Explain the Role of Re-Ranking in the Information Retrieval Process GENERATIVE-AI-ENGINEER-ASSOCIATE Practice Questions (Page 3)

Questions 11–15

1. 11application · medium

   A team is building a job search engine. They have a dataset of user interactions (clicks, applications) and want to improve the ranking of job postings. They plan to use a re-ranking model that can learn from these interactions. Which re-ranking technique is most appropriate?

   • AA cross-encoder that scores the semantic similarity between the job description and the user's resume.
   • BA bi-encoder that embeds jobs and resumes separately and uses cosine similarity.
   • CA learning-to-rank model, such as LambdaMART, trained on click and application data.
   • DA rule-based system that boosts jobs from preferred companies.
   Reveal Solution Discussion Select an answer first
2. 12foundation · easy

   Which of the following is a common re-ranking technique used to improve the relevance of retrieved results?

   • AUsing a cross-encoder model to score query-document pairs
   • BUsing a random shuffle of the retrieved documents
   • CUsing a simple term-frequency (TF) scoring function
   • DUsing a bi-encoder to embed the query and documents separately and then computing cosine similarity
   Reveal Solution Discussion Select an answer first
4. 13foundation · easy

   In a typical retrieval pipeline, where does the re-ranking stage occur?

   • AAfter the initial retrieval stage and before the final result set is returned
   • BDuring the indexing phase, when documents are added to the index
   • CAfter the user has reviewed the results and provided feedback
   • DBefore the initial retrieval stage
   Reveal Solution Discussion Select an answer first
5. 14expert · hard

   A search team is evaluating re-ranking models for a multilingual customer support portal. They have a small labeled dataset for each language. They need a re-ranking approach that can generalize across languages with limited training data. Which approach is most suitable?

   • AA rule-based system that boosts documents containing query terms in the same language.
   • BA bi-encoder that uses a multilingual embedding model and cosine similarity.
   • CA learning-to-rank model with language-specific features, trained separately for each language.
   • DA cross-encoder based on a multilingual transformer, fine-tuned on the available data.
   Reveal Solution Discussion Select an answer first
6. 15foundation · easy

   How does re-ranking affect the overall efficiency of a retrieval system?

   • AIt increases efficiency by reducing the number of documents that need to be indexed
   • BIt decreases efficiency because re-ranking models are computationally expensive, but it improves result quality
   • CIt has no effect on efficiency because it is performed offline
   • DIt increases efficiency by using the same model for both initial retrieval and re-ranking
   Reveal Solution Discussion Select an answer first