After running the community detection, you now have multiple sets of community member nodes. Each set represents a semantic topic within your knowledge graph. The community reporting step abstracts across concepts from different documents in your knowledge base. I integrated this into the Microsoft implementation, adding a function call for structured output parsing.

    
      You are an AI assistant that aids a human analyst in general information discovery. The aim is to identify and evaluate relevant information related to entities (e.g., organizations, individuals) in a network.
    
  

The community report generation highlights the challenge of knowledge graph retrieval. A document could potentially impact many existing communities. I addressed this issue by detecting changed communities and regenerating reports only for them.