Kvíz spojený s průvodcem · Střední úroveň
Kvíz o strategiích rozbíjení dokumentů
Prohlédněte si hranice bloků, překrývání, sémantické a strukturální rozdělení a jak se mohou empirické výsledky lišit podle datové sady. Porovnejte konfigurace na korpusu.
Otázka 1 z 8
What does Amazon Bedrock say its default chunking approach does?
Náhled každé otázky8 otázky
Odpovědi a vysvětlení se zobrazují, jak budete v kvízu.
- What does Amazon Bedrock say its default chunking approach does?
- Splits text into roughly 300-token chunks while preserving complete sentence boundaries
- Splits every document into one token
- Uses one chunk for every knowledge base regardless of size
- Groups documents by alphabetical title
- Why can the buffer size matter in semantic chunking?
- It changes the source document’s spelling
- It determines how many neighboring sentences are included when creating embeddings
- It identifies the correct speaker
- It specifies the model’s output vocabulary
- Jaký je hlavní kompromis mezi velmi malými a velmi velkými kousky?
- Malé kousky se vždy vkládají pomaleji
- Velké kusy nelze nikdy získat
- Malé kousky jsou přesné, ale ztrácejí kontext; velké kusy zachovávají kontext, ale snižují relevanci
- Není v tom žádný smysluplný rozdíl
- In Amazon Bedrock, what may its parsed-content chunker preserve as logical boundaries?
- Only individual characters
- Image colors and crop boxes
- The prompt’s output format
- Pages or sections
- How does Amazon Bedrock describe hierarchical chunking at retrieval time?
- It retrieves child chunks and can replace them with broader parent chunks
- It discards parent text and never returns broader context
- It merges all documents into one vector
- It converts every chunk to an image
- How does Amazon Bedrock’s semantic chunking propose boundaries between sentences?
- By sorting sentences alphabetically
- Using sentence-distance thresholds to identify less-similar adjacent sentences
- By splitting every document at the same character count
- By detecting speaker identity
- What did a 2026 chunking study find about overlap in its tested Natural Questions setup?
- It always doubled answer accuracy
- It eliminated all retrieval errors
- It increased indexing cost without a measurable answer-quality benefit
- It reduced indexing cost to zero
- What happened when retrieved context grew beyond about 2,500 tokens in that study?
- Accuracy rose without limit
- All questions became exact matches
- The retriever stopped using embeddings
- Answer quality showed a context cliff