Vungsovanreach Kong
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Vungsovanreach Kong has a total of 22 co-authors. Here are the top co-authors by number of shared publications:
- Tae-Kyung Kim
Department of Management Information Systems, Chungbuk National University, Seowon-gu, Cheongju-si, Chungcheongbuk-do, South Korea
4 shared publications - Anand Nayyar
Graduate School, Faculty of Information Technology, Duy Tan University, Da Nang, Vietnam
3 shared publications - Ho Sun Shon
Medical Research Institute, School of Medicine, Chungbuk National University, Cheongju, Republic of Korea
3 shared publications - Kyung Ah Kim
Department of Biomedical Engineering, School of Medicine, Chungbuk National University, Cheongju 28644, Korea
3 shared publications Chang Kue Seok
2 shared publications- Eun-Young Lee
Department of Oral and Maxillofacial Surgery, College of Medicine and Medial Research Institute, Chungbuk National University, Cheongju 28644, Republic of Korea
2 shared publications
Latest Publications
- Open Access
ASG-LLM: An LLM-based pipeline for automated scraping code generation in security-sensitive environments
- Open Access
- 9 June 2026
- Published by Elsevier in SSRN Electronic Journal
ASG-LLM: An LLM-Based Pipeline for Automated Scraping Code Generation in Security-Sensitive Environments
Automating interactions with security-sensitive web services remains difficult due to dynamic frontends, session-bound tokens, and undocumented parameter-generation logic. Traditionally, developers must manually analyze captured traffic, hypothesize parameter-generation routines, and iteratively test internally defined functions until the interaction succeeds. We propose ASG-LLM, an execution grounded pipeline that generates scraping modules from a single successful browser session trace. ASG-LLM captures full HTTP(S) traffic, removes irrelevant requests through multi-stage filtering, and constructs a dependency-aware representation of the core interaction flow. When security-critical parameters cannot be reconstructed from the trace alone, the FUNC Retriever identifies a small set of reusable internal function descriptions using HyDE-guided selective retrieval, enabling the LLM to select compatible parameter-generation routines based on behavioral similarity rather than keyword matching, without exposing proprietary implementation code. Generated modules are executed in a controlled environment and iteratively repaired using structured failure feedback. ASG-LLM was evaluated on representative banking and public-sector portals. In our case studies, ASG-LLM reduces developer effort by cutting traffic analysis and basic code implementation from hours to minutes and reducing security-handling effort from a full-day task to a couple of hours. At the model level, HyDE-based selective retrieval improves internal function recommendation accuracy from 30% to 70% Top-1 and from 60% to 90% Top-3 compared with a full-context baseline. These results suggest that combining trace grounding, selective retrieval, and execution-time validation is a promising direction for robust web automation in security-constrained environments.
- 9 June 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessASG-LLM: An LLM-Based Pipeline for Automated Scraping Code Generation in Security-Sensitive Environments
Automating interactions with security-sensitive web services remains difficult due to dynamic frontends, session-bound tokens, and undocumented parameter-generation logic. Traditionally, developers must manually analyze captured traffic, hypothesize parameter-generation routines, and iteratively test internally defined functions until the interaction succeeds. We propose ASG-LLM, an execution grounded pipeline that generates scraping modules from a single successful browser session trace. ASG-LLM captures full HTTP(S) traffic, removes irrelevant requests through multi-stage filtering, and constructs a dependency-aware representation of the core interaction flow. When security-critical parameters cannot be reconstructed from the trace alone, the FUNC Retriever identifies a small set of reusable internal function descriptions using HyDE-guided selective retrieval, enabling the LLM to select compatible parameter-generation routines based on behavioral similarity rather than keyword matching, without exposing proprietary implementation code. Generated modules are executed in a controlled environment and iteratively repaired using structured failure feedback. ASG-LLM was evaluated on representative banking and public-sector portals. In our case studies, ASG-LLM reduces developer effort by cutting traffic analysis and basic code implementation from hours to minutes and reducing security-handling effort from a full-day task to a couple of hours. At the model level, HyDE-based selective retrieval improves internal function recommendation accuracy from 30% to 70% Top-1 and from 60% to 90% Top-3 compared with a full-context baseline. These results suggest that combining trace grounding, selective retrieval, and execution-time validation is a promising direction for robust web automation in security-constrained environments.
