> ## Documentation Index > Fetch the complete documentation index at: https://docs.galileo.ai/llms.txt > Use this file to discover all available pages before exploring further. # Chunk Relevance (Deprecated) > Understand how to measure and optimize the relevance of retrieved chunks to user queries in your RAG pipeline (deprecated - use Chunk Relevance instead) export const DefinitionCard = ({children}) => { return
{children}
; }; export const Scale = ({low, mid, high, lowLabel = "Low", midLabel = "Mid", highLabel = "High", lowDescription, midDescription, highDescription, midColor = "yellow", inverted = false}) => { const lowColor = inverted ? "green" : "red"; const highColor = inverted ? "red" : "green"; const gradientId = inverted ? "greenToRed" : "redToGreen"; return

{low}

{mid &&

{mid}

}

{high}

{lowLabel}

{lowDescription &&

{lowDescription}

}
{mid &&

{midLabel}

{midDescription &&

{midDescription}

}
}

{highLabel}

{highDescription &&

{highDescription}

}
; }; Chunk Relevance measures the proportion of text in each retrieved chunk that contains useful information to address the user's query in a RAG pipeline. Chunk Relevance is a continuous metric ranging from 0 to 1: A chunk with low relevance contains "unnecessary" text that is not pertinent to the user's query, indicating potential inefficiencies in your retrieval strategy. ## Calculation method Chunk Relevance is computed using a fine-tuned in-house Galileo evaluation model: A transformer-based encoder model trained to identify relevant information in the provided query, context, and response. The same model computes multiple metrics (Chunk Adherence, Chunk Completeness, Chunk Attribution, and Utilization) in a single inference call for efficiency. All metrics are computed simultaneously in a single inference call, optimizing performance and resource usage. The model processes each token in the context and outputs a relevance probability - the likelihood that the token is useful for answering the query. The model is trained on carefully curated RAG datasets and optimized to align closely with established RAG Plus metrics for accurate evaluation. ## Explainability The model identifies which parts of the chunks were relevant to the query: * These sections can be highlighted in your retriever nodes by clicking on the icon next to the Chunk Utilization metric value in your Retriever nodes * This visualization helps you understand exactly which portions of your chunks are being utilized effectively ## Optimizing your RAG pipeline

Addressing Low Relevance Scores

Low Chunk Relevance scores indicate that your chunks are probably longer than they need to be. To improve your system:
Tune your chunking strategy: Experiment with different chunking methods to create more focused chunks.
Reduce chunk size: Consider using smaller chunks that contain more concentrated relevant information.
Improve chunk boundaries: Ensure chunks are divided at logical content boundaries rather than arbitrary character counts.
Benefits: Improved system efficiency, lower cost, and reduced latency.
## Best practices Experiment with different chunking methods (sentence-based, paragraph-based, semantic-based) to find the approach that maximizes relevance. Adjust your retrieval parameters (k value, similarity thresholds) to prioritize chunks with higher relevance scores. Use Chunk Relevance alongside [Chunk Attribution Utilization](/concepts/metrics/rag/generation-quality/chunk-attribution) for a complete picture of retrieval effectiveness. Regularly analyze queries with low relevance scores to identify patterns and improve your document preprocessing or embedding strategy. When optimizing for Chunk Relevance, remember that the goal is to retrieve chunks that contain information relevant to the query. Extremely high relevance might come at the cost of missing important context if chunks are too small or too narrowly focused.