学生のWIANGNAK, Patiponさん（博士後期課程2年、人間情報学研究領域、Natthawut研究室）が、The 24th International Semantic Web Conference（ISWC 2025）において、Best Poster Awardを受賞しました。

　ISWCは、セマンティックウェブ（Semantic Web）およびリンクドデータ（Linked Data）コミュニティにおける主要な国際会議です。
　ISWC 2025は、令和７年11月2日～6日まで、奈良県コンベンションセンターにて開催され、研究者や実務者、産業の専門家が一堂に会し、セマンティック技術の未来について活発な議論が交わされました。

※参考：ISWC 2025

■受賞年月日
　令和7年11月6日

■研究題目、論文タイトル等
Unveiling the Butterfly Effect in Knowledge Editing for Large Language Models Using Knowledge Graph-based Analysis.

■研究者、著者
Patipon Wiangnak, Natthawut Kertkeidkachorn, Kiyoaki Shirai

■受賞対象となった研究の内容
"Unveiling the Butterfly Effect in Knowledge Editing for Large Language Models Using Knowledge Graph-based Analysis" explores how small factual edits in Large Language Models (LLMs) can trigger unexpected ripple effects across related knowledge. We introduce ButterflyKE, a framework that leverages knowledge graphs to trace and evaluate these side effects. Our findings show that while LLMs can successfully update target knowledge, neighboring facts often remain affected, indicating that deeper, connected nodes may also change. This phenomenon helps explain how LLM hallucinations emerge, as models struggle to adapt to real-world updates. By extending the existing Knowledge Editing dataset CounterFact into CounterFact-Probed and proposing the Butterfly Index, we quantify local side effects and confirm the presence of the Butterfly Effect in Knowledge Editing. These insights highlight the need for approaches that maintain both factual accuracy and semantic stability, paving the way for future research on multi-hop and model-wide effects.

■受賞にあたって一言
ISWC 2025 is one of the leading international conferences in the field of Semantic Web and Knowledge Graph research. I could not have accomplished this journey without the kind support and guidance of Associate Professor Natthawut Kertkeidkachorn and Professor Kiyoaki Shirai over the past year. We worked very hard on this research together, and it was truly rewarding to see our efforts recognized with the Best Poster Award for "Unveiling the Butterfly Effect in Knowledge Editing for Large Language Models Using Knowledge Graph-based Analysis."
During the conference, I had the chance to meet many outstanding and insightful researchers in this field. We exchanged ideas and engaged in meaningful discussions, which broadened my perspective and inspired me with new directions for future research.
Thank you, ISWC 2025, for this wonderful opportunity, and to the entire community for your warmth, inspiration, and support. We will continue to work hard to contribute to this amazing research community. What began here as a small and simple idea we built together has now been remembered and recognized.
This is my Butterfly Effect.

　

　

令和8年1月21日