic Regression for Sentiment Classification of the Makan Bergizi Gratis Program on Platform X Slamet Endro Prianto; Berlilana Berlilana; Rujianto Eko Saputro
Journal of Information System and Informatics Vol 8 No 1 (2026): February
Publisher : Asosiasi Doktor Sistem Informasi Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.63158/journalisi.v8i1.1371

Abstract

The rapid growth of e-commerce has made personalized product recommendations a crucial aspect of enhancing customer satisfaction and boosting sales. However, many small-to-medium-sized retail businesses, like Adiva Fashion Store, still rely on manual product selection through customer searches or seller recommendations, which often leads to challenges in meeting customer preferences. This study presents a case study of Adiva Fashion Store, where the Collaborative Filtering method was implemented to develop a personalized clothing product recommendation system. The item-based Collaborative Filtering approach was employed to calculate the similarity between products based on customer ratings and transaction history. These similarity values were then used to predict customer preferences for products that had not yet been purchased. The system was developed using the Waterfall methodology, which involved needs analysis, system design, implementation, testing, and maintenance. The results show that the recommendation system significantly improved the relevance of product suggestions, helping customers make better purchasing decisions and increasing sales effectiveness. This case study illustrates how data-driven recommendation systems can be effectively integrated into small-to-medium-sized retail environments, providing valuable insights for other businesses aiming to adopt similar strategies.
A Decision Support System for Assessing High School Students' Soft Skills Using the Analytical Hierarchy Process Yuwono Wisudo Pramono; Berlilana Berlilana; Azhari Shouni Barkah
Journal of Information System and Informatics Vol 8 No 1 (2026): February
Publisher : Asosiasi Doktor Sistem Informasi Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.63158/journalisi.v8i1.1420

Abstract

Assessing students' soft skills in educational settings is often challenging due to the subjectivity and inconsistency inherent in evaluating qualitative traits. This study employs the Analytical Hierarchy Process (AHP) as a decision support tool to provide a more systematic, consistent, and objective method for evaluating students' soft skills. The assessment model is based on four key criteria—critical thinking, communication, collaboration, and creativity—each further broken down into measurable subcriteria. The study was conducted at MA Mu’allimin Sruweng Kebumen, where evaluations were carried out by a guidance and counseling teacher acting as an expert evaluator, using a numerical scale ranging from 1 to 100. Pairwise comparison matrices were developed using Saaty’s fundamental scale to determine the weights for both criteria and subcriteria, followed by consistency testing using the Consistency Ratio (CR). The findings reveal that critical thinking and collaboration were assigned the highest priority weights, with all comparison matrices meeting the acceptable consistency threshold. The resulting global preference values offer a more objective, proportional representation of students’ soft skills achievements. This AHP-based model enables fairer, more consistent evaluations and provides quantitative outputs that can be utilized for student ranking and structured feedback in educational decision-making.
Modeling EMIS Adoption with PLS-SEM: Integrating the Government Adoption Model and DeLone–McLean IS Success Model Mardiyanto Mardiyanto; Berlilana Berlilana; Purwadi Purwadi
Journal of Information System and Informatics Vol 8 No 1 (2026): February
Publisher : Asosiasi Doktor Sistem Informasi Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.63158/journalisi.v8i1.1445

Abstract

This study explores the key factors influencing the adoption of the Education Management Information System (EMIS) within Indonesia's Ministry of Religious Affairs (Kemenag), which is vital for managing data and distributing Teacher Professional Allowances (TPG). Data inconsistencies have been a significant challenge, leading to delays in TPG disbursement. To understand the determinants of EMIS adoption, this study integrates the Government Adoption Model (GAM) and DeLone & McLean’s (D&M) Information Systems Success Model. A quantitative approach was used, collecting data from 328 valid responses from MTsN teachers in Kebumen Regency, analyzed using Partial Least Squares Structural Equation Modeling (PLS-SEM). The results show that Perceived Uncertainty (PU), Perceived Security (PSC), and Perceived Privacy (PP) positively contribute to Perceived Trust (PT). Additionally, Information Quality (IQ) emerged as the strongest predictor of EMIS adoption, followed by System Quality (SYQ), Service Responsiveness (PSR), and Trust. The study emphasizes that improving data accuracy (IQ), ensuring system reliability (SYQ), and strengthening security measures (PSC) are critical for accelerating EMIS adoption. The findings offer practical implications for Kemenag to optimize the implementation of EMIS, ultimately improving the efficiency and timeliness of TPG disbursements for educators.
ANALISIS POLA PENYEBARAN PENYAKIT MENGGUNAKAN PENDEKATAN CLUSTERING HIERARKIS DAN K-MEANS Setiyawan, Dilliana Tugas; Berlilana, Berlilana; Barkah, Azhari Shouni
JIPI (Jurnal Ilmiah Penelitian dan Pembelajaran Informatika) Vol 10, No 4 (2025)
Publisher : STKIP PGRI Tulungagung

