Smart Techno (Smart Technology, Informatics and Technopreneurship)

Sentiment Analysis of Gojek Application User Reviews Using the Long Short-Term Memory (LSTM) Algorithm

Ahmad Firdaussani — 2025-12-05

This study was conducted to perform sentiment analysis by identifying patterns or trends in user reviews of the Gojek application using the Long Short-Term Memory (LSTM) algorithm, which was implemented in the form of a simple web-based application or dashboard. In today’s digital era, technological advancements have significantly influenced various aspects of life, particularly the mobile-based transportation service industry. One of the most widely used online transportation services in Indonesia is Gojek. It is essential for Gojek to listen to customer reviews; therefore, sentiment analysis is required to identify patterns or trends within user feedback so the application can better respond to user needs. This research utilizes the Long Short-Term Memory (LSTM) algorithm, a variant of the Recurrent Neural Network (RNN) that incorporates a cell state and gating mechanisms (input, forget, and output gates) to regulate the flow of information. This structure enables LSTM to retain relevant information while discarding irrelevant data, allowing it to capture both short-term and long-term patterns in text reviews. The model was used to analyze sentiment within a dataset collected from 2021 to 2024. The experimental results show that LSTM achieved an optimal accuracy of 78% using a 70:30 dataset split, providing balanced performance across both majority and minority classes, with a significant improvement in the f1-score for each class (0: 0.73; 1: 0.75; 2: 0.85) after applying the SMOTE technique to address class imbalance. Without SMOTE, the highest accuracy reached 83% with the same split (70:30); however, the neutral class could not be detected (f1-score = 0). With SMOTE, although accuracy slightly decreased, the overall performance became more balanced as the neutral class could be properly recognized.

Design and Implementation of a Web-Based Extracurricular Management System at SMP Negeri 04 Kotabumi

Reni Agustina — 2025-12-10

This study aims to design and implement a web-based extracurricular management information system at SMP Negeri 04 Kotabumi as a solution to issues in managing activities that are still conducted manually. The previous manual process often resulted in registration delays, data recording errors, and difficulties in activity reporting. The system was developed using the Waterfall method, which includes five main stages: requirements analysis, design, coding, testing, and maintenance.

During the analysis stage, observations and interviews were conducted on eight active extracurricular activities, involving 15 supervisors and 120 students. The system was designed using PHP, MySQL, and Bootstrap to create a dynamic and easily accessible interface. Testing was conducted by 10 admins/supervisors and 20 students using Black Box Testing, with five trials for each main feature. The results showed a 100% success rate with no functional errors. Moreover, the implementation of this web-based system increased the average time efficiency by 74.5% compared to the previous manual process and enhanced transparency and effectiveness in managing extracurricular activities at the school.

Clustering of Planted Area, Harvested Area, and Rice Production in Each Village of Jember Regency Using K-Means Clustering and the Davies Bouldin Index

Hestina Restu Astika — 2025-12-10

Rice (Oryza sativa L.) is a cultivated crop that serves as the primary staple food for the majority of Indonesia’s population. East Java Province is one of the regions with the highest rice production in the country. Therefore, increasing rice production is essential to meet national food demands. This study aims to classify villages in Jember Regency based on the variables of planted area, harvested area, and rice production, using data obtained from the official publications of the Jember Regency Central Bureau of Statistics for 2022 and 2023, covering a total of 248 villages. The data were processed using the K-Means Clustering algorithm, followed by determining the optimal number of clusters using the Davies Bouldin Index. The clustering results were visualized in an interactive web-based map through a Geographic Information System. Based on testing cluster counts from 2 to 10, the optimal number of clusters was found to be three, with a Davies Bouldin Index value of 0.605. This study is expected to provide benefits for the Jember Regency Central Bureau of Statistics, the community, and farmers in storing, managing, and disseminating information regarding rice crops in Jember Regency.

User Satisfaction Analysis of the E-Monev Application Using the Integration of the EUCS and TAM Methods: A Case Study of the Jombang Regency Government

Mochamad Imamudin — 2025-12-16

The E-Monev application is used as a system for monitoring and evaluating program performance. However, based on preliminary field observations, it is found that some E-Monev users still experience various difficulties. Several commonly reported problems include an application interface that is considered difficult to understand, data access processes that are often slow and require a long time to load information, and confusion in operating the menus available in the E-Monev application.

