Need clarification on three RAG evaluation metric

[vr25]

Hi:

Really interesting work on RAG evaluation metrics. However, I am unable to understand how these evaluation functions work. Could you please point me to some relevant documentation that explains the mathematical formula for all three:

1. Context Relevance
2. Groundedness
3. Answer Relevance

Would really appreciate your help...thanks much in advance.

[joshreini1]

Hi @vr25 - these are model-based metrics.

1. Context relevance can be measured with smaller BERT-style models, embedding distances, or with LLMs.
2. Groundedness can be measured using smaller NLI models or LLMs
3. Answer relevance can BERT-style models, embedding distances, or with LLMs.

TruLens gives you the flexibility to use these different options for metric generation.

Does that help?

[vr25]

@joshreini1 This helps...thanks for the clarification.