epstein-semantic-explorer

Semantic clusters, entity extraction, and BM25 search across recently released Congressional Epstein documents.

The full semantic dataset is hosted on Kaggle:

👉 https://www.kaggle.com/datasets/cjc0013/epstein-bge-large-hdbscan-bm25/data

Epstein Semantic Explorer v5 — README

A lightweight, semantic investigation toolkit for exploring the House Oversight Epstein text dump.

This notebook does not add any new allegations. It simply organizes, clusters, and indexes what is already public — thousands of text fragments released by Congress.

All analysis is done locally inside Colab, and no external services, models, or APIs are required.

---

📦 What This Notebook Does

The Epstein Semantic Explorer v5 takes a messy pile of thousands of text documents and gives you:

1. Cluster Browser

Explore themed groups of documents — legal strategy, PR coordination, iMessage logs, internal disputes, etc.

view_cluster(96)

---

2. Keyword Search (BM25-lite)

Instantly find related documents by keyword or phrase.

search("Prince Andrew")
search("Ghislaine")
search("Clinton")

---

3. Cluster Summaries

Get a quick human-readable overview of what a cluster likely contains.

summarize_cluster(96)

---

4. Topic Modeling

Shows “top terms” for each cluster — a lightweight, dependency-free way to understand cluster themes.

show_topics()

---

5. Entity Extraction

Find the most frequently mentioned people, places, and organizations inside a cluster.

cluster_entities(12)

---

6. Timeline Extraction

Automatically extract dates and sort them chronologically.

show_timeline()

---

7. Cluster Similarity Matrix

Shows how thematically related different clusters are.

cluster_similarity()

---

8. Cross-cluster Entity Search

Instantly see which clusters mention a specific name the most.

entity_to_clusters("Epstein")
entity_to_clusters("Maxwell")
entity_to_clusters("Barak")

---

📁 What You Need

Only one file:

epstein_semantic.jsonl

Your dataset should be structured like:

{"id": "HOUSE_OVERSIGHT_023051", "cluster": 96, "text": "…document text…"}
{"id": "HOUSE_OVERSIGHT_028614", "cluster": 122, "text": "…document text…"}

That’s it. No images, no PDFs needed.

---

🚀 How to Use the Notebook

Step 1 — Upload the Notebook

Open Google Colab → upload:

epstein_explorer_v5.ipynb

Step 2 — Run All Cells

Runtime → Run all

Step 3 — Upload Your Data

When prompted:

Upload epstein_semantic.jsonl

Step 4 — Start Exploring

Use the tools:

view_cluster(96)
search("Prince Andrew")
show_topics()
cluster_entities(96)

Everything runs instantly on CPU; no GPU required.

---

🧩 FAQ

Q: Does this notebook create new allegations?

No. It only organizes existing public documents released by the House Oversight Committee.

Q: Does this send data to external servers?

No. Everything runs inside your Colab session.

Q: Can reporters use this without technical experience?

Yes — the notebook is designed to be point-and-click simple. You can search, browse clusters, and explore without writing code.

Q: Is this safe for publication?

Yes — as long as you make clear that:

• all data is public
• no new claims were added
• this is a semantic reorganization of public records

---

📝 Summary

This notebook:

• converts the chaotic House Oversight Epstein text archive into a structured, searchable dataset
• enables investigators to quickly identify themes, contradictions, legal strategies, and PR coordination patterns
• gives reporters the ability to explore thousands of documents in minutes

This tool makes the archive usable, but does not alter or create any content.