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.29100/jipi.v10i4.7328

Abstract

Penyebaran penyakit, baik yang bersifat menular maupun tidak menular, merupakan isu penting yang harus diidentifikasi secara tepat untuk mendukung upaya pencegahan dan pengendalian kesehatan masyarakat. Identifikasi pola sebaran penyakit menjadi krusial karena setiap penyakit memiliki karakteristik penyebaran yang berbeda, baik berdasarkan faktor lingkungan, demografi, maupun perilaku masyarakat. Penerapan K-Means Cluster Analysis merupakan metode yang digunakan untuk mengelompokkan data menjadi beberapa kelompok (cluster) berdasarkan kesamaan karakteristik. Selain itu, pendekatan Hierarchical Clustering diterapkan untuk memvisualisasikan hubungan antar data secara hierarkis, memungkinkan analisis yang lebih mendalam. Tujuan penelitian ini adalah untuk menganalisis pola penyebaran penyakit menggunakan pendekatan Clustering Hierarkis dan K-Means. Data dari 35 Puskesmas dianalisis berdasarkan jumlah pasien dan prevalensi penyakit, termasuk Tuberkulosis, Diabetes, Hipertensi, dan penyakit menular lainnya. Hasil penelitian menunjukkan bahwa kedua metode memberikan wawasan yang saling melengkapi. K-Means efektif dalam membagi data menjadi cluster yang merata dan efisien, sementara Hierarchical Clustering memungkinkan identifikasi hubungan hierarkis dan distribusi granular antar Puskesmas
Utilitarian vs Human-Centered AI Acceptance: Explaining Students’ Adoption of ChatGPT in Higher Education Parameswara, Dwi Angesti Dinda; Berlilana, Berlilana; Saputro, Rujianto Eko
Jurnal Pendidikan Informatika (EDUMATIC) Vol 10 No 1 (2026): Edumatic: Jurnal Pendidikan Informatika
Publisher : Universitas Hamzanwadi

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.29408/edumatic.v10i1.34218

Abstract

The growing use of generative artificial intelligence (AI) in higher education raises questions about how students assess and adopt these systems, particularly whether traditional utilitarian models are sufficient to explain their use. This study compares the Technology Acceptance Model (TAM) and the Human-Centered AI Acceptance Model (HCAIAM) in explaining students’ behavioral intention to use ChatGPT, while examining how functional and human-centered factors operate within the same framework. A cross-sectional design was used, involving 100 undergraduate students in Indonesia selected through convenience sampling, and the data were analyzed using Partial Least Squares Structural Equation Modeling (PLS-SEM). The results show that TAM provides stronger explanatory power and better model fit (R² = 0.765; SRMR = 0.073) than HCAIAM (R² = 0.709; SRMR = 0.136). Perceived usefulness and perceived ease of use emerge as the main drivers of intention, indicating that students tend to use ChatGPT primarily as a tool to support academic tasks. In contrast, human-centered factors such as transparency and ethical alignment influence intention indirectly through trust and attitude. The autonomy construct shows weak reliability and overlaps with other variables, suggesting limitations in its measurement. These findings indicate that utilitarian factors remain central in this context, while human-centered aspects play a more conditional role, and point to a layered pattern of AI acceptance in which different types of factors operate at different levels.
A Comprehensive Evaluation of CatBoost and LightGBM Algorithms for Honorarium Prediction on Categorical Datasets with Class Imbalance Slamet Widodo; Fandy Setyo Utomo; Berlilana
JUITA: Jurnal Informatika JUITA Vol. 13 Issue 3, November 2025
Publisher : Department of Informatics Engineering, Universitas Muhammadiyah Purwokerto

