Affiliation

Master's Program (2nd Year), Computer Science, Graduate School of System Design,
Tokyo Metropolitan University, Takama Laboratory

Research Interests

• Agentic LLM
• Mechanistic Interpretability
• LLM for Recommendation

International Conference

• LIT-RAGBench: Benchmarking Generator Capabilities of Large Language Models in Retrieval-Augmented Generation LREC 2026
  Koki Itai, Shunichi Hasegawa, Yuta Yamamoto, Gouki Minegishi, Masaki Otsuki
  paper
• Proposal of Modeling Personal Values Using Large Language Models for Extracting Mentions of Item Attributes and Evaluation Polarity from Review Texts TAAI 2024
  Koki Itai, Hiroki Shibata, Yasufumi Takama
  paper poster

Domestic Conference / Symposium

• Automatic Synthesis of Japanese Multi-hop QA Dataset Based on Knowledge Graphs for Agentic RAG The 32nd Annual Meeting of the Association for Natural Language Processing (NLP2026)
  Koki Itai (TMU / SB Intuitions), Koji Matsuda, Shigehiko Fukuchi (SB Intuitions)
  paper poster
• neoAI-InstructBench: A Japanese Composite Instruction-Following Benchmark Based on Practical Scenarios The 32nd Annual Meeting of the Association for Natural Language Processing (NLP2026)
  Toshiki Kawamoto (neoAI), Koki Itai (neoAI / TMU), Masaki Otsuki (neoAI / UTokyo)
  paper poster
• Construction of Generator Evaluation Benchmark for Japanese RAG The 265th Natural Language Processing Research Meeting
  Koki Itai, Shunichi Hasegawa, Yuta Yamamoto, Gouki Minegishi, Masaki Otsuki
  paper slide
• Training Japanese Agentic RAG Model Based on Chain of Abstraction The 20th Natural Language Processing Symposium for Young Researchers
  Koki Itai, Masaya Ohagi, Akihiko Fukuchi
  poster movie
• Proposal of Value Modeling from Review Text Using Large Language Models The 38th Annual Conference of the Japanese Society for Artificial Intelligence
  Koki Itai, Hiroki Shibata, Yasufumi Takama
  paper
  

Talks

Media