This study seeks to investigate user satisfaction and acceptance of the E-Monev application by combining two frameworks: the End User Computing Satisfaction (EUCS) and the Technology Acceptance Model (TAM). A quantitative approach was employed, involving the distribution of questionnaires to 134 E-Monev users within the Jombang Regency Government. Data were analyzed using Partial Least Squares Structural Equation Modeling (PLS-SEM) through SmartPLS 4. The analysis encompassed both the outer model assessing convergent validity, discriminant validity, and reliability and the inner model, which included evaluation of R-square, effect size, predictive relevance, and hypothesis testing.The results show that several variables, namely Content (t-value = 2.759; p-value = 0.006), Perceived Ease of Use (t-value = 3.637; p-value = 0.000), and Attitude (t-value = 14.965; p-value = 0.000), have a significant effect on the actual use of the application. Meanwhile, the variables of Accuracy, Perceived Usefulness, Ease of Use, Format, and Timeliness do not show a significant influence on user attitude. The factor with the strongest effect is Attitude toward Actual Use; therefore, enhancing users’ positive attitudes becomes the main key to optimizing E-Monev utilization. These findings provide insights for system administrators to improve content quality, ease of interaction, and user experience so that the implementation of E-Monev can be carried out more effectively.

Predicting Employee Attrition Using the Random Forest Algorithm Based on IBM HR Analytics Data

Putu Satya Saputra — 2026-01-14

The phenomenon of employee attrition has become a serious challenge for organizations, as it directly affects productivity, recruitment costs, and long-term performance stability. Understanding the factors that lead to employee turnover can no longer rely solely on manual observation; therefore, data-driven approaches are required to identify hidden patterns within workforce data. This study aims to predict employee attrition using the Random Forest algorithm applied to the IBM HR Analytics Employee Attrition & Performance dataset, which consists of 1,470 records and 35 attributes. The research stages include data preprocessing, handling class imbalance using the Synthetic Minority Over-sampling Technique (SMOTE), model training, and performance evaluation using accuracy, precision, recall, F1-score, ROC-AUC, and a confusion matrix. The results indicate that the baseline model without SMOTE exhibits low recall for the attrition class, whereas the application of SMOTE significantly improves model performance, particularly for the minority class, achieving a final accuracy of 83.96%. The most influential features identified are Stock Option Level, MonthlyIncome, and JobSatisfaction. These findings provide a comprehensive understanding of the factors influencing employee attrition and can serve as a foundation for organizations in designing more adaptive and data-driven employee retention strategies.

Sentiment Analysis of the Hindu Community Toward Religious Issues on Social Media: A Case Study of Kapuas Regency Using a Text Mining Approach

I Wayan Sindia Griya Danika — 2026-01-19

This study aims to analyze the sentiment of the Hindu community toward religious issues on social media, with a specific focus on Kapuas Regency, Central Kalimantan. The development of social media as a digital public sphere has positioned platforms such as Instagram and Twitter as primary spaces for the spontaneous and open expression of opinions, perceptions, and religious attitudes. This study is important because religious issues in digital spaces often influence social harmony and shape religious discourse within society. The research employs a quantitative approach using text mining and sentiment analysis based on Natural Language Processing (NLP). Primary data were collected through web scraping techniques, utilizing Apify for Instagram and asynchronous Python scripts for Twitter, with relevant keywords, hashtags, and geographic indicators. The analysis process includes text preprocessing (cleaning, tokenization, stopword removal, and stemming), followed by sentiment classification using a lexicon-based approach with the InSet dictionary into three categories: positive, negative, and neutral. The analysis results were evaluated using a confusion matrix, along with precision, recall, and F1-score metrics to assess model reliability. The findings indicate that positive sentiment predominates on both Instagram and Twitter, followed by neutral sentiment, while negative sentiment appears in only a small proportion. Positive sentiment is generally associated with expressions of prayer, gratitude, tolerance, and calls for togetherness, whereas negative sentiment tends to emerge in discussions related to ritual differences or responses to socio-religious controversies. The sentiment analysis model achieved an accuracy of 100% on Instagram data (self-evaluation) and 74.4% on Twitter data (manual evaluation), with relatively high precision and recall values, indicating that the results are statistically reliable.

Implementation of DeepFace for Gender Prediction Based on Facial Images

Aditya Wijaya — 2026-01-19

This study evaluates the performance of a pretrained DeepFace model for gender classification based on facial images using the UTKFace dataset. A total of 100 facial images were employed as test data, consisting of 50 male and 50 female samples selected through controlled random sampling to maintain class balance. Image preprocessing was conducted automatically using the DeepFace.analyze() function, which includes face detection, alignment, size normalization, and facial cropping. The study did not involve model retraining and relied solely on the inference capability of the pretrained DeepFace model. The experimental results show that the model correctly classified 45 male and 44 female images, achieving accuracies of 90% and 88% for the male and female classes, respectively, with an overall accuracy of 89%. Confusion matrix analysis indicates that misclassifications were primarily influenced by image quality factors such as lighting variations, camera angles, and facial expressions. Overall, the findings demonstrate that DeepFace is effective for gender classification without retraining; however, further improvements in preprocessing techniques and dataset diversity may enhance classification performance in future research.

