Reranking

« Back to Glossary Index

Re-ordering of results using a secondary post-process applied to a previously generated candidate set. Reranking algorithms can be more computationally expensive as they’re applied, typically, to smaller subsets of the corpus. An example might be applying cosine similarity scoring between doc and query embeddings (vectors) – too expensive to apply to all potential results in large corpi.





Glossary Comments

If you think we've got something wrong or you can suggest a term we should feature in the Glossary, let us know here:

« Back to Glossary Index