📍 Tokyo ｜ 👨‍🔬 Researcher ｜ 👨‍💻 Engineer ｜ 👨‍💼 Manager

Specializing in natural language processing (NLP), especially applied NLP, text generation, and evaluation.

• BannerBench: Benchmarking Vision Language Models for Multi-Ad Selection with Human Preferences [Paper] [Dataset]
  • Hiroto Otake, Peinan Zhang, Yusuke Sakai, Masato Mita, Hiroki Ouchi, and Taro Watanabe
  • In EMNLP 2025 Findings
• Distilling Many-Shot In-Context Learning into a Cheat Sheet [Paper]
  • Ukyo Honda, Soichiro Murakami, and Peinan Zhang
  • In EMNLP 2025 Findings
• AdParaphrase v2.0: Generating Attractive Ad Texts Using a Preference-Annotated Paraphrase Dataset [Paper] [Dataset]
  • Soichiro Murakami, Peinan Zhang, Hidetaka Kamigaito, Hiroya Takamura, and Manabu Okumura
  • In ACL 2025 Findings
• AdTEC: A Unified Benchmark for Evaluating Text Quality in Search Engine Advertising [Project Page] [Paper] [Dataset]
  • Peinan Zhang, Yusuke Sakai, Masato Mita, Hiroki Ouchi, and Taro Watanabe
  • In NAACL 2025

See all publications here.

Prototyping to production, from research experiments to user-facing services.

Programming Languages
Frameworks
Environment	(See my dotfiles for more)
Tools

All in Japanese

• Do Androids Dream of Blue Traffic Light? ― Do VLMs in different languages see colors differently?
• Google Cloud Batch Parallelization ― With LLM pre-processing as an example
• NLP Benchmark for Understanding and Generating Ads ― Presents two important benchmarks
• Effective Ad Text Generation ― Presents ad text generation methods in CyberAgent's "Kiwami" products
• Effective Joint Research ― Deep dives into a joint research project with the Tokyo Institute of Technology
• CyberAgent's AI Research and Business Implementation ― 5-min overview featuring ad creative AI

See more here.