Patterns
πŸ•ΈοΈ

Graph-of-Thought(GoT)

Non-linear reasoning with thoughts as nodes and dependencies as edges

Complexity: highReasoning Techniques

🎯 30-Second Overview

Pattern: Non-linear reasoning network with nodes (thoughts) and edges (dependencies) enabling cross-path synthesis

Why: Supports complex interdependent problems through backtracking, merging, and conflict resolution

Key Insight: Thoughts as graph nodes β†’ Explicit dependencies β†’ Cross-path merging β†’ Coherent synthesis

⚑ Quick Implementation

1Graph Init:Create nodes (thoughts) & edges (dependencies)
2Expand:Generate successor thoughts from frontier
3Evaluate:Score nodes/subgraphs with evaluators
4Merge:Connect compatible paths, resolve conflicts
5Distill:Extract coherent solution from best subgraph
Example: Problem β†’ [Node A ← Node B β†’ Node C] β†’ Merge paths β†’ Synthesize solution

πŸ“‹ Do's & Don'ts

βœ…Use explicit dependency tracking between thoughts
βœ…Implement robust cycle detection & deduplication
βœ…Design evaluators for scoring subgraphs not just nodes
βœ…Cache intermediate results & maintain visited states
βœ…Enable cross-path synthesis & conflict resolution
❌Allow unlimited graph expansion (combinatorial explosion)
❌Merge incompatible branches without conflict checks
❌Skip thought granularity planning (too coarse/fine)
❌Use weak evaluators that mis-rank solution paths
❌Ignore token costs from graph exploration overhead

🚦 When to Use

Use When

  • β€’ Complex problems with interdependent sub-goals
  • β€’ Research synthesis from conflicting sources
  • β€’ Multi-constraint optimization problems
  • β€’ Strategic planning with trade-offs
  • β€’ When path merging creates synergies

Avoid When

  • β€’ Simple linear reasoning tasks
  • β€’ Strict latency/cost constraints
  • β€’ Domains without reliable evaluators
  • β€’ When Tree-of-Thought suffices
  • β€’ Real-time applications

πŸ“Š Key Metrics

Solution Quality
Accuracy vs linear/tree baselines
Graph Efficiency
Useful nodes / total explored nodes
Synthesis Quality
Coherence of merged solution paths
Dependency Accuracy
Correctness of thought relationships
Exploration Cost
Tokens per node + evaluation overhead
Convergence Rate
Steps to reach stable solution

πŸ’‘ Top Use Cases

Research Analysis: Node[Theory A] ← Node[Evidence 1] β†’ Node[Synthesis] ← Node[Theory B]
Strategic Planning: Connect market analysis β†’ product strategy β†’ resource allocation β†’ risk assessment
System Design: Link requirements β†’ architecture β†’ implementation β†’ validation with feedback loops
Policy Analysis: Map stakeholder interests β†’ regulatory constraints β†’ implementation options β†’ outcomes
Creative Problem-Solving: Merge diverse solution approaches through cross-pollination nodes

References & Further Reading

Deepen your understanding with these curated resources

Contribute to this collection

Know a great resource? Submit a pull request to add it.

Contribute

Patterns

closed

Loading...

Built by Kortexya