// LLM Researcher · Tokyo

Chengguang Gan

LLM Researcher @ Techtouch · Ph.D. in Informatics

I build and study large language models for information extraction, web agents, and the Mutual Reinforcement Effect — and I ship the datasets and models to back it up.

View Publications Get in touch
Chengguang Gan
Tokyo, Japan
// 01

About

I’m an LLM Researcher at Techtouch in Tokyo. I earned my Ph.D. in Informatics in March 2025 from Yokohama National University, advised by Prof. Tatsunori Mori. My work spans natural language processing and large language models — with a focus on information extraction, prompting, multimodal IE, and web agents.

I introduced the Mutual Reinforcement Effect and built the Japanese (and later multilingual) IE Mix datasets — covering sentence, text, sentiment and POS classification, relation and event extraction — then fine-tuned a line of open LLMs on top of them. Everything ships: you can grab the papers, code, datasets, and models on Hugging Face and GitHub.

  • Large Language Models
  • Information Extraction
  • Prompting
  • Mutual Reinforcement Effect
  • Multimodal IE
  • Web Agents
  • Japanese NLP

Reviewer for NeurIPS, COLING, and COLM.

// 02

News

// 03

Publications

Bold = me. Citation counts via Google Scholar (Jul 2026).

  1. 2026

    A Learning-Rate-Gated Failure of GRPO in a Small Language and Vision-Language Model Web Agent: A Controlled Null and Its Mechanism

    Chengguang Gan, Zhixi Cai, Yunhao Liang, Hanjun Wei, Shiwen Ni, Qinghao Zhang

    Preprint arXiv
  2. 2026

    MAG: A Web-Agent Benchmark and Harness for Multimodal Action and Guide Generation

    Chengguang Gan, Hanjun Wei, Yunhao Liang, Zhixi Cai, Qinghao Zhang, Shiwen Ni

    Preprint arXiv
  3. 2026

    A Multilingual Dataset and Empirical Validation for the Mutual Reinforcement Effect in Information Extraction

    Chengguang Gan, Sunbowen Lee, Qingyu Yin, Yunhao Liang, Xinyang He, Hanjun Wei, Younghun Lim, Shijian Wang, Hexiang Huang, QingHao Zhang, Shiwen Ni, Tatsunori Mori

    ACL 2026 Findings ACL Anthology
  4. 2026

    GuideWeb: A Benchmark for Automatic In-App Guide Generation on Real-World Web UIs

    Chengguang Gan, Yoshihiro Tsujii, Yunhao Liang, Tatsunori Mori, Shiwen Ni, Hiroki Itoh

    Preprint arXiv
  5. 2025

    GIELLM: Japanese General Information Extraction Large Language Model Utilizing Mutual Reinforcement Effects

    Chengguang Gan, Qinghao Zhang, Tatsunori Mori

    IJCNN 2025cited by 16 IEEE
  6. 2025

    USA Model: Japanese Universal Sentiment Analysis Model & Construction of Japanese Sentiment Text Classification and Part-of-Speech Dataset

    Chengguang Gan, Qinghao Zhang, Tatsunori Mori

    PACLIC 2025 ACL Anthology
  7. 2025

    RECODE: Leveraging Reliable Self-generated Tests and Fine-Grained Execution Feedback to Enhance LLM-Based Code Generation

    Yunhao Liang, Ruixuan Ying, Takuya Taniguchi, Chengguang Gan, Zhe Cui

    ICIC 2025 Springer
  8. 2025

    Exploring Behavior-Driven Development for Code Generation

    Yunhao Liang, Chengguang Gan, Ruixuan Ying, Zhe Cui

    ICIC 2025 Springer
  9. 2025

    Retrieval and Distill: A Temporal Data Shift-Free Paradigm for Online Recommendation System

    Lei Zheng, Ning Li, Chengguang Gan, Yong Yu, Weinan Zhang

    APWeb-WAIM 2025 Springer
  10. 2025

    M-MRE: Extending the Mutual Reinforcement Effect to Multimodal Information Extraction

    Chengguang Gan, Zhixi Cai, Yanbin Wei, Yunhao Liang, Shiwen Ni, Tatsunori Mori

    Preprint arXiv
  11. 2025

    Decoding Prokaryotic Whole Genomes with a Product-Contextualized Large Language Model

    Shiwen Ni, Sheng Li, Shijian Wang, Xin Bi, Yang Li, Chengguang Gan, et al.

    bioRxiv bioRxiv
  12. 2024

    Application of LLM Agents in Recruitment: A Novel Framework for Resume Screening

    Chengguang Gan, Qinghao Zhang, Tatsunori Mori

    Journal of Information Processingcited by 151 J-STAGE
  13. 2024

    II-Bench: An Image Implication Understanding Benchmark for Multimodal Large Language Models

    Ziqiang Liu, Feiteng Fang, Xi Feng, Xinrun Du, Chenhao Zhang, Zekun Wang, Yuelin Bai, Qixuan Zhao, Liyang Fan, Chengguang Gan, et al.

