JURNAL INFOTEL

A Systematic Literature Review of BERT-based Models for Natural Language Processing Tasks

2024-12-16T12:22:11+00:00

Research area in natural language processing (NLP) domain has made major advances in recent years. The Bidirectional Encoder Representations from Transformers (BERT) and its derivative models have been at the vanguard, gaining notice for their exceptional performance across a variety of NLP applications. As a response to this context, hence, this study aims to conduct a systematic literature review on current research in BERT-based models in order to describe their characteristic variations on three frequently demanded natural language processing (NLP) tasks, i.e. text classification, question answering, and text summarization. This study employed a systematic literature review method as prescribed by Kitchenham. We collected 4,120 papers from publications indexed by Scopus and Google Scholar from which 42 complied to our defined review criteria and finally chosen for further analysis. Our review came up with three conclusions. First, in order to select appropriate models for particular NLP tasks, three primary concerns should be considered: i) the type of NLP problem to be resolved (i.e. NLP task to be served), ii) the specific domain to be handled (such as financial, medical, law/legal or others), and iii) the intended language to be applied (such as English or others). Second, learning rate, batch size, and the type of optimizer were the three most considered hyperparameters to be properly arranged in model training. Third, the most widely used metrics for text classification tasks were F1-score, accuracy, precision, and sensitivity (recall), while question answering, and text summarization tasks were mostly used the Exact Match and ROUGE respectively.

A Random Oversampling and BERT-based Model Approach for Handling Imbalanced Data in Essay Answer Correction

2024-12-20T03:00:56+00:00

The task of automated essay scoring has long been plagued by the challenge of imbalanced datasets, where the distribution of scores or labels is skewed towards certain categories. This imbalance can lead to poor performance of machine learning models, as they tend to be biased towards the majority class. One potential solution to this problem is the use of oversampling techniques, which aim to balance the dataset by increasing the representation of the minority class. In this paper, we propose a novel approach that combines random oversampling with a BERT-base uncased model for essay answer correction. This research explores various scenario of text pre-processing techniques to optimize model accuracy. Using a dataset of essay answers obtained from eighth-grade middle school students in Indonesian language, our approach demonstrates good performance in terms of precision, recall, F1-score and accuracy compared to traditional methods such as Backpropagation Neural Network, Naïve Bayes and Random Forest Classifier using FastText word embedding with Wikipedia 300 vector size pretrained model. The best performance was obtained using the BERT-base uncased model with 2e-5 learning rate and a simplified pre-processing approach. By retaining punctuation, numbers, and stop words, the model achieved a precision of 0.9463, recall of 0.9377, F1-score of 0.9346, and an accuracy of 94%.

Blockchain-based Supply Chain Solution Using IPFS and QR Technology for Traditional Weave in West Nusa Tenggara

2024-12-20T03:14:25+00:00

Traditional weaving is a significant cultural heritage in West Nusa Tenggara, Indonesia, renowned for its unique and intricate woven fabrics. However, the industry faces challenges such as product counterfeiting and the need for more transparency in the supply chain, hindering its growth and economic potential. This paper proposes a solution by leveraging blockchain technology to enhance traceability, security, and efficiency throughout the supply chain of traditional woven products. Integrating the InterPlanetary File System (IPFS) and Quick Response (QR) codes further fortifies data integrity and provides consumers with comprehensive product information. A prototype web application is developed, demonstrating the practical implementation of this framework. Rigorous testing of the prototype validates the correct functionality of the proposed solution. This innovative framework aims to safeguard the authenticity and sustainability of the traditional woven products from West Nusa Tenggara, fostering a more competitive and secure industry while preserving cultural heritage.

Solar Radiation Prediction using Long Short-Term Memory with Handling of Missing Values and Outliers

2024-12-20T04:03:20+00:00

The pyranometer sensor is an instrument for measuring Global Horizontal Irradiance (GHI) which is used as parameter for analyzing and predicting weather. GHI data which is processed into prediction model for photovoltaics is useful for determining the performance of solar power generation systems in distributed energy operations. However, GHI sensor data has weaknesses in missing values and outliers due to measurement errors. The research designed a GHI sensor data prediction model using data preprocessing by the imputation of missing values using linear, polynomial, and Piecewise Cubic Hermite Interpolating Polynomials (PCHIP) interpolation and eliminating outliers using Random Sample Consensus (RANSAC) on the dataset. Previous researches show that Long Short-Time Memory (LSTM) can improve the performance of predictions compared to machine learning. This research designs an LSTM prediction model with data preprocessing and without data preprocessing. The results of the imputation of missing values obtained the best performance in PCHIP with Mean Absolute Error (MAE) 39.708 W/m², Root Mean Absolute Error (RMSE) 76.224 W/m², Normalized Root Mean Absolute Error (NRMSE) 0.433, and Coefficient Determination (R²) 0.903 then imputation from outlier elimination obtained MAE 44.377 W/m², RMSE 86.738 W/m², NRMSE 0.500, and R² 0.886. RANSAC testing succeeded in eliminating 100% outliers. The results of LSTM with data preprocessing obtained better performance with the best evaluations on MAE, RMSE, NRMSE, and R² for test data of 42.863 W/m², 82.396 W/m², 0.396 and 0.918. This study contributes to GHI prediction model that can handle missing values and outliers from sensors to support solar power plants.

