eISSN : 3022-7011
ISSUER : KIPS
ISSUER : KIPS
After the Korea Information Processing Society (KIPS) Transactions journal was founded in 1994, it was reorganized into the KIPS Transactions: Computer and Communication Systems(2287-5891/2734-049X ) and the KIPS Transactions: Software and Data Engi neering(2287-5905/2734-0503) in 2012. Through the KIPS official meeting on January 8th, 2024, the new KIPS Transaction journal was founded by integrating two KIPS Journals, KIPS Transactions: Computer and Communication Systems and KIPS Transactions: Software and Data Engineering. The new journal aims to realize social value and contribute to the development of South Korea’s science and technology with support from the lottery fund of the Ministry of Strategy and Finance and the science/technology promotion fund of the Ministry of Science and ICT. It is indexed in the Korea Science Academic Database, Korea Citation Index (KCI), and EBSCO.
HighlightsMore
Performance Evaluation and Consideration of Shadow Stack on RISC-V ArchitectureKang Ha Young Han Go Won Park Sung Hwan Kwon Dong Hyun |
|
RISC-V is an open-source instruction set architecture, used in various hardware implementations, and can be flexibly expanded to meet system requirements through the RV64I base instruction set and 16 standard extensions. Currently, the RISC-V archit... | |
A Study on the Evaluation Methods for Assessing the Understanding of Korean Culture by Generative AI ModelsSon Ki Jun Kim Seung Hyun |
|
Recently, services utilizing large-scale language models (LLMs) such as GPT-4 and LLaMA have been released, garnering significant attention. These models can respond fluently to various user queries, but their insufficient training on Korean data ra... | |
Protection on WebAssembly JIT-Compiled Code with Randomized Memory Protection KeyShin Chae Won Jeong Yun Seo Bae Myeong Jin Kwon Dong Hyun |
|
WebAssembly(Wasm) is a powerful platform that enables compiled code in various programming languages to be executed in web browser and in varied runtime environments. Specifically, for performance optimization, Wasm runtime provides Just-in-Time (JI... | |
A Study on Automatic Metrics for Korean Text Abstractive SummarizationSehwi Yoon Youhyun Shin |
|
This study aims to analyze and validate automatic evaluation metrics for Korean abstractive summarization. The unique linguistic characteristics of each language require evaluation metrics designed for them, underscoring the importance of research f... | |
Latest Publication (Vol. 14, No. 6, Jun. 2025)
k-obstacle Based Anonymity Trajectory(k-oATY) Methods to Protect the User’s k-trajectories in Continuous Query Processing
Song Doo Hee
https://doi.org/10.3745/TKIPS.2025.14.6.379
Location-based service information protection continuous query processing k-trajectories(k-T) Guaranteed k-trajectories(k-TG)
Most smartphone users have options that must be agreed upon when installing or using the application. In particular, the number
of cases requiring the location of users is gradually increasing. The reason why the service provider(server) requests location information
from the user is that it aims to provide a quick and accurate service to the user. However, when analyzing the user’s behavioral patterns
over time, it is very likely to include sensitive personal information. In particular, when grasping the building or place where they stayed
(shopping malls, clubs, etc.), the user’s trajectory can expect not only behavioral patterns but also interests. Therefore, this paper propose
a k-obstacle based anonymity trajectory(k-oATY) method that can protect the user’s trajectory. Through experiments, it was proved that
the proposed method has a higher probability of protecting individual location information than the existing method.
The Transactions of the Korea Information Processing Society,
Vol. 14, No. 6, pp. 379-385,
Jun.
2025

