1Z0-184-25 Question 85
Single answerYour company is developing an AI-driven knowledge base in Oracle Cloud Infrastructure that uses multiple embeddings (title, abstract, and body) for each document to improve search relevance. You plan to leverage Oracle Database� vector capabilities for multi-vector similarity searches across thousands of documents. Which approach best ensures efficient retrieval while allowing you to apply different weights to each embedding during queries?
- A
Store all embeddings in a single vector column and use a single index for similarity search.
- B
Create separate vector columns (e.g., TITLE_VEC, ABSTRACT_VEC, BODY_VEC) within the same table and index each column individually.
- C
Encode all embeddings into a single JSON column and use standard text indexing for multi-field queries.
- D
Use Object Storage to persist embeddings as files and run a full-table scan to compare vector similarities.
Show answer and explanation
Correct answer: B
Explanation
In real-world scenarios, separate vector columns allow more flexible manipulation of embeddings and efficient multi-vector similarity queries. By indexing each vector column independently, Oracle Database can quickly rank documents based on each embedding's contribution to overall relevance. Refer to Oracle� Database Vector Search documentation (available for Oracle Database 23c and later) for detailed guidelines on how to create, configure, and query vector columns for multi-document search use cases.
- A. Incorrect.
Option 1 is incorrect because combining all embeddings into a single vector makes it hard to weight them differently for searches. You also lose granularity for partial matches on individual embeddings.
- B. Correct.
Option 2 is correct. Storing each embedding in a separate column allows you to tune and index each embedding independently. Oracle Database� vector search feature supports creating multiple vector columns and corresponding indexes, enabling you to apply distinct weights or filters for each column during multi-vector similarity searches.
- C. Incorrect.
Option 3 is incorrect because standard text indexing on JSON data does not properly handle vector similarity�we need a vector index for accurate multi-vector similarity search. JSON indexing focuses on text-based or structural queries, not vector-based.
- D. Incorrect.
Option 4 is incorrect because storing embeddings as files in Object Storage and then performing a scan is highly inefficient and does not leverage the native vector index capabilities in Oracle Database.