1Z0-184-25 Question 119
Single answerYou are designing a semantic search application on Oracle Cloud Infrastructure where you need to store high-dimensional vector embeddings generated from textual data. The goal is to achieve efficient similarity queries (e.g., nearest neighbor lookups) on these embeddings. In Oracle Database 23c and later, which approach is recommended to enable fast vector operations and indexing for these embeddings?
- A
Store the embeddings in a dedicated vector column type introduced in Oracle Database 23c, and create a specialized vector index for approximate nearest neighbor queries.
- B
Store the embeddings as CSV strings in a VARCHAR2 column for easy readability, and rely on full-text indexes for searching.
- C
Store the embeddings in a single NUMBER column by concatenating all dimensions into one numeric field.
- D
Store the embeddings in a JSON column without creating any additional indexes to keep the schema flexible.
Show answer and explanation
Correct answer: A
Explanation
When designing applications that rely on semantic similarity, storing vector embeddings in Oracle Database 23c or later with the built-in vector data type and vector index is the most efficient approach. This ensures proper handling of high-dimensional data and fast similarity queries through approximate nearest neighbor methods. Refer to Oracle Database 23c documentation for details on the syntax and best practices for defining vector columns and creating vector indexes.
- A. Correct.
Correct. Oracle Database 23c introduces a specialized vector column type and a vector index that enable efficient storage and similarity searches on high-dimensional embeddings. By using this native approach, you can leverage built-in approximate nearest neighbor search capabilities, reducing both development complexity and query latency.
- B. Incorrect.
Incorrect. Storing embeddings as CSV strings in a VARCHAR2 column makes them human-readable but does not allow optimized similarity queries. Full-text indexes are designed for linguistic token-based searches, not high-dimensional vector distances.
- C. Incorrect.
Incorrect. Concatenating dimensions into a single NUMBER column sacrifices clarity and makes it difficult to apply any indexing strategy optimized for multi-dimensional data. It also complicates insert and update logic since each dimension must be properly parsed and packed.
- D. Incorrect.
Incorrect. Storing embeddings in a JSON column without specialized indexing may offer flexibility but will not provide the performance needed for similarity or nearest neighbor queries. Native vector indexing is recommended for this purpose.