enron / README.md
SuccessfulCrab's picture
Create README.md
d996d63 verified
|
Raw
History Blame Contribute Delete
4.2 kB

Enron Email Dataset — Entity Extraction

This dataset contains the results of running the Entity Extractor service over the complete Enron email corpus. The Entity Extractor is a Python-based text analysis tool that identifies and extracts structured data from unstructured text using a combination of regex pattern matching and deep learning NER models.

Dataset Overview

Property Value
Source Enron Email Corpus
Processing Entity Extractor (Azure ML Endpoint)
Split train

Schema

Column Type Description
Unnamed: 0 int64 Row index
file string Original file path (e.g., allen-p/_sent_mail/1.)
message string Raw email content including Message-ID and headers
extracted_entities list Array of extracted entity objects

Extracted Entity Format

Each item in the extracted_entities list contains:

{
  "text": "extracted value",
  "label": "entity type",
  "confidence": 0.95,
  "start_pos": 0,
  "end_pos": 10,
  "source_column": "message"
}

Supported Entity Types

The Entity Extractor identifies the following entity categories:

Type Code Description
Email Addresses email Standard email patterns including subdomains
Phone Numbers phone Australian and international formats
BSB Numbers bsb Australian Bank-State-Branch identifiers
ABN abn Australian Business Numbers (with checksum validation)
Account Numbers acct Bank account numbers with contextual matching
Dates date Multiple formats: numeric, written, partial
Payment Methods payment Cards, digital wallets, BNPL, transfers
Cryptocurrency crypto 8 blockchain types with cryptographic validation
Named Entities ner People, organisations, locations (Flair NER model)

Cryptocurrency Detection

Addresses from 8 major blockchains are detected with cryptographic validation:

  • Bitcoin (BTC) — Base58Check / Bech32 checksum
  • Ethereum (ETH) — EIP-55 mixed-case checksum
  • Litecoin (LTC) — Base58Check / Bech32 checksum
  • Monero (XMR) — Length and format validation
  • Cardano (ADA) — Bech32 checksum
  • Solana (SOL) — Base58 format validation
  • Tron (TRX) — Base58Check checksum
  • XRP (Ripple) — Base58Check checksum

Named Entity Labels (NER)

Deep learning-based extraction using Flair:

Label Description
PERSON Individual names
ORG Organisations and companies
GPE Geopolitical entities (cities, countries)
LOC Locations
DATE Temporal expressions

Use Cases

  • Compliance & Fraud Detection — Identify financial identifiers and suspicious patterns
  • Network Analysis — Map communication patterns via extracted emails and contacts
  • Document Analysis — Parse contact information, dates, and named entities
  • Cryptocurrency Investigation — Detect wallet addresses in historical communications

Accuracy Considerations

Potential False Positives

  • Solana wallets: Base58 patterns may match other alphanumeric strings
  • Account numbers: May match numeric sequences if contextual keywords are present
  • Phone numbers: International format variations may capture unintended patterns
  • Dates: Ambiguous formats (e.g., 01/02/03) cannot reliably distinguish day/month/year

Potential False Negatives

  • Non-standard separators or spacing may cause valid entities to be missed
  • Context-dependent entities require surrounding text that may not always be present
  • Truncated data will not match expected patterns

Technical Details

  • Platform: Azure Machine Learning (real-time inference endpoint)
  • Compute: NVIDIA T4 GPU
  • NER Model: Flair-based Named Entity Recognition

License & Attribution

The Enron email dataset is publicly available and was released by the Federal Energy Regulatory Commission during its investigation of the Enron Corporation.

Entity extraction performed using the Entity Extractor service.