Better RAG Results With Reciprocal Rank Fusion (RRF) and Hybrid Search

RRF is a compound type of retriever that combines results obtained from different types of retrievers into one unified ranking.

Some practical use cases of RRF are in information retrieval in search engines, document retrieval, recommendation systems, and e-commerce search systems.

Reciprocal rank score is a metric used to evaluate the effectiveness of a search system to return relevant results. It is the inverse of the rank of the particular result of the query and is calculated as 1/(rank+constant k).