Location-based service information protection continuous query processing k-trajectories(k-T) Guaranteed k-trajectories(k-TG)
Effectiveness of Digital Health-Enabled Pelvic Floor Muscle Training for Urinary Incontinence: A Systematic Review and Meta-Analysis
In Gyeom Yoo Gaeun Kim
https://doi.org/10.3745/TKIPS.2025.14.6.386
Urinary incontinence Pelvic floor muscle training Digital health Mobile health Wearable Devices System design meta-analysis
This review investigates the clinical effectiveness of pelvic floor muscle training (PFMT) interventions delivered through digital health
modalities for individuals experiencing urinary incontinence (UI). It also explores embedded system-level features that may shape adherence
behavior and health outcomes. Extensive searches were undertaken across global (PubMed, EMBASE, CINAHL, CENTRAL) and Korean
(RISS, KISS) academic databases for relevant studies published from 2000 to 2023. Only randomized controlled trials (RCTs) evaluating
digital PFMT programs delivered through mobile apps, telehealth systems, or wearable technologies were selected. Risk of bias was assessed
using the Cochrane RoB 2 tool. Meta-analytic synthesis with a random-effects model was used to calculate pooled effects. Additionally,
a narrative synthesis examined digital functionalities including feedback mechanisms, adherence tracking, and user engagement tools.
Ten eligible RCTs involving 1,342 participants were analyzed. Digital PFMT yielded statistically significant benefits in reducing symptom
severity (SMD = −0.46; 95% CI: −0.73, −0.19), alleviating urinary distress (SMD = −0.41; 95% CI: −0.68, −0.14), and enhancing quality
of life (SMD = 0.39; 95% CI: 0.11, 0.67). Platforms incorporating real-time feedback loops, gamified interfaces, and structured reminders
were associated with better adherence and sustained benefits. In contrast, few interventions integrated AI-based personalization or ensured
interoperability with other digital systems. Digital PFMT approaches demonstrate clinical value in managing UI and improving
patient-reported outcomes. Emphasizing real-time interactivity, tailored feedback, and user-centered system architecture could further
optimize intervention adherence and impact. The findings advocate for the continued evolution and integration of smart PFMT tools
within comprehensive digital health strategies.
The Transactions of the Korea Information Processing Society,
Vol. 14, No. 6, pp. 386-396,
Jun.
2025

Urinary incontinence Pelvic floor muscle training Digital health Mobile health Wearable Devices System design meta-analysis
Compressed LLM-Based Secure Coding Support
Chan Woo Lee Junyoung Heo
https://doi.org/10.3745/TKIPS.2025.14.6.397
Secure Coding Model Compression Large Language Model(LLM)
As the importance of security in the IT industry continues to grow,
secure coding has become essential. However, small-scale enterprises
often struggle to fully implement secure coding guidelines due to a lack
of specialized security personnel. This study proposes a compressed
Large Language Model (LLM)-based approach to support secure
coding, enabling cost-effective adoption of secure coding guidelines.
The proposed method leverages an open-source model fine-tuned on
a security dataset for domain-specific learning and applies the
Low-Rank Adaptation (LoRA) technique to optimize training efficiency
in a single GPU environment. Additionally, it incorporates BF16
transformation and 8-bit GGUF quantization to reduce model size,
ensuring operability in standard computing environments. Through
experiments, we evaluated the proposed model’s ability to detect
security vulnerabilities and generate improved code suggestions. The
results demonstrate superior performance over the baseline model in
key evaluation metrics, including F1 score and BLEU score.
The Transactions of the Korea Information Processing Society,
Vol. 14, No. 6, pp. 397-399,
Jun.
2025

Secure Coding Model Compression Large Language Model(LLM)
A Systematic Framework for Enhancing Retrieval-Augmented Generation for Tabular Data
Eunbin Lee Younghan Lee Ho Bae ·
https://doi.org/10.3745/TKIPS.2025.14.6.400
RAG Tabular data performance optimization Empirical Analysis
This study proposes a novel framework approach combining structural data processing, user query classification, and a self-feedback
loop to optimize the performance of Retrieval-Augmented Generation (RAG) models for effectively handling tabular data. While RAG models
excel at complex data processing and question answering by integrating retrieval-based and generative capabilities, specific design
strategies tailored to the unique characteristics of tabular data have not been sufficiently explored. To address this gap, this study presents
a detailed design framework focused on improving the key components of RAG models in tabular data processing. The proposed framework
systematically enhances the efficiency and accuracy of RAG models by refining their ability to structure and interpret tabular datasets.
Furthermore, an empirical analysis was conducted to determine the optimal combination of generative models, embedding models, and
the number of retrieved documents for effective tabular data processing. To support this study, a new QA dataset was constructed to
better evaluate RAG models on tabular data tasks. Experimental results demonstrate that the structural approach explicitly preserves the
relationships and context within the data, while the user query classification strategy contributes to maximizing the efficiency of the
RAG process. Additionally, the self-feedback loop enhances the quality of generated responses through iterative evaluation and refinement,
effectively mitigating hallucination issues and ensuring reliable, high-quality responses even for complex queries. By integrating these
optimization strategies, this study refines the design direction for RAG models tailored to tabular data and provides practical insights
into their deployment. This work expands the applicability of RAG models specialized in tabular data processing and enhances their
potential across various application domains. These findings provide a foundational resource for RAG model design and performance
optimization, offering valuable guidance for addressing practical challenges in tabular data processing and advancing future AI-driven
data analysis systems.
The Transactions of the Korea Information Processing Society,
Vol. 14, No. 6, pp. 400-416,
Jun.
2025

