1Z0-184-25 Question 87
Select 2A research organization has thousands of technical documents and wants to implement a multi-vector similarity search on Oracle Cloud Infrastructure (OCI). They plan to store and index several embeddings per document�such as the main text embedding, a summary embedding, and a metadata embedding�to improve retrieval accuracy. Which two actions should they take to ensure high-quality multi-document search results using multi-vector similarity on OCI?
- A
Enable vector indexing for each distinct embedding field and store embeddings consistently in the same dimensional format.
- B
Manually concatenate embeddings across different semantic aspects for each document into a single vector field.
- C
Split large documents into logical �chunks,� generating separate embeddings for each chunk to capture localized semantic context.
- D
Configure each vector index to support unstructured text matching without any chunking or dimensional consistency.
Show answer and explanation
Correct answers: A, C
Explanation
When implementing multi-vector similarity search in OCI, best practices include creating multiple embeddings that capture different aspects of each document and indexing them separately (e.g., text, metadata, summaries). The documents should be split into logical sections, and each section should have a consistent dimensional embedding for robust semantic comparison. Oracle documentation recommends enabling vector indexing on your Autonomous Database or other vector-enabled services and ensuring each embedding field matches in dimensions. This approach enhances retrieval accuracy by allowing the system to compare each semantic dimension independently.
- A. Correct.
Correct. For multi-vector similarity search, it is crucial to enable a vector index for each embedding type (e.g., main text, summary, metadata) so they can be searched independently or in combination. Maintaining a consistent dimensional format (e.g., 512-dimensional embeddings for all fields) ensures proper indexing and comparison.
- B. Incorrect.
Incorrect. Simply concatenating embeddings for different semantic aspects into a single field can cause overlapping vector dimensions and reduce accuracy. Each embedding field should be indexed separately to capture unique facets of the document.
- C. Correct.
Correct. Using a chunking strategy to split large documents into smaller pieces (e.g., paragraphs or sections) lets you index each chunk� embedding accurately. This approach captures local context and increases the relevance of the search results.
- D. Incorrect.
Incorrect. Relying on unstructured text matching alone or ignoring dimensional consistency removes the core benefit of semantic embedding-based search. Proper chunking and consistent vector dimensions are essential for high-quality multi-vector searches.