The Comparative Analysis Of Multi-Criteria Decision-Making Methods (MCDM) In Priorities Of Industrial Location Development

2024-12-21T09:17:55+00:00

The process of prioritizing the development of an industrial area's site is a matter that necessitates a mature approach. The establishment of an industrial region has significant social implications for the surrounding locality. However, it is also necessary to take into account the availability of variables that facilitate the functioning of such an industrial zone. The goal of the study "A Comparative Analysis of Multi-Criteria Decision Making Methods (MCDM) for Determining the Priority of Industrial Area Location Development" is to compare and contrast different MCDM methods in the context of deciding which industrial area locations should be developed first. A case study was undertaken, examining various possible industrial sites for future development. Multiple approaches, namely MOORA, WASPAS, ARAS, COPRAS, and AHP, are employed to ascertain the prioritization of industrial area development locations. This study presents a comparative analysis of each approach by using the Spearman Rank correlation and utilizing the factual data obtained from the Department of Capital Plantation and Integrated One Door Services (DPMPTSP). The external research is anticipated to involve a comprehensive review of the literature on the efficacy of Multiple Criteria Decision Making (MCDM) methods. This research has the potential to assist both governmental bodies and private entities in establishing priorities for the development of industrial areas, taking into account prevailing circumstances and conditions while also considering various significant factors and criteria.

Two Omnidirectional Antenna Models for Ship Hull Corrosion Detection Radar System

2024-11-30T13:34:09+00:00

This paper discusses two modified monopole antenna models. These antenna models are designed to detect hull corrosion of ships, which is predicted to be the cause of hull leaks. The modification was carried out by adding a conical parasitic element aimed at widening the bandwidth and increasing the antenna gain. The conical parasitic on the first antenna is directed upwards, while the second is the opposite. The first model uses a cone with a larger diameter than the second model. The calculation of the cone diameter is based on half the wavelength of the frequencies generated by the antenna. With a smaller diameter, PTFE is added between the monopole and the parasitic cone to avoid short circuits. This configuration is intended to see the effective antenna model widen the bandwidth and increase the gain. Antenna performance testing was carried out using a vector network analyzer and software-defined radio. The test results show that both antenna models are omnidirectional in radiation. The first model operates at a frequency of 3904.7 - 7793.1 MHz with a bandwidth of 3888.4 MHz and a highest gain of 10 dBi. The second model operates at a frequency of 2282.4 - 3324.1 MHz with a bandwidth of 1041.7 MHz and a highest gain of 8dBi. Thus, the first antenna model has a higher bandwidth and gain than the first model, but both have met the requirements for ship hull corrosion detection antennas.

Revolutionizing Classroom Attendance: A Wireless Smart System Using ESP-NOW Protocol

2024-12-16T12:26:25+00:00

In an era where technological advancements drive improvements across various sectors, enhancing efficiency in educational management systems is crucial. This paper presents a novel wireless attendance system that leverages the ESP-NOW protocol, which offers advantages over traditional Wi-Fi by enabling low-power, low-latency, and direct device-to-device communication without the need for an intermediary network. The system employs ESP32 modules configured as both slave and master devices. Slave devices, positioned on the lecturer’s desk, interact with students' smartphones when the lecturer initiates the class, while master devices, strategically placed at multiple locations within the classroom, compile and consolidate attendance data for each room. The system incorporates RSSI-based restrictions via the ESP-NOW protocol to prevent overlapping attendance between rooms and ensure that students can only record their presence if they are physically within the designated classroom. Attendance data is automatically logged and made accessible in real-time through a dedicated mobile application for lecturers. Empirical testing demonstrates 100% accuracy in attendance recording, with an average verification time of less than 1 second and a data transmission rate ranging from 700 to 800 Bytes/seconds.

Design and Implementation of Current Control for Step Up–Down Zeta H-Bridge Inverter Using STM32F407VET6

2024-12-20T10:42:52+00:00

The development of electrical power conversion equipment is increasing, along with the utilization of new and renewable energy sources. Power conversion equipment from DC to AC voltage, known as inverters, is extensively researched and implemented in this sector. These inverters commonly operate as step-down voltage in specific applications used as step-ups with limited operating ranges. A step-up-down inverter with a single power circuit is developed to overcome this issue. Still, the number of power switches used correlates with the complexity of its control strategy. This paper investigates a step-up-down inverter using the Zeta H-Bridge Inverter with the implementation of six power switches. Furthermore, this type of inverter is operated with a controlled output current utilizing the STM32VET407 microcontroller. The control method is derived based on possible operational modes. An HX10-P current sensor detects the output current. It maintains itself according to the current reference by installing a proportional-integral controller. The initial verification utilizes computational simulation with power simulator software, ensuring the system operates as intended. The final stage involves implementation in the laboratory and testing with standardized equipment. The test results meet the IEEE 519 standard, where the output current has a THD of 1.1%.