Skip to content

Graph generation quality benchmark #5

Description

@MaximilianRau04

Description

The knowledge graph is only useful if the relationships it surfaces between documents, parties, and clauses are actually correct.

Proposed Solution

Build a benchmark comparing generated graph structure (nodes/edges) against a labeled ground-truth graph for a set of sample documents.

Acceptance Criteria

  • Ground-truth graph exists for a set of sample documents
  • Benchmark compares generated nodes/edges against ground truth
  • Benchmark reports precision/recall on relationships, not just node presence
  • Results are reproducible and comparable across runs

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions