1Z0-184-25 Question 118
Select 2An e-commerce company is building a product recommendation engine that relies on vector embeddings to measure semantic similarity between products. They plan to store these embeddings in an Oracle Database 23c environment on Oracle Cloud Infrastructure. Which two approaches represent best practices for storing and querying these vector embeddings for optimal performance?
- A
Use the native vector data type in Oracle Database 23c and create an Approximate Nearest Neighbor (ANN) index on the vector column.
- B
Use the SDO_GEOMETRY data type from Oracle Spatial, leveraging geospatial indexes for approximate vector similarity searches.
- C
Split each embedding dimension into individual numeric columns and rely on B-tree indexes for each dimension.
- D
Store embeddings as JSON arrays in a column with the JSON data type and create an ANN index that supports vector queries on JSON.
Show answer and explanation
Correct answers: A, D
Explanation
Oracle Database 23c introduces native vector data type support and specialized vector indexes (ANN) for efficient similarity searches on high-dimensional data. Storing embeddings in a dedicated vector column offers straightforward indexing and query optimization. Alternatively, storing them as JSON arrays is also supported, provided you configure the proper ANN indexing. For more details, refer to the official Oracle Database 23c documentation on vector data and indexing best practices.
- A. Correct.
Option 1 is correct. Oracle Database 23c introduces a native vector data type and the ability to create specialized indexes (ANN indexes) for vector similarity queries. Storing and indexing vectors natively offers performance advantages for high-dimensional data.
- B. Incorrect.
Option 2 is incorrect. SDO_GEOMETRY is designed for spatial data rather than general machine learning embeddings. While it can handle some vector-like operations, it is not optimized for high-dimensional embeddings and does not directly leverage the new vector search capabilities in Oracle Database 23c.
- C. Incorrect.
Option 3 is incorrect. Splitting embeddings into many numeric columns and relying on B-tree indexes results in suboptimal performance for high-dimensional vectors. This approach can become cumbersome and is not recommended for large embeddings.
- D. Correct.
Option 4 is correct. Oracle Database 23c supports storing vector embeddings in JSON arrays and creating an ANN index on these JSON columns. This leverages the vector search engine without requiring a native vector data type column, providing flexibility for JSON-based applications.