179 questions
No questions match those filters.
← All questions
Retrieval & EmbeddingsHard· Length Normalization· BM25· Chunking
A teammate argues length normalization is irrelevant be...
This is one of the questions in the full AI/ML interview bank. Pro unlocks all 1789 questions; Premium includes the same bank plus the highest daily Practice limit.
See plansEven with a 512-token chunking strategy, documents often vary in length due to PDF tables, code blocks, or short trailing chunks. Setting b = 0 removes the length penalty, which can cause short fragments to win on raw term sums. Instead of relying on theory, you should pre-commit to a measurement: compare b = 0 against b = 0.75 by stratifying nDCG@10 by chunk-length decile to see if the parameter actually affects the top-ranked results.