1Z0-184-25 Question 117
Single answerYou manage an AI-driven product recommendation system deployed on Oracle Cloud Infrastructure. Your data science team generates vector embeddings to represent product descriptions. You want to store these embeddings in your Oracle Database and run efficient similarity searches in real time. Which approach should you implement to achieve this goal?
- A
Store the vector embeddings as JSON arrays in a JSON column, and rely on a standard B-tree index for queries.
- B
Use Oracle Database� specialized vector column type and create a vector-based index to perform approximate nearest neighbor searches.
- C
Save the embeddings in CLOB columns as comma-separated values, then leverage Oracle Text indexes for text-based matching.
- D
Persist the embeddings in Oracle Cloud Object Storage in binary format and reference them via external tables for similarity queries.
Show answer and explanation
Correct answer: B
Explanation
Storing vector embeddings in Oracle Database and using its specialized vector-based indexing offers the most efficient approach for serving similarity queries. Oracle� native vector data type and approximate nearest neighbor (ANN) indexes are designed to handle high-dimensional embeddings. Refer to Oracle� Database documentation (for example, Oracle Database 23c or later) on vector search to learn how to create vector columns and build vector indexes for AI-driven use cases.
- A. Incorrect.
Option 1 is incorrect. Storing embeddings as JSON arrays is possible in Oracle Database, but a standard B-tree index on JSON data is not optimized for high-dimensional similarity queries. You would likely see significant performance degradation when dealing with typical embedding sizes.
- B. Correct.
Option 2 is correct. Oracle Database (starting with certain releases that support vector search) offers a dedicated vector data type and specialized indexing to handle approximate nearest neighbor queries. This provides optimized similarity searches for high-dimensional embeddings and is widely considered a best practice for AI-driven applications.
- C. Incorrect.
Option 3 is incorrect. Storing embeddings in a CLOB column and using Oracle Text indexes targets text-pattern or keyword searches, not high-dimensional vector similarity. This approach would not produce accurate or efficient embedding-based results.
- D. Incorrect.
Option 4 is incorrect. While Oracle Cloud Object Storage can hold large binary objects, retrieving them for each query and performing vector similarity on external data would be inefficient. This setup does not leverage the built-in vector indexing capabilities of Oracle Database.