How Interface Design Nudges Instagram Users Toward Posting Less Permanent Content

Putu Dhanu Driya — 2026-01-19

This study examines the effect of Instagram interface design nudges on Gen Z users’ preference for ephemeral content over permanent feed posts, and the mediating roles of cognitive biases and self-presentation concerns. A survey of 347 Gen Z users was analyzed using parallel mediation (PROCESS Model 4). Results indicated that interface nudges significantly predicted cognitive biases (b = 0.903, p = 0.047) and self-presentation concerns (b = 0.807, p = 0.039), but neither mediator significantly influenced ephemeral posting (indirect effect M1 = 0.0021, 95% CI [-0.0244, 0.0192]; M2 = 0.0003, 95% CI [-0.0165, 0.0236]). The direct effect of nudges on ephemeral posting was significant (b = 0.060, p = 0.031), indicating that UI design directly encourages temporary content sharing. These findings highlight the dominant role of interface design in guiding user behavior, suggesting that nudges influence ephemeral posting primarily through direct behavioral effects rather than mediated psychological mechanisms.

Implementation of Latent Dirichlet Allocation in a Cookie-Based Final Project Topic Recommendation System

Fiddar Tahwifa Putri — 2026-01-20

The selection of a final project topic is a crucial stage in the academic journey of students, as it determines the direction of research while serving as a means to apply the knowledge acquired during their studies. However, in practice, many students experience difficulties in choosing a topic that aligns with their interests and areas of expertise. This challenge is largely attributed to the absence of systems capable of providing personalized recommendations. To address this issue, this study develops a final project topic recommendation system by integrating the Latent Dirichlet Allocation (LDA) algorithm with a cookie-based approach to accommodate user preferences. The dataset used consists of 200 final project documents from the Informatics Engineering program, with titles and abstracts serving as the primary features for topic modeling during model training and perplexity evaluation. In addition, users’ search histories and relevance feedback stored in cookie sessions are utilized as personalization features to generate more tailored recommendations. FastText is employed to produce more contextual word vector representations, while cosine similarity is applied to measure the closeness between search keywords and final project topic documents. Evaluation results based on perplexity indicate that the model with 22 topics yields the most statistically optimal performance. Furthermore, testing using Click-Through Rate (CTR) demonstrates that the combination of topic modeling and user preference personalization produces the highest relevance, achieving a CTR of 15.6%, which is significantly higher than the baseline CTR of 2.2%. These findings confirm that the proposed system is capable of delivering more relevant, adaptive, and user-oriented final project topic recommendations.

Optimization of the Payment Process at Toko Pertanian Kurnia Manokwari Using Business Process Reengineering and Throughput Efficiency

Faizal Qadri Trianto — 2026-01-20

Rapid advances in digital technology within the agribusiness sector require Micro, Small, and Medium Enterprises (MSMEs) to adapt their operational strategies to remain competitive. This study presents a single-case study conducted at Toko Pertanian Kurnia Manokwari, an agribusiness MSME in West Papua, which experiences inefficiencies in its payment and order fulfillment processes due to reliance on manual bank transfers and centralized owner-based verification. The study aims to optimize the payment process through the application of Business Process Reengineering (BPR) by modeling the existing (As-Is) and redesigned (To-Be) processes using Business Process Model and Notation (BPMN) and evaluating process performance with the ASME Standard Process Chart through throughput efficiency measurement. The analysis identifies centralized verification as a single point of failure that prolongs transaction cycle times. The proposed solution integrates an API-based automated payment gateway to replace manual verification. The results indicate that the As-Is process achieves a throughput efficiency of 35.48% with a total cycle time of 186 minutes, whereas the evaluation of the redesigned To-Be process model indicates a potential increase in throughput efficiency to 100% and a reduction in cycle time to 23 minutes. These findings demonstrate that BPR supported by digital payment system integration, based on To-Be process modeling, can significantly improve transaction efficiency and operational scalability in agribusiness MSMEs.

Design And Development of An Internet of Things-based Smartbell Using ESP32-Cam And Telegram

Puji Utami Rakhmawati — 2026-02-01

When guests or couriers arrive, homeowners must go to the front door to identify them due to the limitations of conventional doorbells, which only produce sound. If the homeowner is not at home, they may miss the arrival of guests, and couriers might leave packages elsewhere, posing a risk of loss since there is no real-time notification system. To address this issue, a Smartbell based on the Internet of Things (IoT) was designed using the ESP32-CAM module as an image capture device, integrated with Telegram to provide homeowners with real-time visual information. This study applied the prototype method, which consists of stages such as requirements identification, system design, coding, functionality and time testing, as well as system evaluation before implementation. The test results show that the Smartbell successfully performed as expected. In the simulation, when the bell button was pressed, the buzzer sounded, and the ESP32-CAM camera automatically captured an image and sent it to Telegram in real-time. Since the Smartbell was successfully connected to the Telegram bot, it can be operated remotely. Testing with a Wi-Fi network resulted in an average response time of 0.74 seconds, while using a cellular data network achieved 0.54 seconds. With a response time of less than one second from the integration of ESP32-CAM and Telegram, this system supports the homeowner’s needs as a remote and real-time guest monitoring solution.

