Explain the Role of Re-Ranking in the Information Retrieval Process GENERATIVE-AI-ENGINEER-ASSOCIATE Practice Questions (Page 2)

Questions 6–10

1. 6foundation · easy

   In an information retrieval system, what is the primary purpose of a re-ranking stage?

   • ATo reduce the total number of documents stored in the index
   • BTo speed up the initial retrieval by using a simpler scoring function
   • CTo improve the relevance of the top results by reordering them based on more sophisticated scoring
   • DTo increase the number of documents retrieved from the index
   Reveal Solution Discussion Select an answer first
2. 7foundation · easy

   Which statement best describes the relationship between initial retrieval and re-ranking in a retrieval pipeline?

   • AInitial retrieval and re-ranking are two independent paths that both produce the final results
   • BInitial retrieval produces a candidate set, and re-ranking refines the order of that set
   • CInitial retrieval and re-ranking are performed simultaneously on the entire corpus
   • DRe-ranking replaces initial retrieval entirely, making the initial stage unnecessary
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A search team is evaluating re-ranking models for a customer support portal. They have a labeled dataset of query-document relevance pairs. They need a re-ranking approach that can be trained on this data and that captures fine-grained interactions between the query and each document. Which approach best fits these requirements?

   • AA rule-based system that boosts documents containing the query terms in the title.
   • BA bi-encoder that embeds queries and documents separately and uses cosine similarity.
   • CA BM25 scorer with tuned k1 and b parameters.
   • DA cross-encoder that concatenates the query and document and outputs a relevance score.
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   A data science team is designing a retrieval pipeline for a legal document search application. They need to ensure that no potentially relevant document is missed, even if it means the initial list contains many irrelevant items. They plan to use a re-ranking model to refine the final order. Which configuration best aligns with the roles of initial retrieval and re-ranking?

   • AUse a single model that combines retrieval and re-ranking in one step.
   • BUse a high-recall retriever (e.g., BM25 with a large candidate set) followed by a precision-focused re-ranker (e.g., a cross-encoder).
   • CUse a precision-focused retriever (e.g., a cross-encoder on the whole corpus) and skip re-ranking.
   • DUse a high-precision retriever (e.g., a strict filter) and then re-rank with BM25.
   Reveal Solution Discussion Select an answer first
6. 10expert · hard

   A search platform is experiencing high infrastructure costs. The current pipeline retrieves 50 candidates per query and re-ranks them with a large cross-encoder. The team observes that the re-ranking stage consumes 80% of the total query latency and cost. They need to reduce costs while minimizing the impact on search quality. Which approach is most cost-effective?

   • AIncrease the candidate set to 100 to improve recall before re-ranking.
   • BKeep the candidate set at 50 but replace the cross-encoder with a bi-encoder for re-ranking.
   • CMove the re-ranking stage to a batch process that runs nightly.
   • DReduce the candidate set to 10 and use a smaller, distilled cross-encoder for re-ranking.
   Reveal Solution Discussion Select an answer first