Since community summaries are generated by LLMs summarizing thousands of nodes and edges, they are highly prone to hallucinating connections that don't exist in the raw text.
How do you automate the evaluation of community summary faithfulness?
AI Engineer Problems › Vector Databases & RAG
Hard · AI Engineering · Vector Databases & RAG · Asked at Scale AI, Microsoft
Since community summaries are generated by LLMs summarizing thousands of nodes and edges, they are highly prone to hallucinating connections that don't exist in the raw text.
How do you automate the evaluation of community summary faithfulness?