Analysis of the Effect of Spectral Feature Dimensionality on Audio Classification Performance

Tria Hikmah Fratiwi — 2026-02-01

This study examines the impact of spectral feature quantity on the classification performance of dangdut music sub-genres, namely classical dangdut, dangdut rock, and dangdut koplo. Previous studies reported relatively low classification accuracy, which is presumed to be influenced by spectral features with small numerical values and dense feature distributions. To address this issue, two feature configurations were evaluated six and five spectral features using the K-Nearest Neighbor (KNN) algorithm and a Genetic Algorithm-optimized KNN (GA- KNN). Model performance was assessed using accuracy, precision, recall, and F1-score, supported by confusion matrix analysis. The results show that the six-feature configuration consistently outperforms the five- feature configuration for both methods. GA-KNN achieved the best performance with six spectral features, yielding an accuracy of 71.53%, precision of 0.7147, recall of 0.7153, and an F1-score of 0.7140, outperforming conventional KNN, which achieved an accuracy of 62.50% and an F1-score of 0.6135. When reduced to five spectral features, performance declined for both methods; GA-KNN reached an accuracy of 66.67% with an F1-score of 0.6611, while conventional KNN dropped to 52.08% accuracy with an F1- score of 0.5121, accompanied by increased misclassification between sub-genres with similar spectral and rhythmic characteristics. These findings indicate that spectral features with small numerical values still contribute meaningful discriminative information and should be carefully evaluated before applying feature reduction in music genre classification tasks.

Sentiment Analysis of YouTube Comments for the Jumbo Movie Trailer Using IndoBERT

Fardan Zamakhsyari — 2026-01-31

The film industry in Indonesia has experienced significant growth, from cinematography to animation. Along with this growth, public opinion has also varied, from assessments of the storyline to the production process. To assess public sentiment on social media, a system is needed that can accommodate this process. This study aims to analyse public sentiment towards the trailer for the animated film ‘Jumbo,’ which was released on the YouTube platform. Using an NLP approach, two fine-tuned IndoBERT models were compared: ‘Aardiiiiy/indobertweet-base-Indonesian-sentiment-analysis’ and ‘rikidharmawan/finetuning-sentiment-model-indobertweet-v2’. The data to be processed was obtained from 1,468 YouTube comments through a crawling process using the YouTube API. The data was then analysed using both models to classify the comments into positive, neutral, and negative sentiments. Evaluation was conducted using a confusion matrix with accuracy, precision, recall, and F1-score metrics. The evaluation results show that ‘Aardiiiiy/indobertweet-base-Indonesian-sentiment-analysis’ is superior, with an accuracy of 57.2% and a higher average F1-score compared to ‘rikidharmawan/finetuning-sentiment-model-indobertweet-v2,’ which has an accuracy of 51.3%. This research contributes to the selection of sentiment analysis models for Indonesian-language data, particularly in the domains of social media and the film industry.

An Explainable Deep Learning for Malaria Blood Cell Classification Using DenseNet121 and Grad-CAM

Octavian — 2026-01-31

Malaria diagnosis based on microscopic examination of blood smears is time-consuming and highly dependent on skilled laboratory personnel, which limits its scalability in resource-constrained environments. This study investigated whether an explainable deep learning approach could provide reliable and interpretable malaria blood cell classification using a convolutional neural network based on the DenseNet121 architecture combined with Gradient-weighted Class Activation Mapping to visualize the image regions influencing model predictions. Five-fold cross-validation was applied to ensure a stable and unbiased performance evaluation. The model achieved a mean classification accuracy of 0.8285 with low variation across folds, and the precision, recall, and F1-score values were balanced between the parasitized and uninfected classes. Visual explanations consistently highlighted intracellular regions associated with parasite presence in infected cells and more uniform cytoplasmic regions in uninfected samples, indicating that the network learned the biologically meaningful features of the cells. The results demonstrated that DenseNet121 provided a stable and interpretable solution for malaria blood cell classification when supported by a visual explanation, thereby enabling transparent automated screening. The proposed framework is suitable for integration into smart healthcare and medical informatics systems, where both predictive reliability and interpretability are required.