- Open Access
- 11 February 2026
- Published by PeerJ in PeerJ Computer Science
- Vol. 12, e3525
QA-SQL: query-augmented SQL generation using few-shot prompting with data augmentation
Natural Language to Structured Query Language (NL-to-SQL) models make it easier to retrieve data from structured databases by converting plain language queries into Structured Query Language (SQL) commands. However, current methods often struggle with complex database structures, unclear user questions, and generating efficient SQL queries. These issues make it hard to use them effectively in real-world, large-scale databases. To tackle these problems, a novel framework titled “Query-Augmented SQL Generation (QA-SQL)”, is introduced that combines retrieval augmented generation (RAG) with data augmentation with advanced techniques to enhance the accuracy and efficiency of natural language to SQL (NL-to-SQL) generation. The proposed approach uses Schema Linking, RAG, and Dynamic Prompt Engineering along with few-shot learning and vector database retrieval, to enhance SQL query generation through query augmentation. To test and validate the proposed model, extensive experimentations were conducted on the BIg Bench for LaRge-scale Database Grounded Text-to-SQL Evaluation (BIRD) benchmark, which includes 1,534 natural language queries across various relational databases and the results highlighted a execution accuracy of 46.68% and a valid efficiency score of 87.66%, outperforming existing methods. The proposed research introduces a context-aware retrieval system that significantly enhances SQL accuracy, paving the way for more scalable and reliable NL-to-SQL solutions.
- 11 February 2026
- Published by PeerJ in PeerJ Computer Science
- Vol. 12, e3525
Open AccessQA-SQL: query-augmented SQL generation using few-shot prompting with data augmentation
Natural Language to Structured Query Language (NL-to-SQL) models make it easier to retrieve data from structured databases by converting plain language queries into Structured Query Language (SQL) commands. However, current methods often struggle with complex database structures, unclear user questions, and generating efficient SQL queries. These issues make it hard to use them effectively in real-world, large-scale databases. To tackle these problems, a novel framework titled “Query-Augmented SQL Generation (QA-SQL)”, is introduced that combines retrieval augmented generation (RAG) with data augmentation with advanced techniques to enhance the accuracy and efficiency of natural language to SQL (NL-to-SQL) generation. The proposed approach uses Schema Linking, RAG, and Dynamic Prompt Engineering along with few-shot learning and vector database retrieval, to enhance SQL query generation through query augmentation. To test and validate the proposed model, extensive experimentations were conducted on the BIg Bench for LaRge-scale Database Grounded Text-to-SQL Evaluation (BIRD) benchmark, which includes 1,534 natural language queries across various relational databases and the results highlighted a execution accuracy of 46.68% and a valid efficiency score of 87.66%, outperforming existing methods. The proposed research introduces a context-aware retrieval system that significantly enhances SQL accuracy, paving the way for more scalable and reliable NL-to-SQL solutions.
- Open Access
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 149604-149625
MulFD-Net: Per-Scale Multi-Layer Feature Decomposition for Transparent and Reflective Object Segmentation
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 149604-149625
Open AccessMulFD-Net: Per-Scale Multi-Layer Feature Decomposition for Transparent and Reflective Object Segmentation
- Open Access
- 31 August 2025
- Published by Korea Disease Control and Prevention Agency in Osong Public Health and Research Perspectives
- Vol. 16 (4) , 333-347
Deep learning-based prognosis of major adverse cardiac events in patients with acute myocardial infarction: a retrospective observational study in the Republic of Korea
Objectives: This study developed deep neural network (DNN) models capable of accurately classifying major adverse cardiac events (MACE) in patients with acute myocardial infarction (AMI) after hospital discharge, across 3 follow-up intervals: 1, 6, and 12 months.Methods: DNN models were constructed to predict post-discharge MACE across 4 categories. Multiple traditional machine learning models were implemented as controls to benchmark the performance of our DNN approach. All models were evaluated based on their ability to predict MACE occurrence during the specified follow-up periods.Results: The DNN models demonstrated superior predictive performance over conventional machine learning methods, achieving high accuracies of 0.922, 0.884, and 0.913 for the 1-month, 6-month, and 12-month follow-up periods, respectively.Conclusion: The high accuracy of our DNN models highlights their practical advantages for AMI diagnosis and guidance of follow-up treatment. These models can serve as valuable decision support tools, enabling clinicians to optimize the overall management of AMI patients and potentially enhance their hospitalization experience.
