Gmail Entity Extraction

Create people entries from contacts data using confidence-based filtering.

Confidence-Based Filtering

Always Create (High Confidence)

• confidence: 'high' — 6+ interactions over 14+ days
• Strong sustained relationships with clear signal

Usually Create (Medium Confidence)

• confidence: 'medium' AND totalInteractions >= 4
• Regular contacts with sufficient history

Skip (Low Confidence)

• confidence: 'low' — Unless exceptional signalScore
• Insufficient data to assess relationship

Quality Signals

Strong signals (create entity):

• bidirectional: true — Mutual communication
• spanDays > 30 — Sustained relationship
• avgMessagesPerThread > 2 — Multi-turn conversations
• uniqueThreads > 1 — Multiple separate conversations

Red flags (skip or flag):

• spanDays < 7 with totalInteractions <= 4 — Recent burst, likely one-off
• Only 1 thread — Single exchange
• breakdown.timeSpan is negative — Burst penalty applied

Entity Description Templates

Include confidence context in the relationship description:

High confidence:

"Frequent collaborator (sustained 90-day relationship)"

Medium confidence:

"Regular contact (multiple conversations)"

Low confidence (if included due to exceptional signal):

"Recent contact (limited interaction history)"

What to Include

• Full name
• Email address
• Relationship context (what you know about them)
• Confidence level context (see templates above)
• Source: Gmail contact

What to Skip

• Newsletters and automated senders
• Low-confidence one-off contacts
• Contacts with only scheduling/logistics exchanges