⚠️ Please note that the reranking extension feature is in closed beta phase.
The reranking extension provides the possibility to rerank your search results based on your specific logic, for example reranking using:
- A custom diversification and merchandising strategy (e.g. based on number of impressions)
- A custom personalization strategy (e.g using a query time computed user score)
- A custom machine learning model (e.g using a semantic reranker)
The reranking extension is a post-processing step for a search request that gives you more control on the final search results ranking.
The sequence diagram below shows how the reranking extensions works.
- A user triggers a new search
- The engine performs a search and sends a specified chunk of results to the reranking extension endpoint
- The reranking service performs the reranking and returns the reordered result list to the engine
- The engine performs the highlighting and pagination before returning the reordered results to the user
The reranking extension joins the existing ranking pipeline as the last step of the tie-breaking algorithm.
If you have activated the advanced ranking features such as AI-reranking and Personalization, the resulting ranking pipeline will look as follows: