1Z0-184-25 Question 115
Single answerYou are adding a semantic search feature to your e-commerce application, which generates 256-dimensional vector embeddings for each product description. You plan to store these embeddings in Oracle Autonomous Database and need to efficiently query them for nearest neighbors across thousands of products. Which approach should you recommend for optimal storage and retrieval performance?
- A
Store the embeddings in a JSON array column and leverage Oracle Text indexes for approximate nearest neighbor searches.
- B
Use the native Vector Data Type in Oracle Database 23c or later and create a specialized Vector Index for similarity queries.
- C
Convert each embedding into a comma-separated string and store it in a VARCHAR2 column, then apply a standard B-tree index.
- D
Store all embeddings in a single BLOB column and run custom SQL queries without indexing, relying on full table scans.
Show answer and explanation
Correct answer: B
Explanation
Oracle Database 23c (and subsequent releases) provide built-in support for vector data, allowing you to store embeddings as a specialized Vector Data Type. You can then create a Vector Index to perform similarity searches (e.g., k-nearest neighbor queries) efficiently. This new feature is documented in Oracle� official documentation on Vector Search functionality, providing a significant performance advantage over generic indexing or text-based approaches when working with large-scale vector embeddings.
- A. Incorrect.
Option 1 is incorrect. JSON arrays are flexible, but Oracle Text indexes are designed for textual data, not numeric high-dimensional vectors. They cannot provide efficient nearest neighbor vector searches.
- B. Correct.
Option 2 is correct. Oracle Database 23c introduces a native Vector Data Type and a specialized Vector Index, enabling efficient similarity queries on high-dimensional embeddings. This approach significantly reduces query latency compared to generic indexing methods.
- C. Incorrect.
Option 3 is incorrect. Storing vectors as a comma-separated string in a VARCHAR2 column prevents efficient vector-based similarity searches. A standard B-tree index will not help with approximate nearest neighbor operations on numerical embeddings.
- D. Incorrect.
Option 4 is incorrect. Storing all embeddings in a single BLOB column without a specialized index forces full table scans, which is highly inefficient and does not leverage vector-specific capabilities available in Oracle Database 23c.