Reranking Search Results
Reranking is the process by which an initial set of query results is rescored to produce a more refined and accurate ranking. It's usually performed with a machine learned model that, in contrast to the extremely fast retrieval step, is slower but more accurate.
Reranking is a beta feature in the platform, and only available as a single
reranker for English results, and can be enabled by specifying 272725717 as
In order to use the reranker you need to specify two additional parameters in
numResults. These tell Vectara
which results to rerank. In most scenarios, it makes sense to set
0 so that you're reranking all of the best initial results, and then to set
numResults to the total number of documents you wish to rerank.