MS MARCO
Real search queries with passages and answers, the standard retrieval benchmark.
Overview
MS MARCO collects questions people actually asked a search engine, paired with passages from the web that answer them and human-written answers. It solves the problem that retrieval benchmarks built by annotators produce artificial queries that do not look like what users type. It exists as the dataset rerankers and embedding models are trained and compared on, which is why its quirks show up in production behaviour.
Technology Graph
How MS MARCO connects to the rest of the stack.
Applied to
- AI Search application Applications MS MARCO used for AI Search High ✓ source
- Embedding concept Techniques MS MARCO used for Embedding High ✓ source
- Reranking technique Techniques MS MARCO used for Reranking High ✓ source