- 31 August 2025
- Published by Korea Disease Control and Prevention Agency in Osong Public Health and Research Perspectives
- Vol. 16 (4) , 333-347
Open AccessDeep learning-based prognosis of major adverse cardiac events in patients with acute myocardial infarction: a retrospective observational study in the Republic of Korea
Objectives: This study developed deep neural network (DNN) models capable of accurately classifying major adverse cardiac events (MACE) in patients with acute myocardial infarction (AMI) after hospital discharge, across 3 follow-up intervals: 1, 6, and 12 months.Methods: DNN models were constructed to predict post-discharge MACE across 4 categories. Multiple traditional machine learning models were implemented as controls to benchmark the performance of our DNN approach. All models were evaluated based on their ability to predict MACE occurrence during the specified follow-up periods.Results: The DNN models demonstrated superior predictive performance over conventional machine learning methods, achieving high accuracies of 0.922, 0.884, and 0.913 for the 1-month, 6-month, and 12-month follow-up periods, respectively.Conclusion: The high accuracy of our DNN models highlights their practical advantages for AMI diagnosis and guidance of follow-up treatment. These models can serve as valuable decision support tools, enabling clinicians to optimize the overall management of AMI patients and potentially enhance their hospitalization experience.
- Open Access
- 17 June 2025
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 20 (6) , e0325253
PIFR: A novel approach for analyzing pose angle-based human activity to automate fall detection in videos
Falls pose a significant health risk for elderly populations, necessitating advanced monitoring technologies. This study introduces a novel two-stage fall detection system that combines computer vision and machine learning to accurately identify fall events. The system uses the YOLOv11 object detection model to track individuals and estimate their body pose by identifying 17 key body points across video frames. The proposed approach extracts nine critical geometric features, including the center of mass and various body angles. These features are used to train a support vector machine (SVM) model for binary classification, distinguishing between standing and lying with high precision. The system’s temporal validation method analyzes sequential frame changes, ensuring robust fall detection. Experimental results, evaluated on the University of Rzeszow Fall Detection (URFD) dataset and the Multiple Cameras Fall Dataset (MCFD), demonstrate exceptional performance, achieving 88.8% precision, 94.1% recall, an F1-score of 91.4%, and a specificity of 95.6%. The method outperforms existing approaches by effectively capturing complex geometric changes during fall events. The system is applicable to smart homes, wearable devices, and healthcare monitoring platforms, offering a scalable, reliable, and efficient solution to enhance safety and independence for elderly individuals, thereby contributing to advancements in health-monitoring technology.
- 17 June 2025
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 20 (6) , e0325253
Open AccessPIFR: A novel approach for analyzing pose angle-based human activity to automate fall detection in videos
Falls pose a significant health risk for elderly populations, necessitating advanced monitoring technologies. This study introduces a novel two-stage fall detection system that combines computer vision and machine learning to accurately identify fall events. The system uses the YOLOv11 object detection model to track individuals and estimate their body pose by identifying 17 key body points across video frames. The proposed approach extracts nine critical geometric features, including the center of mass and various body angles. These features are used to train a support vector machine (SVM) model for binary classification, distinguishing between standing and lying with high precision. The system’s temporal validation method analyzes sequential frame changes, ensuring robust fall detection. Experimental results, evaluated on the University of Rzeszow Fall Detection (URFD) dataset and the Multiple Cameras Fall Dataset (MCFD), demonstrate exceptional performance, achieving 88.8% precision, 94.1% recall, an F1-score of 91.4%, and a specificity of 95.6%. The method outperforms existing approaches by effectively capturing complex geometric changes during fall events. The system is applicable to smart homes, wearable devices, and healthcare monitoring platforms, offering a scalable, reliable, and efficient solution to enhance safety and independence for elderly individuals, thereby contributing to advancements in health-monitoring technology.
- Open Access
- 8 November 2024
- Published by MDPI AG in Bioengineering
- Vol. 11 (11) , 1130
Integrating Super-Resolution with Deep Learning for Enhanced Periodontal Bone Loss Segmentation in Panoramic Radiographs
Periodontal disease is a widespread global health concern that necessitates an accurate diagnosis for effective treatment. Traditional diagnostic methods based on panoramic radiographs are often limited by subjective evaluation and low-resolution imaging, leading to suboptimal precision. This study presents an approach that integrates Super-Resolution Generative Adversarial Networks (SRGANs) with deep learning-based segmentation models to enhance the segmentation of periodontal bone loss (PBL) areas on panoramic radiographs. By transforming low-resolution images into high-resolution versions, the proposed method reveals critical anatomical details that are essential for precise diagnostics. The effectiveness of this approach was validated using datasets from the Chungbuk National University Hospital and the Kaggle data portal, demonstrating significant improvements in both image resolution and segmentation accuracy. The SRGAN model, evaluated using the Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) metrics, achieved a PSNR of 30.10 dB and an SSIM of 0.878, indicating high fidelity in image reconstruction. When applied to semantic segmentation using a U-Net architecture, the enhanced images resulted in a dice similarity coefficient (DSC) of 0.91 and an intersection over union (IoU) of 84.9%, compared with 0.72 DSC and 65.4% IoU for native low-resolution images. These results underscore the potential of SRGAN-enhanced imaging to improve PBL area segmentation and suggest broader applications in medical imaging, where enhanced image clarity is crucial for diagnostic accuracy. This study also highlights the importance of further research to expand the dataset diversity and incorporate clinical validation to fully realize the benefits of super-resolution techniques in medical diagnostics.
