Duplicate Detection

Automatically identify duplicate issues

Duplicate Detection

Simili Bot uses AI to automatically identify when a new issue is a duplicate of an existing one.

How it works

1. Find Similar Issues - Use semantic search to find related issues
2. Analyze with AI - Use Gemini LLM to determine if truly duplicates
3. Calculate Confidence - Return confidence score (0.0-1.0)
4. Post Comment - Notify about duplicates with reasoning

Example

New Issue: “Can’t log in with email”

Simili Bot finds similar issues and analyzes:

• #456: “Email login broken” (94% similarity, 87% duplicate confidence)
• #123: “Authentication issues” (88% similarity, 76% duplicate confidence)

Posts comment:

🔍 Duplicate Analysis

This issue appears to be a duplicate of #456:

Duplicate Confidence: 87%
Analysis: Both issues describe the same authentication timeout
problem in the login flow with identical error messages.

Suggested Action: Close as duplicate and reference #456

Automatic Closure

Simili Bot can take duplicate detection a step further by automatically managing the closure of confirmed duplicates.

How to Enable

In your simili.yaml, add the auto_close section:

auto_close:
  grace_period_hours: 48 # Allow 48 hours for human review

The Triage Flow

1. Detection: Bot identifies a duplicate.
2. Labeling: Bot adds the duplicate label.
3. Grace Period: The issue remains open for review.
4. Auto-Close: A scheduled worker (run via simili auto-close) closes the issue if the grace period has passed.

For more details on setting up the scheduled worker, see the Auto-Close Configuration.

Configuration

Control duplicate detection via workflow selection:

# In workflow
- uses: similigh/simili-bot@v0.2.0
  with:
    workflow: "issue-triage"  # Includes duplicate detection
    # or
    workflow: "similarity-only"  # Skip duplicate detection

Confidence scores

• 90-100%: Very likely duplicate
• 75-90%: Probably duplicate
• 50-75%: Possibly related
• <50%: Not a duplicate

Benefits

✅ Prevent scattered discussions ✅ Consolidate related issues ✅ Reduce duplicate work ✅ Improve issue tracking

Next steps

Semantic search

How semantic search powers duplicate detection