BM25 and Lexical Retrieval Quiz

Answers and explanations appear as you take the quiz.

1. What does BM25 primarily use to rank documents?
   • Dense semantic embeddings
   • Term frequency, inverse document frequency, and document length
   • The publication date of each document
   • User click-through history only
2. What role does inverse document frequency (IDF) play in BM25?
   • It boosts very common words like 'the'
   • It measures how long each document is
   • It gives rarer, more distinctive terms higher weight
   • It counts the total number of documents only
3. Why does BM25 apply document-length normalization (the b parameter)?
   • So that long documents are not unfairly favored just for containing more words
   • To make every document the same length
   • To delete short documents
   • To translate documents into a standard language
4. A key feature of BM25's term-frequency handling is that it:
   • Increases linearly forever with each repeat
   • Ignores term frequency entirely
   • Doubles the score for every repeated word
   • Saturates, so extra repeats add diminishing relevance
5. What data structure makes BM25 fast at scale?
   • An inverted index mapping terms to the documents containing them
   • A neural network
   • A single sorted array of all text
   • A graph database of citations
6. In modern search systems, BM25 is often used to:
   • Replace all neural models entirely
   • Serve as a fast first-stage retriever before a neural reranker, or be fused with dense scores
   • Generate brand-new documents
   • Encrypt the search index
7. In the standard BM25 formula, what effect does setting b to zero have?
   • It removes all inverse-document-frequency weights
   • It makes every matching term contribute zero
   • It changes the index from lexical terms to dense vectors
   • It disables document-length normalization in the term-frequency denominator
8. When BM25 is used to mine DPR training negatives, what makes a retrieved nonanswer passage a hard negative?
   • It has no words in common with any document
   • It is automatically the positive passage because its lexical score is high
   • It looks relevant through term overlap but does not support the question's answer
   • It contains no readable text and is never retrieved