Research
Our laboratory aims to understand phenomena in business, society, and science & technology using data-science methodologies, generating new insights across these domains. We focus on machine learning, deep learning, natural language processing, and network analysis, while also pursuing foundational research on these methods.
Our analyses cover diverse datasets including large-scale scholarly corpora, social media, corporate transaction and M&A information, patent data, urban mobility datasets, and review texts. We are also among the first to analyze large language models (LLMs) via inverse learning and to utilize LLMs in the analysis of scientific and social systems.
Students are encouraged to choose their research topics autonomously, leading to a rich variety of themes. Active collaboration across domains is a defining characteristic of our research environment.
We also collaborate with external organizations such as the Technology Informatics Social Collaboration Program and the NEDO/AIST Science-Technology Trend Forecasting Project, emphasizing societal impact and engagement with broader research communities.
Science of Science
We analyze how scientific knowledge and technological innovation emerge and diffuse through papers, patents, and corporate activities. Social factors such as diversity and gender are also key themes in this rapidly growing field.
Generative AI and Society
We quantitatively analyze how the spread of generative AI is reshaping scientific research and technology development, using large-scale paper and patent data. We also run social simulations with LLM agents to experimentally examine collective behaviors such as conformity and cooperation, revealing how AI may influence collective decision-making and norm formation.
Recent Research Highlights
- Normative conformity ( arXiv 2026 ) and cooperation breakdown under communication delays ( arXiv 2026 ) in groups of LLM agents
- Analyzing how expanding use of generative AI reshapes research topics and technology development
- Developing social-simulation methods with LLM agents
AI-driven Technology Discovery
We develop methods and systems that embed generative AI into large-scale analysis of papers and patents to support the discovery of promising technologies and collaboration partners. Our work includes LLM-based mapping and summarization of the literature, linking papers to patents with LLM annotation, and quantifying the advancement from basic to applied research, pursued jointly with industry through the Technology Informatics Social Collaboration Program.
Recent Research Highlights
- Developing and providing Science Planet+ , an LLM-powered map of scientific literature
- Partner firm selection via patent-embedding-derived technology graphs and synergy ( Complex Networks 2025 )
- Linking papers to patents using LLM annotation (to appear at METSTI 2026)
- Quantifying basic-to-applied research advancement and the role of basic scientists during applied shifts ( Technovation 2024 , arXiv 2025 )
Computational Social Science
We analyze human behavior and online interactions using large-scale data. Our work includes Google Maps and Twitter (X) analysis, LLM-based social data analytics, and studies on techno-economic interactions using firm networks and patent information.
Recent Research Highlights
- Tracking structural changes in political communities on Twitter
- Understanding urban interaction patterns using Google Maps data
- Analyzing scientists’ social media use and research impact
- Distance characteristics in regional-bank-centered transaction networks
- Evaluating corporate portfolio strategies using patent-space models
- Commentary on mental health as both outcome and determinant in climate adaptation ( Nature Climate Change 2026 )
Foundations of Generative AI in Natural Language Processing
We develop advanced methods such as inverse learning, model composition, and training-data distillation to understand how language models acquire generalization and reasoning abilities. We also pursue improved transparency, safety, and efficiency, while exploring new frameworks that use LLMs as autonomous agents.
Recent Research Highlights
- Analyzing training data of large language models via inverse learning ( ACL 2024 )
- Improving language model safety through inverse distillation of corpora ( ICLR 2025 )
- Discussions on the legal validity of guardrails and related mechanisms
- Research on efficient communication methods among agents
Research Highlights
Science Planet+: LLM-powered Map of Scientific Literature
An LLM-powered system that maps hundreds of millions of papers for interactive exploration
Technology Informatics Social Collaboration Program
Industry-collaboration program on AI-driven technology discovery (Apr 2020 – Mar 2029)
Large Language Models Exhibit Normative Conformity
Experimental evidence that groups of LLM agents exhibit human-like normative conformity
Quantifying progress in research topics across nations
Quantitative time-series comparison of research topics across nations
Scientific Attention to Sustainability and SDGs: Meta-Analysis of Academic Papers
Analyzing how sustainability literature addresses the SDGs
Large-scale analysis of delayed recognition using sleeping beauty and the prince
長い間注目されなかった論文であるSleeping Beautyとそのきっかけとなる論文であるPrinceを大規模に取得し、その特徴を明らかにしました。非連続な発見は遠い分野間だけではなく、近くの分野の見落としにも多く存在していることがわかりました。
Evaluating Nodes of Latent Mediators in Heterogeneous Communities
異質なコミュニティをつなぐ結節点(ノード)を検出する新しいネットワーク指標(PW)を提案
Dense and influential core promotion of daily viral information spread in political echo chambers
Twitterの政治的エコーチャンバーにおいて効率的な情報拡散が行われている過程を明らかにしました。
Unsupervised Abstractive Opinion Summarization by Generating Sentences with Tree-Structured Topic Guidance
木構造上のトピックから各トピックに関する要約文を生成する教師なし要約手法を提案