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30595/juita.v13i3.27363

Abstract

Determining income, including honoraria in the academic environment, is often done manually and subjectively, necessitating a predictive model to objectively determine the honorarium amount. However, the development of the prediction model faces challenges due to the dataset's characteristics, which include categorical data and an imbalanced class distribution. This research aims to evaluate the predictive performance and computational resource efficiency of the CatBoost and LightGBM algorithms in predicting honorariums. The dataset used includes 58,332 actual honorarium data of employees from higher education institution "A" in Purwokerto for the period from January 2024 to February 2025. The methods used include data preprocessing, dataset splitting using Stratified Splitting, modeling with CatBoost, LightGBM, Random Forest, Neural Network, and Linear Regression, as well as evaluation using MSE, RMSE, MAE, R² metrics, and computational resources (execution time, memory, CPU time). LightGBM achieved an RMSE of 665.960 and an R² of 0.54, while recording the lowest memory usage at only 2.67 MB. CatBoost produced an RMSE of 667.395 and an R² of 0.53, excelling in processing categorical features without one-hot encoding. Meanwhile, Linear Regression showed the lowest accuracy and high memory usage. These results confirm that LightGBM is the most optimal choice for fast, efficient, and accurate honorarium predictions. However, this research is limited to testing in a laboratory environment. Further research is recommended to implement direct integration with an active database and the integration of information retrieval methods to enhance the effectiveness and security of real-time honorarium predictions, as well as to integrate interpretability methods such as SHAP to improve decision-making transparency.
Literature Review: Comparison of Machine Learning Algorithms for Sentiment Analysis of Free Nutritious Meals Mukhlisin Mukhlisin; Berlilana Berlilana; Rujiyanto Eko Saputro
Sistemasi: Jurnal Sistem Informasi Vol 15, No 3 (2026): Sistemasi: Jurnal Sistem Informasi
Publisher : Program Studi Sistem Informasi Fakultas Teknik dan Ilmu Komputer

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32520/stmsi.v15i3.6201

Abstract

The Free Nutritious Meal (FNM) program has triggered massive public responses on social media, driving numerous machine learning–based sentiment analysis studies. However, there has been no comprehensive review comparing the effectiveness of these methods. This study adopts a Systematic Literature Review (SLR) approach on 18 studies (2024–2026) to evaluate the performance of computational algorithms and map trends in public sentiment. The main contribution of this research is to provide an empirical guide for selecting Indonesian-language text classification models, while also offering insights into shifts in public perception. Key findings indicate that Support Vector Machine (SVM) is the most frequently used method, whereas the highest accuracy (97%) was achieved by a combination of Logistic Regression, SVM, and Random Forest on large datasets. Temporally, sentiment trends shifted from budget skepticism (2024) to positive acceptance during program implementation (2025–2026). The study’s implications support policymakers in evaluating program effectiveness in real time. The scope and limitations of this research focus on literature within a specific timeframe, with performance evaluation emphasizing quantitative accuracy metrics.
Prediksi Harga Cryptocurrency Multi-Aset Menggunakan Machine Learning dan Deep Learning Yusuf Nur Alam; Berlilana
Jurnal Algoritma Vol 23 No 1 (2026): Jurnal Algoritma
Publisher : Institut Teknologi Garut

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33364/algoritma/v.23-1.3473

Abstract

Cryptocurrency price volatility requires predictive models capable of accurately capturing non-linear patterns. This study predicts the price of Bitcoin (BTCUSDT) as the main asset, as well as Ethereum (ETHUSDT) and Ripple (XRPUSDT) as comparison assets, using Decision Tree, Random Forest, XGBoost, and LSTM models. The novelty of this study lies in the analysis of temporal data leakage and the evaluation of model extrapolation capability within a uniform experimental framework. Daily historical data were processed through cleaning, correlation analysis, variable selection, and sequential 70:30 data splitting. The prediction target was defined as the next-day closing price to avoid data leakage, and the models were evaluated using time-series cross-validation with RMSE, MAPE, and R² metrics. The results show that the best-performing model differs for each asset: LSTM outperformed other models for BTC and XRP, while Random Forest performed best for ETH, with R² values ranging from 0.60 to 0.98. Tree-based models tended to produce flat predictions when test prices exceeded the training data range. These findings emphasize the importance of defining prediction targets, applying temporal validation, and conducting cross-asset evaluation in selecting appropriate models for cryptocurrency price prediction.
Development of interactive learning media through 2D animated video in Indonesian language learning Zhafran Afif Nurdiyansah; Berlilana Berlilana
Journal of Educational and Learning Studies Vol 8, No 2 (2025): Journal of Educational and Learning Studies
Publisher : Global Econedu

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.32698/02282

Abstract

The development and advancement of technology play a crucial role in various fields, including education. Learning media holds significant importance in the teaching and learning process. Furthermore, the use of learning media can also have positive psychological effects on students. As we know, first-grade students are still at a stage where it can be challenging for teachers to provide explanations that they can fully comprehend. One example is at MI Muhammadiyah Kalilandak, where many students still struggle with reading, especially in the first grade. Since fluent reading is essential, the use of learning media that can support these students is necessary, such as through 2D animated videos. With the inclusion of videos equipped with animation, text, and music, it is hoped that the students will be more motivated in their learning. Therefore, this research will focus on the creation of 2D animated videos to enhance the reading abilities of the students.