    NeurIPS 2024cited by 21 Proceedings
  14. 2024

    Think from Words (TFW): Initiating Human-Like Cognition in Large Language Models Through Think from Words for Japanese Text-level Classification

    Chengguang Gan, Qinghao Zhang, Tatsunori Mori

    NLDB 2024 Springer
  15. 2024

    Empirical Study of Mutual Reinforcement Effect and Application in Few-shot Text Classification Tasks via Prompt

    Chengguang Gan, Tatsunori Mori

    Preprint arXiv
  16. 2024

    Demonstrating Mutual Reinforcement Effect through Information Flow

    Chengguang Gan, Xuzheng He, Qinghao Zhang, Tatsunori Mori

    Preprint arXiv
  17. 2023

    Sensitivity and Robustness of Large Language Models to Prompt Template in Japanese Text Classification Tasks

    Chengguang Gan, Tatsunori Mori

    PACLIC 37cited by 48 ACL Anthology
  18. 2023

    A Few-Shot Approach to Resume Information Extraction via Prompts

    Chengguang Gan, Tatsunori Mori

    NLDB 2023cited by 16 Springer
  19. 2023

    Sentence-to-Label Generation Framework for Multi-task Learning of Japanese Sentence Classification and Named Entity Recognition

    Chengguang Gan, Qinghao Zhang, Tatsunori Mori

    NLDB 2023cited by 12 Springer
  20. 2023

    Mutual Reinforcement Effects in Japanese Sentence Classification and Named Entity Recognition Tasks

    Chengguang Gan, Qinghao Zhang, Tatsunori Mori

    Preprint arXiv
  21. 2023

    Construction of English Resume Corpus and Test with Pre-trained Language Models

    Chengguang Gan, Tatsunori Mori

    ANLP 2023 arXiv
  22. 2021

    英文履歴書データ抽出システムへの BERT 適用性の検討

    Chengguang Gan, Yoshihide Takahashi

    IPSJ Kansai 2021 IPSJ
// 04

Experience & Education

Experience

  • 2025 — Present
    LLM ResearcherTechtouch
  • 2025
    Data ScientistGeneric Solution
  • 2024 — 2025
    Part-time ResearcherNII Large Language Model Center
  • 2024
    Research Part-timerRIKEN AIP
  • 2022 — 2023
    Research AssistantYokohama National University

Education

  • 2022 — 2025
    Ph.D. in InformaticsGraduate School of Environment and Information Sciences, Yokohama National University
  • 2020 — 2022
    M.S. in Information TechnologyThe Kyoto College of Graduate Studies for Informatics
  • 2014 — 2018
    East China Jiaotong UniversityNanchang, China
// 05

Open Source

Datasets, models, and code behind the papers — free to use.

Dataset 139 · 15

resume_seven_class

~78k-entry resume sentence-classification dataset across seven section types. My most-downloaded release.
Dataset 58

MMM Multilingual Test Set

Trilingual (EN / JA / ZH) test set for the Mutual Reinforcement Effect IE benchmark — ACL 2026 Findings.
Model7B · JA/EN/ZH

Yoko-7B-Japanese

LLaMA2-based Japanese LLM. The most-liked model on my Hugging Face profile.
Model8B · Instruct

OIELLM-8B

Open-domain information-extraction LLM (LLaMA3-8B) for trilingual NER & IE — the model behind MMM.
Model7B

GIELLM-7B

General Information Extraction LLM behind the GIELLM / Mutual Reinforcement Effect paper (IJCNN 2025).
CodeHub repo

Mutual Reinforcement Effect

The home for my flagship research line — USA, GIELLM, OIELLM and the MMM mix datasets, all linked in one place.

More on Hugging Face and GitHub.

// 06

Beyond Research

When I’m away from the terminal.

Gaming

Mainly a AAA gamer — big-budget, single-player worlds are how I unwind, and honestly the same curiosity that got me into taking models apart.

Steam profile

Outdoors

I ski all through winter; once the snow melts I’m out climbing mountains and on long hikes. Fresh air, big views.

Photography

Landscape photography and urban street shooting — the header shot is one of mine, somewhere up in the hills.

500px gallery
// 07

Get in touch

Open to research collaborations and roles in LLMs, NLP, and agents. Best reached by email — or find me on the platforms below.

ganchengguan@yahoo.co.jp