RAG Tabular data performance optimization Empirical Analysis
Machine Learning-Based Gambling Addiction Risk Prediction Model
Lee Seo Jin Choi Ji Woong Hong Seung Hyeon Kim Jongwan
https://doi.org/10.3745/TKIPS.2025.14.6.417
Machine Learning Supervised Learning Gambling Addiction Logistic Regression Counseling Psychology
This paper proposes a machine learning-based prediction model to distinguish gambling addicts according to their level of addiction
to mitigate gambling-related problems. With the widespread use of the Internet and smartphones, gambling accessibility has increased,
leading to a growing number of individuals experiencing gambling addiction. Gambling addiction is not just a personal issue but a significant
societal problem that requires early detection and intervention. To prevent gambling problems and strengthen treatment interventions,
a model capable of quickly and accurately identifying gambling addicts is necessary. In this study, we propose a method to predict addiction
risk levels based on an individual’s gambling behavior using a machine learning model trained on data from the Korean version of the
Canadian Problem Gambling Index (K-CPGI). We compared and analyzed various machine learning models and found that logistic
regression demonstrated relatively high performance. Considering its interpretability and reliability, it was selected as the final model.
However, due to the limitations of the dataset used in this study, we discuss the need for additional data collection and validation of
generalization performance. Through this research, we expect to enhance the early and accurate identification of gambling addicts, thereby
strengthening treatment linkage and contributing to the development of gambling addiction prevention and intervention strategies.
The Transactions of the Korea Information Processing Society,
Vol. 14, No. 6, pp. 417-423,
Jun.
2025

Machine Learning Supervised Learning Gambling Addiction Logistic Regression Counseling Psychology
Fast Generation of Large Maps Using Variable Block Based Wave Function Collapse Algorithm
Kim Tae Hwan Sung Man Kyu
https://doi.org/10.3745/TKIPS.2025.14.6.424
PCG WFC Model Synthesis Algorithm Game Level Design
This paper proposes a method to quickly generate large-scale 2D maps using the Wave Function Collapse algorithm using variable-sized
blocks. Procedural Content Generation (PCG) is one of the techniques for automatically generating maps. The Wave Function Collapse
algorithm is one of the PCG techniques and has been used to generate 2D maps to satisfy constraints. However, it has the problem
that the generation time increases exponentially as the map size increases. The experimental results of the proposed algorithm in this
study showed that it was at least five times faster than existing algorithms when generating large-scale maps.
The Transactions of the Korea Information Processing Society,
Vol. 14, No. 6, pp. 424-430,
Jun.
2025

PCG WFC Model Synthesis Algorithm Game Level Design
A Study on Machine Learning -based River Water Level Prediction Model for Flood Prevention
DeokHyun You Heesun Roh Suyeon Lim Seongbok Baik Yong-Geun Hong
https://doi.org/10.3745/TKIPS.2025.14.6.431
River Water Level Prediction Flood Prediction Machine Learning-based Predcition Models
This study compares various data-driven modeling approaches for river water level prediction to support the development of accurate
forecasting systems for disaster response. Initially, a CNN-based model was applied for image-based classification of water level stages.
Although the model achieved high accuracy, limitations in generalization were observed due to class imbalance and location-specific
bias. Techniques such as dropout and data augmentation were introduced, but they offered limited improvement. Subsequent time series
prediction experiments employed linear regression, polynomial regression, and LSTM models, comparing Shift and Sliding Window input
methods. Across all models, the Sliding Window method consistently outperformed the Shift method in terms of R², NSE, and PBIAS,
with especially notable differences in long-term forecasts. The multivariate LSTM model, which incorporated additional variables such
as flow rate, demonstrated superior accuracy and stability compared to its univariate counterpart, achieving R² values exceeding 0.96
for 3-hour forecasts. Overall, the combination of the Sliding Window approach and multivariate modeling was shown to be highly effective
for improving prediction performance. Future work will explore advanced nonlinear models such as Transformers, as well as multimodal
data integration including rainfall and CCTV imagery to enhance the accuracy and applicability of river level forecasting systems.
The Transactions of the Korea Information Processing Society,
Vol. 14, No. 6, pp. 431-450,
Jun.
2025

