| 金地 琳太郎 | M, 1回目発表 | ソフトウェア設計学 | 飯田 元, | 門林 雄基, | 柏 祐太郎, | Reid Brittany |
|
title: An Empirical Study of Security-Policy Related Issues in Open Source Projects
abstract: GitHub recommends that projects adopt a security file that outlines vulnerability reporting procedures. However, the effectiveness and operational challenges of such files are not yet fully understood. This study aims to clarify the challenges that security files face in the vulnerability reporting process within open-source communities. Specifically, we classified and analyzed the content of 711 randomly sampled issues related to security files. We also conducted a quantitative comparative analysis of the close time and number of responses for issues concerning six community health files, including security files. Our findings offer practical insights for improving security reporting policies and community management, ultimately contributing to a more secure open-source ecosystem. language of the presentation: English | ||||||
| 小松 聖矢 | D, 中間発表 | 情報基盤システム学 | 藤川 和利, | 門林 雄基, | 林 優一, | 新井 イスマイル |
|
title: A Study on Network-Environment-Adaptive Malware Detection by Integrating Time-Window Traffic Segmentation with Dynamic Graph Learning
abstract: In network-traffic-based malware detection, approaches relying on IP flow records are attractive due to their low deployment overhead. However, they still suffer from several limitations: (i) detection latency caused by waiting for flow termination, (ii) susceptibility to evasion because transient behavioral changes within an ongoing flow are difficult to capture, and (iii) difficulty in identifying hosts located behind NAT. This study first enables early detection and improved sensitivity to short-lived changes by segmenting each flow into fixed time windows and performing feature extraction and classification on a per-window basis. Furthermore, in the doctoral phase, we target diverse cloud environments, including MEC and enterprise settings. We represent host behaviors, including both intra- and inter-segment communications, as communication graphs and learn their representations using graph neural networks (GNNs). This aims to improve robustness against environmental changes and to expand detection coverage. Finally, we design and evaluate a fast and environment-adaptive malware detection framework by integrating time-window features with graph-based representations. language of the presentation: Japanese 発表題目: トラフィックの時間窓分割と動的グラフ学習を統合したネットワーク環境適応型のマルウェア検知に関する研究 発表概要: ネットワークトラフィックに基づくマルウェア検知では，IPフロー情報を用いる方式が軽量である一方で，(i) フロー終了待ちに起因する検知遅延，(ii) フロー途中の短時間変化を捉えにくいことによる回避可能性，(iii) NAT配下ホストの識別困難といった課題が残る．本研究ではまず，フローを時間窓に分割し，窓ごとに特徴抽出・判定することで，早期検知と短時間変化の捕捉を実現する．さらに博士研究では，MECやエンタープライズ向けを含む多様なクラウド環境を想定し，セグメント内・間通信を含むホスト挙動を通信グラフとして表現してGNNにより学習することで，環境変化に対するロバスト性の向上と検知対象範囲の拡大を狙う．時間窓特徴とグラフ表現を統合した，高速かつ環境適応的なマルウェア検知手法の設計と評価を行う． | ||||||
| MUTABAZI PATRICK | D, 中間発表 | サイバーレジリエンス構成学 | 門林 雄基, | 笠原 正治, | 林 優一, | 妙中 雄三 |
| GAIDUCHENKO SOFIA | M, 2回目発表 | ソーシャル・コンピューティング | 荒牧 英治, | 松本 健一, | 若宮 翔子, | PENG SHAOWEN, | 久田 祥平 |
|
title: Modeling Native Speaker Intuition: Evaluating LLM Judgments on Idiomatic Morphological Transformations
abstract: This study investigates whether large language models (LLMs) exhibit behavior resembling native speaker intuition in the domain of idiomatic morphology. We compiled a dataset of 800 Japanese idioms with sentence-like structure and analyzed a subset of 112 items in the current stage of the project. The focus is on morphological transformations (e.g., nominalization) and whether such transformations preserve idiomatic meaning and grammatical acceptability. Using GPT and LLaMA models, we prompted the systems to evaluate transformed forms for semantic preservation and naturalness, and to generate contextualized examples. The outputs were analyzed to identify patterns of agreement, systematic divergences, and clustering behavior across transformation types. Preliminary results suggest that LLM judgments are sensitive to certain morpho-syntactic constraints, while also revealing systematic overgeneralization in borderline or non-productive transformations. These findings provide a foundation for modeling LLMs as approximations of native speaker intuition and open avenues for expanding the dataset to additional transformation types. language of the presentation: English | |||||||
| JOSEPH AYOBAMI JOSHUA | M, 1回目発表 | ソフトウェア工学 | 松本 健一, | 安本 慶一, | 嶋利 一真, | Fan Youmei | |
|
Title: How Does Context Impact AI-Generated Pull Request Descriptions?
