How to Evaluate Retrieval Quality in RAG Pipelines (Part 3): DCG@k and NDCG@k

Unlike binary metrics that label results as simply relevant or not, graded evaluation metrics assess retrieval quality for RAG pipelines on a spectrum. A key graded metric, Discounted Cumulative Gain (DCG@k), scores retrieved items based on their degree of relevance while penalizing those that appear lower in the search results. However, DCG scores are difficult to compare across different-sized result sets, a limitation addressed by Normalized Discounted Cumulative Gain (NDCG@k). NDCG@k solves this by normalizing the score against an ideal, perfectly ranked result list, producing a final value between 0 and 1 for fair and consistent performance evaluation.