Reranker Model
A model that performs fine-grained re-ranking of candidate documents retrieved in an initial recall stage. By evaluating the relevance between a query and documents, it improves retrieval quality and is commonly used in the reranking phase of RAG systems.
Timeline (last 90 days)
In recommender systems, the weight the ranking model assigns to "political content" and its preference for "credible sources" both systematically affect visibility.
If negative signals such as high bounce rates and short dwell times are used as training data, they may contaminate the ranking model and further limit the exposure of high-quality content.
Every answer a user receives in Perplexity begins with embedding models and ranking models filtering for relevant results.
Ranking models perform fine-grained sorting on top of initially recalled results, pushing the most relevant content to the front.
All Facts (4)
In recommender systems, the weight the ranking model assigns to "political content" and its preference for "credible sources" both systematically affect visibility.
50%UnverifiedIf negative signals such as high bounce rates and short dwell times are used as training data, they may contaminate the ranking model and further limit the exposure of high-quality content.
50%UnverifiedEvery answer a user receives in Perplexity begins with embedding models and ranking models filtering for relevant results.
50%UnverifiedRanking models perform fine-grained sorting on top of initially recalled results, pushing the most relevant content to the front.
50%