Anthropic has developed a groundbreaking technique called “Contextual Retrieval” that significantly improves how AI systems access and utilize information from large knowledge bases. This innovative method addresses a critical weakness in traditional Retrieval-Augmented Generation (RAG) systems, enhancing their accuracy by up to 67%.
Preserving Context in Knowledge Retrieval
Traditional RAG systems often lose critical context when documents are split into smaller chunks for processing. This can lead to retrieval failures, where the AI fails to find the most relevant information. Contextual Retrieval tackles this issue by adding relevant contextual information to each chunk before it’s embedded or indexed. This ensures that retrieval systems can accurately identify and utilize the correct information.
Contextual Retrieval: A Two-Pronged Approach
Contextual Retrieval employs two key components: Contextual Embeddings and Contextual BM25. Contextual Embeddings enhance the embeddings used for retrieval, while Contextual BM25 improves the BM25 scoring used for ranking retrieved chunks. Together, these techniques dramatically reduce retrieval failures, with Contextual Retrieval embeddings alone cutting failures by 35%, and the combination of Contextual Retrieval embeddings and Contextual BM25 reducing failures by 49%. Adding a reranking step on top of these techniques slashes failures by an impressive 67%.
Why Should You Care?
These gains in accuracy directly translate to better performance in downstream tasks, potentially improving the quality of AI-generated responses across a wide range of applications.
– Enhances retrieval accuracy by up to 67%
– Improves performance in various knowledge domains
– Cost-efficient implementation with prompt caching
– Consistent improvements across different embedding models
– Customizable for specific domains with tailored prompts