AI-Accelerated News Media

My research interests lie in the realization of AI-accelerated news media. This involves creating new experiences and improving operational efficiency in all stages of news production. In particular, we focus on exploring opportunities and challenges for pre-trained models in news media. Furthermore, I am involved in leveraging data science competitions as a means to promote the democratization of data science.

Building and Applying Information Technology Such as Pre-trained Models

There are many opportunities to apply information technology within the business processes of news media. As in-house engineers, we collaborate closely with editors. Some of our work is published externally in the form of press releases and research papers.

• Domain-specific pre-trained models Press release
• Improving operational efficiency:
  • News summarization Journal of NLP
  • Entity linking system Journal of NLP
• Creating new experiences:
  • Recommendation from daily scenery KES-2025
  • Crossword puzzle generation CIKM 2023
• Service & User analysis:
  • Effects of delivery formats IC2S2 2026
  • Semantic shift analysis IC2S2 2023
  • Reading time estimation BigData 2022 Industrial & Government Track
  • Revisiting news recommendation NLDB 2025

Quantifying memorization of LLMs

The memorization of training data by large language models (LLMs) is recognized as a significant challenge related to copyright, security, and sound evaluation. We are conducting empirical experiments to quantify memorization, primarily using Japanese newspaper articles. We have also published survey papers and released a library for quantification.

• Survey paper ACL 2023 Workshop Transactions of JSAI
• Benchmarking membership inference attacks ACL 2026 Demo
• Experiments on Japanese newspaper INLG 2024 ACL 2025 Workshop
• Monitoring time-series performance degradation AACL-IJCNLP 2022 Journal of NLP

Promoting data science competitions

Data science competitions, such as those on Kaggle, are a powerful tool for the social implementation and democratization of information technology. Through publishing books and organizing competitions, we are dedicated to promoting this approach.

• Books:
  • 高野海斗, 齋藤慎一朗, 石原祥太郎 (2026). Kaggleではじめる大規模言語モデル入門　自然言語処理〈実践〉プログラミング. 講談社.
  • 小嵜耕平, 秋葉拓哉, 林孝紀, 石原祥太郎 (2023). Kaggleに挑む深層学習プログラミングの極意. 講談社.
  • Abhishek Thakur (著), 石原祥太郎 (翻訳) (2021). Kaggle Grandmasterに学ぶ 機械学習 実践アプローチ. マイナビ出版.
  • 石原祥太郎, 村田秀樹 (2020). PythonではじめるKaggleスタートブック. 講談社. Chinese Korean
• Website:
  • Weekly Kaggle News
  • Everyday Kaggle News
• Hosting competitions:
  • Kaggle Days Tokyo 2019, Dec 2019.
  • JOAI2025, April 2025.
  • JAPAN AI CUP 2025 予測部門, Feb 2026.
• Results as a participant
• Committee:
  • 一般社団法人人工知能学会企画委員会（コンペティション担当）, April 2024 -.
  • 地球惑星科学における AI 研究会専門委員, April 2024 -.
  • 国際人工知能オリンピック（IOAI）日本委員会スタッフ, Dec 2024 -.

Reference

You can see Publications for more details.