1Z0-184-25 Question 113
Single answerYou have developed a product recommendation feature for your e-commerce platform on Oracle Cloud Infrastructure. To enable semantic search for product descriptions, you have generated vector embeddings using a large language model. Your goal is to store these embeddings in your Oracle Database so that you can quickly perform similarity searches to find the most relevant products. Which approach should you use to store and index these vector embeddings for optimal search performance?
- A
Store the embeddings in a CLOB column and use Oracle Text indexes to handle similarity queries.
- B
Store each dimension of the embeddings in separate NUMBER columns and create a standard B-Tree index on these columns.
- C
Use Oracle Database's native vector data type and build a specialized vector index to enable efficient similarity queries.
- D
Use a JSON column to store all vectors in one document and rely on a functional index for vector-based searches.
Show answer and explanation
Correct answer: C
Explanation
Oracle Database� specialized vector data type and indexing (introduced in Oracle Database 23c) provide efficient mechanisms for storing and querying vector embeddings. Using these features allows for quick k-nearest neighbor (kNN) searches and other similarity calculations. Refer to the Oracle Database 23c documentation, particularly the sections on vector data type and vector indexing, for best practices in defining schemas and indexes that handle large-scale semantic search workloads.
- A. Incorrect.
Option 1: Storing embeddings in a CLOB and using Oracle Text focuses on text-based searching, not numerical similarity. Text indexes cannot efficiently handle vector similarity calculations, so this approach is unsuitable.
- B. Incorrect.
Option 2: Splitting embeddings into multiple numeric columns and using B-Tree indexes might work for exact or range queries, but it is inefficient for high-dimensional similarity searches. B-Tree indexes are not designed for vector-based nearest neighbor queries.
- C. Correct.
Option 3: (Correct) Native vector data type and specialized vector indexes in Oracle Database (available in the latest releases) are designed to optimize similarity searches in high-dimensional spaces. This approach leverages built-in functions for vector queries, significantly improving performance for semantic or nearest neighbor searches.
- D. Incorrect.
Option 4: Although storing embeddings in a JSON column is possible, functional indexes derived from JSON content are not optimized for high-dimensional vector similarity. You would still have to parse and calculate similarity manually.