Abstract: The use of Large Language Models (LLMs) in software development has become increasingly popular, providing assistance in code generation, reviews, and other open-source contribution activities. This increase in AI-assisted development has led to a rise in AI-generated pull requests, prompting concerns and pushback from some open-source maintainers. This study evaluates the detectability of AI-generated pull request descriptions. Using the JabRef repository as a case study, we analyse a dataset of 308 human-written and 5,544 AI-generated pull request descriptions. We apply existing AI-generated text detection tools to assess their accuracy and to identify textual features that correlate with AI-generated content in software projects. In addition, we examine how the amount of context provided to the LLM (such as code diffs or referenced issues) affects the quality, detectability, and similarity of AI-generated pull request descriptions compared to human-written ones. Finally, we aim to provide empirical evidence on the performance of AI-text detectors in software engineering use cases and contribute to discussions on authorship attribution and transparency in AI-assisted open-source development. language of the presentation: English | |||||||
| SETTEWONG TASHA | M, 2回目発表 | ソフトウェア工学 | 松本 健一, | 安本 慶一, | 嶋利 一真, | Fan Youmei | |
|
title: Human vs AI Comparing Competition Notebooks on Kaggle
abstract: Data scientists increasingly rely on computational notebooks for analysis and collaboration, as they uniquely integrate code with descriptive documentation. With the rise of generative AI (GenAI), understanding the differences between human and GenAI, particularly in competitive environments, has become essential. This research investigates the respective strengths of humans and GenAI within the context of notebook-based coding and documentation tasks. By first examining 25 distinct features in medal-winning Kaggle notebooks, we identified that gold medalists are largely characterized by their extensive and detailed documentation. Furthermore, a comparison between human and GenAI notebooks reveals a trade-off: while GenAI typically produces higher-quality code as measured by metrics like code smells and technical debt), with fewer technical issues, human-written notebooks demonstrate superior structural complexity and creative problem-solving. language of the presentation: *** English *** | |||||||
| JAISRI PONGCHAI | D, 中間発表 | ソフトウェア工学 | 松本 健一, | 安本 慶一, | 嶋利 一真, | Fan Youmei | |
|
title: How Software Repository are Mined: A Preliminary Systematic Literature Review
abstract: GitHub provides access to public software repositories, allows other users especially researchers to utilize real-world projects in empirical studies. However, numerous of available repositories presents challenges for researchers when selecting repositories for research purposes. This research proposes a systematic investigation into how software engineering studies construct datasets from GitHub repositories. Using a Systematic Literature Review (SLR) approach, I aim to examine research papers from leading software engineering venues to identify dataset sources, repository selection criteria, and the relationship between selection strategies and research objectives. The study will leverage ChatGPT to extract and classify research goals across publications. By exploring repository selection practices, this research seeks to provide structured guidance for dataset construction in empirical software engineering, such as activity level, topic relevance, and repository structure. The results can highlight the common characteristic of selected repositories and corresponding research aiming. Which can serve as guidelines for future researchers in constructing relevant dataset. Additionally, the findings will provide insight for other software developers on how to improve their repositories for the research purpose. language of the presentation: English | |||||||
| LIU HANZE | M, 2回目発表 | 自然言語処理学 | 渡辺 太郎, | Sakriani Sakti, | 上垣外 英剛, | 坂井 優介 |
|
