GitHub workflow: generate people JSON + Astro MD files from BibTeX entries

[gokceuludogan]

📌 Summary

We currently have publication metadata under data/bib/*.bib (one file per PI).
We want to automatically infer people info from these BibTeX files and:

1. Export a JSON file with basic people info.
2. Generate Astro content files under src/content/people/*.md with a minimal frontmatter.

This should be implemented as a GitHub Actions workflow.

---

🎯 Goals

• Parse all data/bib/*.bib files.

• Extract unique people (authors) across all publications.

• Infer an advisor for each person (based on co-authorship frequency with PIs).

• Export:

  • A machine-readable JSON file with people info.
  • Astro-compatible markdown files in src/content/people/ with minimal frontmatter.

---

📂 Inputs

• BibTeX files:

  data/bib/*.bib
  

  Each file corresponds to a PI (e.g., arzucan-ozgur.bib, tunga-gungor.bib, suzan-uskudarli.bib) and contains multiple @article, @inproceedings, etc. entries with author fields.

• PI mapping (explicit or inferred):

  • Names / slugs of PIs (e.g., from data/googlescholar.json or filename):

    • arzucan-ozgur
    • tunga-gungor
    • suzan-uskudarli

---

🧠 Logic / Requirements

1. People extraction

• For each BibTeX entry in data/bib/*.bib:

  • Parse the author field.
  • Split into individual author names (e.g., "A. Özgür and T. Güngör and A. Köksal" → 3 people).

• Normalize names:

  • Preserve accents (e.g., Özgür, Köksal).
  • Use full names as they appear in BibTeX when possible.

• Build a global map:

  {
    "abdullatif-koksal": {
      "name": "Abdullatif Köksal",
      "advisor": "Arzucan Özgür" | null,
      "category": "student" | "alumni" | etc. (initially just one value),
      // other fields can be added later
    },
    ...
  }

• Slug generation:

  • Lowercase
  • Replace spaces with hyphens
  • Strip punctuation
  • E.g., "Abdullatif Köksal" → abdullatif-koksal.

2. Advisor inference

• A set of PIs is known (from filenames or config), e.g.:

  arzucan-ozgur
  tunga-gungor
  suzan-uskudarli
  

• For each non-PI person:

  • Look at all papers where they are a co-author.
  • Count co-occurrences with each PI.
  • Advisor = PI with highest co-authorship count.
  • If no PI appears with that person, leave advisor empty.

3. JSON export

• Write a JSON file, e.g.:

  data/people.generated.json
  

• JSON schema (minimal for now):

  {
    "abdullatif-koksal": {
      "name": "Abdullatif Köksal",
      "advisor": "Arzucan Özgür",
      "category": "student"
    },
    "tunga-gungor": {
      "name": "Tunga Güngör",
      "advisor": "",
      "category": "pi"
    }
  }

• Category:

  • For now, set only one categorical field:

    • PIs → "pi" (or "faculty")
    • Non-PIs → "student" by default (can be refined later).

4. Astro markdown generation

For each person (key = slug), create a file:

src/content/people/<slug>.md

• Example: src/content/people/abdullatif-koksal.md

• Frontmatter rules:

  • Fill only:

    • name
    • advisor (if inferred)
    • category

  • Leave the rest EMPTY/blank for now (title, photo, bio, email, order, degree, body content).

• Template format:

  ---
  name: "Abdullatif Köksal"
  title: ""
  photo: ""
  bio: ""
  email: ""
  category: "student"
  order: 
  advisor: "Arzucan Özgür"
  degree: ""
  ---
  

• If advisor is unknown:

  advisor: ""

• Do not auto-generate description text for now; keep the body empty.

• The example currently used in the site for reference (not to be fully filled now):

  ---
  name: "Abdullatif Köksal"
  title: "MS Student"
  photo: "/images/people/abdullatif-koksal.jpg"
  bio: "MS student at Boğaziçi University, working on natural language processing under the supervision of Arzucan Özgür."
  email: "abdullatif.koksal@boun.edu.tr"
  category: "student"
  order: 10
  advisor: "Arzucan Özgür"
  degree: "MS"
  ---
  
  Abdullatif Köksal is an MS student at the Computer Engineering Department of Boğaziçi University, working under the supervision of Arzucan Özgür.
  
  ## Research Interests
  
  - Natural Language Processing
  - Machine Learning
  - Cross-lingual NLP
  - Text Classification
  
  ## Advisor
  
  - **Advisor:** Arzucan Özgür

  👉 In this issue, we only want name, advisor, and category populated.

---

⚙️ GitHub Actions workflow

Implement a workflow, e.g. .github/workflows/generate-people-from-bib.yml:

• Trigger:

  • workflow_dispatch (manual for now; later optionally schedule).

• Steps:

  1. Checkout repo.

  2. Set up Python.

  3. Install dependencies (bibtexparser or similar).

  4. Run a Python script, e.g. scripts/generate_people_from_bib.py, that:

     • Parses data/bib/*.bib.

     • Builds the people map.

     • Writes:

       • data/people.generated.json
       • src/content/people/*.md

  5. Optionally:

     • Commit changes on a new branch and open a PR (similar style to existing Scholar workflows).

---

✅ Acceptance Criteria

• Running the workflow on existing data/bib/*.bib generates:

  • data/people.generated.json
  • A set of files src/content/people/<slug>.md for all discovered people.

• Each .md file has:

  • Correct name.
  • advisor populated with the most frequent co-author PI, if any.
  • category set appropriately (at least pi vs student).
  • Other fields exist but are empty ("" or omitted as agreed).

• Existing Astro site builds successfully using these generated people files.

• Workflow is documented in the repo (short note in README or CONTRIBUTING.md).