Computational Social Scientist · NLP Analysis Enthusiast
M.S. student — Tokyo, Japan
This page is updated on 2025-04-25
I’m, above all, a sociologist who wields data science to put theory to the test. My agenda is two‑fold:
- Empirically ground sociological theory – turning abstract concepts of interaction, power, and meaning into variables we can observe and measure.
- Invent methodology – building open, reproducible pipelines that let social scientists ask bigger questions with richer evidence.
- Good argumentation: what makes disagreement productive, and how can we foster cooperative dialogue?
- LLM‑driven agent simulations: using large language models as cognitively rich agents to experiment with communication norms and network structures in silico.
To tackle these, I combine:
- Natural Language Processing (topic models, embeddings, LLMs)
- Agent‑Based Simulation with LLM agents (ABSwLLM)
- Sociological reconstruction of communication theory — formalising conversational norms and power dynamics into testable constructs
- Computational lenses on collective consciousness and meaning‑making
My goal is to bridge the creativity of sociological theory with the rigor of data‑intensive methods and to share the tools so others can do the same.
| Domain | Stack & Tools |
|---|---|
| Programming | Python (Pandas, NumPy, scikit‑learn, PyTorch/Transformers), R, Unix |
| NLP / ML | LDA & CTM, Word 2 Vec / fastText, BERT‑family, Whisper, LangChain, Hugging Face Hub |
| Data Engineering | SQL (SQLite, PostgreSQL), SQLAlchemy, Docker |
| Visualization | Matplotlib, Seaborn, Plotly, NetworkX |
| Cloud & DevOps | Google Colab, GCP (Vertex AI, Cloud Run), GitHub Actions, VS Code Remote‑SSH |
| Collaboration | Git/GitHub, Zotero, Notion, Markdown, LaTeX |
| Year | Project | Brief | Tech Highlights |
|---|---|---|---|
| 2025 | LLM‑powered Agent‑Based Communication Simulation | Simulates dialogues among moral‑profiled agents (Haidt MFT) with LLMs; explores how network structure shapes cooperative communication. | Python, LangChain, NetworkX, Matplotlib, Streamlit |
| 2025 | Whisper‑Driven Interview Transcription Pipeline | End‑to‑end Colab workflow: audio segmentation → Whisper → speaker diarization → structured JSON → Zotero integration. | Python, OpenAI API, PyAnnote, Pandas, Google Drive API |
| 2024 | note.com Discourse Analysis | Identified topic clusters & sentiment dynamics in 1.2 M posts on Japanese platform note; presented at JSS Spring 2024. | Gensim, spaCy‑ja, WordCloud, Matplotlib |
➡︎ Full list & code: see pinned repositories or the projects/ directory.(Work-in-progress)
I maintain a living GitHub Wiki with reproducible snippets:
- Python tricks for large‑scale text mining
- Jupyter & VS Code configurations (
nbstripout, Japanese fonts) - GPT prompt patterns for research & writing
- Google Workspace automations (Apps Script, Gmail API)
- Markdown / LaTeX style guides
- Zotero workflows for collaborative bibliography management
- Tobimatsu, Tomoki. 2024. “Applying Topic‑Model Analysis for the Quantitative Mapping of Discourse Space.” 77th Academic Conference on Japanese Association for Mathematical Sociology, Poster.
飛松大騎，2024，「言説空間の数量的把握に向けたトピックモデル分析の応用」，第77回日本数理社会学会大会，ポスター発表. - Tobimatsu, Tomoki. 2024. “Quantitative Analysis of Online Discourse on Food.”Oral, 97th Academic Conference on the Japan Sociological Society.
飛松大騎，2024，「食をめぐるオンライン上の言説空間の定量的分析」，第97回日本社会学会大会，口頭発表. - Tobimatsu, Tomoki. 2024. “Computational Mapping of Discourse Using À‑la‑Carte Word Embeddings: A Case of Japanese Blog Posts on Veganism.” Poster, 78th Academic Conference on Japanese Association for Mathematical Sociology.
飛松大騎，2024，「アラカルト単語埋め込みモデルによる言説空間の計算論的把握—ヴィーガンについての日本語ブログ記事を対象として—，第78回日本数理社会学会大会， ポスター発表. - Tobimatsu, Tomoki. 2025,05,25.“Social Worlds in Code: Building Virtual Societies with LLM Agents.”，Public Lecture, UTokyo Research Frontiers: 10-Minute Talk.
飛松大騎，2025,05,25，「社会を作って観る—大規模言語モデルと挑む社会学実験—」10分で伝えます！東大研究最前線・公開講演. - Tobimatsu, Tomoki. 2025. "The Dynamics of Immigrant Cultural Transformation through Food: A Study of Zainichi Korean Culture in Osaka Koreatown", "Super-Diversifying Communities and People's Practices: Fieldwork in Ikuno Ward, Osaka City"
飛松大騎, 2025, "食にみる移民文化変容のダイナミズムー大阪コリアタウンの在日コリアン文化を対象としてー," 超多様化する地域と人びとの実践ー大阪市生野区をフィールドとしてー，137-147.
- Graph Neural Networks for narrative networks
- Multi‑modal sentiment analysis (text + image)
- Responsible AI & data ethics frameworks