- 8 November 2024
- Published by MDPI AG in Bioengineering
- Vol. 11 (11) , 1130
Open AccessIntegrating Super-Resolution with Deep Learning for Enhanced Periodontal Bone Loss Segmentation in Panoramic Radiographs
Periodontal disease is a widespread global health concern that necessitates an accurate diagnosis for effective treatment. Traditional diagnostic methods based on panoramic radiographs are often limited by subjective evaluation and low-resolution imaging, leading to suboptimal precision. This study presents an approach that integrates Super-Resolution Generative Adversarial Networks (SRGANs) with deep learning-based segmentation models to enhance the segmentation of periodontal bone loss (PBL) areas on panoramic radiographs. By transforming low-resolution images into high-resolution versions, the proposed method reveals critical anatomical details that are essential for precise diagnostics. The effectiveness of this approach was validated using datasets from the Chungbuk National University Hospital and the Kaggle data portal, demonstrating significant improvements in both image resolution and segmentation accuracy. The SRGAN model, evaluated using the Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) metrics, achieved a PSNR of 30.10 dB and an SSIM of 0.878, indicating high fidelity in image reconstruction. When applied to semantic segmentation using a U-Net architecture, the enhanced images resulted in a dice similarity coefficient (DSC) of 0.91 and an intersection over union (IoU) of 84.9%, compared with 0.72 DSC and 65.4% IoU for native low-resolution images. These results underscore the potential of SRGAN-enhanced imaging to improve PBL area segmentation and suggest broader applications in medical imaging, where enhanced image clarity is crucial for diagnostic accuracy. This study also highlights the importance of further research to expand the dataset diversity and incorporate clinical validation to fully realize the benefits of super-resolution techniques in medical diagnostics.
- Open Access
- 25 August 2022
- Published by MDPI AG in Applied Sciences
- Vol. 12 (17) , 8500
Deep Learning Model for Classifying Periodontitis Stages on Dental Panoramic Radiography
In this study, an integrated deep learning framework was developed for classifying the periodontitis stages of each individual tooth using dental panoramic radiographs. Based on actual patient panoramic radiographs data, the bone loss by periodontitis and cementoenamel junction boundaries were detected, while the tooth number and tooth length were identified using data from AIHub, an open database platform. The two factors were integrated to classify and to evaluate the periodontitis staging on dental panoramic radiography. Periodontitis is classified into four stages based on the criteria of the radiographic bone level, as suggested at the relevant international conference in 2017. For the integrated deep learning framework developed in this study, the classification performance was evaluated by comparing the results of dental specialists, which indicated that the integrated framework had an accuracy of 0.929, with a recall and precision of 0.807 and 0.724, respectively, in average across all four stages. The novel framework was thus shown to exhibit a relatively high level of performance, and the findings in this study are expected to assist dental specialists with detecting the periodontitis stage and subsequent effective treatment. A systematic application will be developed in the future, to provide ancillary data for diagnosis and basic data for the treatment and prevention of periodontal disease.
- 25 August 2022
- Published by MDPI AG in Applied Sciences
- Vol. 12 (17) , 8500
Open AccessDeep Learning Model for Classifying Periodontitis Stages on Dental Panoramic Radiography
In this study, an integrated deep learning framework was developed for classifying the periodontitis stages of each individual tooth using dental panoramic radiographs. Based on actual patient panoramic radiographs data, the bone loss by periodontitis and cementoenamel junction boundaries were detected, while the tooth number and tooth length were identified using data from AIHub, an open database platform. The two factors were integrated to classify and to evaluate the periodontitis staging on dental panoramic radiography. Periodontitis is classified into four stages based on the criteria of the radiographic bone level, as suggested at the relevant international conference in 2017. For the integrated deep learning framework developed in this study, the classification performance was evaluated by comparing the results of dental specialists, which indicated that the integrated framework had an accuracy of 0.929, with a recall and precision of 0.807 and 0.724, respectively, in average across all four stages. The novel framework was thus shown to exhibit a relatively high level of performance, and the findings in this study are expected to assist dental specialists with detecting the periodontitis stage and subsequent effective treatment. A systematic application will be developed in the future, to provide ancillary data for diagnosis and basic data for the treatment and prevention of periodontal disease.