River Water Level Prediction Flood Prediction Machine Learning-based Predcition Models
DQL-based Time Division Duplex Configuration Optimization Technique in 5G New Radio
Muneeb Muhammad Kwang-Man Ko
https://doi.org/10.3745/TKIPS.2025.14.6.451
5G NR Time Division Duplex (TDD) Uplink traffic Downlink traffic Deep Q Learning (DQL)
Emerging 5G and 6G networks demand high bandwidth, ultra-low latency, and adaptable traffic management to facilitate services
which are transitioning to balanced uplink traffic from heavily downlink-dependent traffic. In contrast to 4G Time Division Duplex (TDD),
which uses static Uplink/Downlink (UL/DL) configurations, 5GNR enables dynamic modification of defined procedures for the best possible
pattern design. In order to optimize TDD pattern selection in 5G New Radio (NR) networks, this paper suggests a Deep Q Learning (DQL)
architecture. It was created to model important network characteristics, including as user equipment (UE) density, interference, and uplink
and downlink traffic loads. Optimizing uplink (UL) and downlink (DL) slot designs is essential for 5G networks based on TDD in order
to satisfy dynamic traffic needs while preserving low latency and high throughput. A DQL method that learns to modify UL-DL configurations
in response to real-time traffic patterns is presented in this study. The DQL agent is trained in a simulated 5G environment to maximize
a reward function that combines latency reduction and throughput gain. The DQL-based approach’s performance is evaluated in
comparison to a static 4:4 UL-DL configuration baseline. The suggested paradigm dramatically increases network performance and lowers
latency, according to experimental data. In particular, the DQL agent outperforms the baseline by more than 95% in throughput and
50% in latency, demonstrating its promise for intelligent and flexible slot scheduling in TDD systems of the future.
The Transactions of the Korea Information Processing Society,
Vol. 14, No. 6, pp. 451-458,
Jun.
2025

5G NR Time Division Duplex (TDD) Uplink traffic Downlink traffic Deep Q Learning (DQL)
Transformer Based Korean Emotion Recognition Model through Multi-domain Fusion
Jinhwan Yang Hyuksoon Choi Nammee Moon Jinah Kim
https://doi.org/10.3745/TKIPS.2025.14.6.459
Emotion Recognition Domain Fusion Transformer encoder Information Compression Audio Classification
This study proposes a transformer-based emotion recognition model that enhances performance through multi-domain fusion. The
proposed model extracts and compresses emotion-relevant information from three domains—audio, spectrogram, and text—using a feature
encoder and a transformer encoder, thereby improving recognition accuracy. To evaluate the performance of the proposed model,
experiments were conducted comparing different domain combinations and backbone architectures. The results demonstrate that all three
domains effectively contribute to improved emotion recognition performance, and that ResNet50 is the most suitable backbone. The model
trained on all three domains achieved an accuracy of 0.9306 and an F1-score of 0.9306, outperforming models trained on other domain
combinations. These findings suggest that multi-domain fusion helps enhance the precision of emotion recognition and indicate that
the proposed model can serve as a practical baseline for multimodal emotion recognition research.
The Transactions of the Korea Information Processing Society,
Vol. 14, No. 6, pp. 459-467,
Jun.
2025

