1Z0-184-25 Question 10
Single answerAn e-commerce platform uses a 512-dimensional product embedding to support semantic search on product descriptions in an Oracle Autonomous Database. They need to ensure fast, accurate retrieval of the top relevant products based on these vector embeddings. Which approach best enables efficient semantic queries in this scenario?
- A
Store the 512-dimensional embeddings in a vector column and create an approximate nearest neighbor index on the vector data.
- B
Flatten the 512-dimensional embeddings into 512 separate numeric columns and use a standard B-Tree index on each dimension.
- C
Convert the embeddings to JSON format and rely on full-text indexes for semantic queries.
- D
Store the embeddings in a VARCHAR2 column, concatenated as comma-separated values, and use string comparison to find matches.
Show answer and explanation
Correct answer: A
Explanation
In Oracle Autonomous Database (and Oracle Database 23c), the vector data type allows for storing high-dimensional embeddings in a single column. By creating an approximate nearest neighbor (ANN) index on the vector column, you can efficiently perform similarity or distance-based searches across large sets of embeddings. This approach is documented in Oracle� guidelines on vector data handling and indexing, which recommend using specialized indexes for high-dimensional data to achieve the best performance in semantic query scenarios.
- A. Correct.
Correct. Using the vector data type with a vector index (often approximate nearest neighbor) is the recommended approach. It optimizes queries by storing and searching embeddings natively, supporting high-dimensional similarity or distance-based lookups.
- B. Incorrect.
Incorrect. Splitting the embeddings into many numeric columns places a heavy burden on storage and query performance. Standard B-Tree indexes do not effectively handle high-dimensional vectors, making semantic searches inefficient.
- C. Incorrect.
Incorrect. Although JSON can hold flexible data, it is not optimized for vector-based nearest neighbor queries. Full-text indexes are designed for token-based searches rather than distance calculations between high-dimensional vectors.
- D. Incorrect.
Incorrect. Storing the embeddings as a string is extremely inefficient for similarity queries, as it prevents effective distance or dot-product searches and relies on basic string comparison, which is not suitable for semantic retrieval.