title: Singable Lyrics Translation Using Large Language Models abstract: Lyrics translation must account for rhythm, rhyme, and singability in the translated lyrics. In this study, we focus on singability and investigate effective prompting methods for translating singable lyrics, including verification-guided and multi-round prompting, applied to large language models. We curate a multilingual lyrics translation dataset covering a total of six language directions across Chinese, Japanese, and English, and evaluate seven prompting strategies, with instruction complexity increasing incrementally. The results show that multi-prompt strategies improve singability-related aspects, such as rhythmic alignment and phonological naturalness, compared to naive translation. language of the presentation: English | ||||||
| RIZA SETIAWAN SOETEDJO | M, 2回目発表 | 自然言語処理学 | 渡辺 太郎, | Sakriani Sakti, | 上垣外 英剛, | 坂井 優介 |
|
title: Enhancing Factuality through Consensus and Consistency in Summarization using Minimum Bayes Risk (MBR) Decoding
abstract: Improving the quality of model-generated summaries, especially factuality, the accuracy of a summary with respect to its source content, remains a challenge. While reranking could select the optimal output from multiple generated candidates, it is limited to only using the source as guidance, resulting in unreliable summaries. To address this limitation, we propose ConSUM that reranks candidate summaries by considering two factors: consistency to the source document and consensus among the other candidates. Consensus is established using Minimum Bayes Risk (MBR) decoding over the set of generated summaries, while ensuring consistency by employing factuality-aware metrics that compare the summary against the source. Rigorous testing demonstrates that our system is competitive with existing methods, with human evaluations further confirming that its generated summaries are preferred over those from other systems. language of the presentation: English | ||||||
| ACO ELYANAH MARIE CARIAGA | M, 2回目発表 | ソーシャル・コンピューティング | 荒牧 英治, | Sakriani Sakti, | 若宮 翔子, | PENG SHAOWEN |
|
title: From Scientific Abstracts to Visual Summaries: Frameworks for Targeted
Biomedical Science Communication
abstract: Scientific illustrations are powerful tools for communicating academic content especially to a general audience. This research examines scientific illustrations as visual summaries of scholarly articles, particularly graphical abstracts (GAs) commonly featured in scientific journals. The study introduces a benchmark dataset comprised of paired textual and graphical abstracts from life and health science journals. Frameworks for automated generation of scientific article summaries that are tailored to specific target audiences are also proposed: four-panel data comics or yonkomas for interdisciplinary readers and graphical abstracts for authors. language of the presentation: English | ||||||
| SAIDAH ZAHROTUL JANNAH | D, 中間発表 | ソーシャル・コンピューティング | 荒牧 英治, | Sakriani Sakti, | 若宮 翔子, | PENG SHAOWEN |
|
title: Developing an Emergency Call Dialogue Dataset and System for Symptom-Based Telephone Triage Using LLMs
abstract: Emergency telephone triage is essential for maintaining patient safety while reducing unnecessary ambulance dispatches and emergency department visits. However, real-world medical emergency call datasets annotated with reliable triage labels remain limited, particularly in multilingual contexts. This study presents the development of a synthetic multilingual emergency call dialogue dataset for symptom-based telephone triage research. The dialogues simulate realistic dispatcher–caller interactions and are generated using LLM under symptom-based constraints. To ensure dataset quality, a two-stage validation framework was applied to evaluate the dialogue realism and labeling the triage level. To support multilinguality, culturally and contextually aware translation was performed to adapt dialogues across languages while preserving medical and situational realism. The dataset covers Japanese, English, German, French, Indonesian, and Filipino. Using the validated dataset, baseline systems for automated triage were developed with LLMs using general prompts without symptom-specific reasoning. The baseline systems achieved moderate predictive performance, while undertriage and overtriage rates still did not meet the acceptable thresholds. Future work will investigate symptom-based triage strategies and analyze the multilingual capability behavior. language of the presentation: English | ||||||