normalised discounted cumulative gain
A ranking score that uses graded relevance, discounts results lower down the list, and divides by the best possible score so 1 is a perfect ranking.
In more detail
Järvelin and Kekäläinen introduced cumulated gain in 2002. nDCG is the right metric when some passages are more useful than others and the whole order matters, not just the first hit.
Related terms
- mean reciprocal rank — The average, over queries, of 1 divided by the rank of the first relevant result, so 1 means the answer always came first.
- recall at k — The share of the relevant passages for a query that appear in the top k results.
- labelled query set — A set of real questions, each with the passages judged relevant, often on a graded scale, used to score retrieval.
Taught in
- Measuring Retrieval — AI Engineering: LLM Apps, RAG & Agents
References
- Järvelin and Kekäläinen: Cumulated gain-based evaluation of IR techniques (ACM TOIS, 2002)
- Manning, Raghavan and Schütze: Introduction to Information Retrieval, Evaluation of ranked retrieval results