1Z0-184-25 Question 114
Single answerYour data science team is building a recommendation system on Oracle Cloud Infrastructure using Oracle Database 23c to store vector embeddings for product similarity. They want to perform approximate nearest neighbor searches on these embeddings without sacrificing database performance or manageability. Which approach is recommended for storing and querying the vector embeddings in this scenario?
- A
Store the embeddings as a BLOB and rely on a traditional B-tree index for approximate similarity.
- B
Use the new Vector data type in Oracle Database 23c and create a vector index for performing similarity searches.
- C
Flatten the embedding arrays into separate numeric columns for each dimension and manually build multiple indexes.
- D
Export the embeddings to an external NoSQL data store and periodically synchronize results back to Oracle Database.
Show answer and explanation
Correct answer: B
Explanation
In Oracle Database 23c, the Vector data type and associated vector index provide a first-class solution for storing and querying vector embeddings. This new functionality is designed to optimize similarity searches and most closely aligns with real-world requirements for systems like product recommendations or semantic search. Refer to the Oracle Database 23c documentation on vector data types and indexing for detailed implementation guidance.
- A. Incorrect.
Option 1 is incorrect because BLOB storage with a traditional B-tree index does not natively support approximate nearest neighbor queries or efficient vector similarity searches. This approach could lead to suboptimal performance and complex querying logic.
- B. Correct.
Option 2 is correct. Oracle Database 23c introduces a dedicated Vector data type and vector index for efficient similarity searches. By defining a specialized column and index for vector data, you unlock built-in algorithms for approximate nearest neighbor queries without overly complex workarounds.
- C. Incorrect.
Option 3 is incorrect because flattening embeddings into separate columns increases schema complexity and requires managing many numeric columns (especially for high-dimensional vectors). Multiple indexes on each dimension are generally not practical or efficient for similarity search scenarios.
- D. Incorrect.
Option 4 is incorrect because exporting embeddings to an external NoSQL database for similarity searches introduces additional overhead, potential latency, and complicates data synchronization. While feasible in some architectures, it is not the recommended approach when Oracle Database 23c offers a built-in Vector data type and native similarity search capabilities.