1Z0-184-25 Question 89
Select 2Your organization runs a knowledge base that consolidates technical manuals, marketing materials, and code snippets into a single repository in Oracle Autonomous Database. You plan to generate multiple embeddings per document (e.g., one for core content, one for metadata, and one for context) to improve relevancy. Which practices should you follow to implement multi-vector similarity search across these documents?
- A
Store each embedding in a dedicated vector column so you can weigh or combine them independently during queries.
- B
Embed all vectors into a single JSON array, and Oracle automatically partitions the multi-vector search across different embeddings without additional configuration.
- C
Use configurable distance functions (e.g., weighted average) on embedded columns to combine relevance from multiple vectors in a single query.
- D
Maintain separate tables for each embedding type, then merge the query results manually to achieve multi-vector similarity search.
Show answer and explanation
Correct answers: A, C
Explanation
When implementing multi-vector similarity search in Oracle Autonomous Database, you typically store each vector (representing different aspects of a document) in its own column. This approach allows flexible indexing and querying strategies, including combining multiple embeddings with weighted distance functions. Refer to the Oracle Database 23c Vector Search documentation for detailed steps on creating multi-vector indexes and writing queries that leverage multiple columns simultaneously.
- A. Correct.
Correct. Placing each embedding (e.g., main content, metadata, context) in its own vector column allows fine-grained control over indexing and querying. This also enables you to separately weight or combine embeddings for a more accurate multi-vector search.
- B. Incorrect.
Incorrect. Simply placing all embeddings in one JSON array does not automatically provide granular multi-vector search. You would need to configure or parse them individually. Oracle� vector-based search typically requires defining separate columns for best results.
- C. Correct.
Correct. Oracle Database allows specifying distance functions that can incorporate multiple columns (embeddings). You can apply operations like a weighted average or sum of distances to simultaneously evaluate multiple embeddings within a single query.
- D. Incorrect.
Incorrect. Splitting into multiple tables unnecessarily complicates the design. Oracle� multi-vector indexing and querying can be done within a single table by storing each embedding in separate columns, avoiding the need to merge results manually.