Emotion Recognition Domain Fusion Transformer encoder Information Compression Audio Classification
Evaluating The Openness of Impactful AI Models with A Focus on LLMs
Kil-Won Jeon Hyun-Jun Han Kang-Won Lee
https://doi.org/10.3745/TKIPS.2025.14.6.468
Open Source AI Model AI Model Openness Large Language Model Openness Evaluation Framework
Generative AI models are increasingly driving technical innovations and making societal impacts. Recognizing their significance,
governments(e.g, EU AI Act) and technical communities (e.g., OSI) demand a higher degree of openness and transparency in AI development.
Open sourcing AI models enables deeper scrutiny of their inner workings, accelerates innovation, and mitigates potential risks. Although
numerous AI models are marketed as “open source,” many fall short of the traditional standards of openness. Moreover, despite recent
efforts, there is currently no comprehensive framework for characterizing the openness of AI models. In this paper, we propose a novel
framework to quantify the degree of openness in AI models. We apply our framework to evaluate several high-impact models, including
models developed by Korean companies, and investigate the relationship between openness and performance of AI models with a focus
on large language models (LLMs).
The Transactions of the Korea Information Processing Society,
Vol. 14, No. 6, pp. 468-479,
Jun.
2025

Open Source AI Model AI Model Openness Large Language Model Openness Evaluation Framework
Intelligent NPC AI Based on FSM and Reinforcement Learning: Implementation Study of A 2D Mobile Idle RPG Game
Dongju Kim Seok-Joo Koh
https://doi.org/10.3745/TKIPS.2025.14.6.480
Idle RPG NPC AI Finite state machine Reinforcement Learning Combat System Optimization
This study focuses on the design and implementation of an intelligent NPC AI system that integrates Finite State Machines (FSM) and
reinforcement learning in mobile idle RPG games. By combining FSM with reinforcement learning techniques such as Q-learning and
Deep Q-Networks (DQN), the system optimizes NPCs’ environmental perception, combat strategies, and state management. The NPC AI
dynamically transitions between various states such as idle, attack, and retreat, and experimental evaluation involved collecting and
analyzing performance metrics such as combat win rates, response times, and movement distances to refine behavioral patterns and
maintain game balance. This study proposes a method for enhancing NPC AI performance in idle RPGs through the integrated application
of FSM and reinforcement learning. Through real-time data analysis, the difficulty and behavior patterns of NPCs were dynamically adjusted,
and an intelligent NPC system was implemented that adapts to various combat environments and situations. This research provides a
guide for developing auto-roguelike RPG games and can be used as foundational material for enhancing game experiences based on
real-time AI optimization.
The Transactions of the Korea Information Processing Society,
Vol. 14, No. 6, pp. 480-488,
Jun.
2025

Idle RPG NPC AI Finite state machine Reinforcement Learning Combat System Optimization
Development of An Interactive KoBASIC Using Jupyter
Lee Seok Won Gyun Woo
https://doi.org/10.3745/TKIPS.2025.14.6.489
Korean programming language KoBASIC Jupyter Kernel Educational Programming Language Interactive Environment
Recently, there has been a growing demand for programming education for Korean-based languages that lower entry barriers for
beginners. KoBASIC was developed in response to this need as a Korean-language educational programming language, designed with
an intuitive syntax to enable easy access for novice learners. However, the original implementation of KoBASIC was limited to the Windows
operating system and required a separate executable, causing inconvenience and reduced accessibility. To tackle these problems, this
paper proposes and implements a dedicated Jupyter Notebook kernel for KoBASIC. The proposed system allows users to write and execute
KoBASIC code directly through a web browser, regardless of the operating system. In particular, 0MQ-based message communication
was adopted between the Jupyter server and the KoBASIC kernel, and additional control commands such as list and run were newly
introduced. When applied to an actual university course, the system showed a 20% point increase in the proportion of students achieving
perfect scores, from 46% to 66%, even with minimal instruction time. A brief follow-up survey also indicated that users found using KoBASIC
in the Jupyter environment more convenient than the previous method. Furthermore, compared to existing Korean programming languages
such as HanBASIC and Saesark, KoBASIC was confirmed to be more concise, with the number of syntactic elements reduced to
approximately 41% and 70%, respectively.
The Transactions of the Korea Information Processing Society,
Vol. 14, No. 6, pp. 489-496,
Jun.
2025

Korean programming language KoBASIC Jupyter Kernel Educational Programming Language Interactive Environment