Filter Extraneous Content in Source Documents That Degrades Quality of a RAG Application GENERATIVE-AI-ENGINEER-ASSOCIATE Practice Questions (Page 4)

Questions 16–20

1. 16application · medium

   A retail company is building a RAG application over product reviews. The reviews contain a lot of noise, including 'This is a sponsored review', 'I received this product for free', and repetitive promotional text. The team wants to filter this noise to improve the quality of the RAG responses. Which approach is most appropriate?

   • ARemove all reviews that contain a link, as these are likely to be promotional.
   • BUse a machine-learning classifier trained to detect sponsored or promotional content in reviews, and remove those reviews from the corpus.
   • CRemove all reviews that are shorter than 10 words, as they are likely to be noise.
   • DUse a regex pattern to remove any review that contains the word 'sponsored' or 'free'.
   Reveal Solution Discussion Select an answer first
2. 17foundation · easy

   After applying a new filtering step to a RAG pipeline, what is the MOST direct way to evaluate its impact on retrieval quality?

   • ACheck the size of the vector index in bytes
   • BMeasure the change in retrieval precision and recall on a test set of queries
   • CCompare the total number of chunks before and after filtering
   • DTime the filtering process to ensure it completes quickly
   Reveal Solution Discussion Select an answer first
4. 18foundation · easy

   During document ingestion for a RAG pipeline, which of the following is an example of extraneous content that should be identified and removed?

   • AAn appendix with detailed reference data
   • BA copyright notice that appears on every page
   • CA glossary of technical terms used in the document
   • DA table of contents that lists all chapters and sections
   Reveal Solution Discussion Select an answer first
5. 19expert · hard

   A team is building a RAG application over a corpus of academic papers. They apply a filtering step that removes all 'Acknowledgements' sections, as these are considered noise. After evaluation, they find that retrieval precision has improved, but the answer generation for queries about 'funding sources' has degraded. The team is considering whether to revert the filter. What is the best way to evaluate the impact of this filter?

   • ARevert the filter immediately, as any degradation in answer quality is unacceptable.
   • BRemove the filter only for documents that contain the word 'funding' in the acknowledgements.
   • CRun a side-by-side evaluation on a held-out set of queries, comparing retrieval precision and answer generation quality with and without the filter, and measure the trade-off.
   • DKeep the filter and ignore the degradation, as the improvement in retrieval precision is more important.
   Reveal Solution Discussion Select an answer first
6. 20application · medium

   A healthcare company is building a RAG system over clinical trial protocols. The documents contain a 'Statistical Analysis Plan' section that includes complex formulas and tables, alongside narrative descriptions of the analysis methods. The team applies a filtering step that removes all lines containing mathematical symbols (e.g., '=', '+', 'σ') to clean the text. After deployment, they notice that retrieval for queries about 'how the primary endpoint was analyzed' has degraded significantly. What is the most likely cause?

   • AThe filtering step was too conservative and failed to remove enough of the statistical formulas, leaving noise in the index.
   • BThe retrieval system is not configured to handle mathematical queries, so the degradation is unrelated to the filtering step.
   • CThe filtering removed the narrative text that described the analysis methodology because it also contained mathematical symbols.
   • DThe filtering was applied after chunking, causing the embeddings to be computed on the noisy text.
   Reveal Solution Discussion Select an answer first