1Z0-184-25 Question 13
Select 2You are designing a recommendation system on Oracle Cloud Infrastructure using Oracle Database� vector data type for storing text embeddings. The system needs to support semantic queries to retrieve the most relevant documents based on user input. Which two statements describe best practices for implementing this solution?
- A
Create an approximate nearest-neighbor index on the vector column to optimize similarity searches.
- B
Ensure the dimension of your vector data type aligns exactly with the embedding dimension produced by your ML model.
- C
Store all vectors as Base64-encoded BLOBs so that you can convert them to the vector data type only at query time.
- D
Partition the table by the number of embedding dimensions to distribute storage across multiple tablespaces.
Show answer and explanation
Correct answers: A, B
Explanation
In Oracle Database, storing text embeddings in a vector column and creating an approximate nearest-neighbor index are established best practices when leveraging semantic queries. Matching the column� vector dimension to the ML model� embedding dimension ensures optimal indexing and prevents alignment errors. Refer to Oracle� documentation on the vector data type and ANN indexing in Autonomous Database for more guidance on performance tuning and design considerations.
- A. Correct.
CORRECT. Creating an approximate nearest-neighbor (ANN) index on the vector column is a recommended practice. This index allows the database to efficiently locate embedding vectors that are most similar to a given query vector, greatly improving performance in large-scale environments.
- B. Correct.
CORRECT. The embedding-dimensionality (e.g., 256, 512, etc.) must match the defined dimension of the vector data type in the table. Misalignment of dimensions can lead to errors or result in suboptimal indexing and query performance.
- C. Incorrect.
INCORRECT. Converting from Base64-encoded BLOBs to vector data types at query time adds unnecessary overhead and complexity. It is more efficient to store and index data directly in the vector data type to enable instantaneous similarity searches.
- D. Incorrect.
INCORRECT. Partitioning by embedding dimension is not a recognized best practice for vector-based queries. Partitioning strategies may be used for other reasons (e.g., data lifecycle management), but splitting data by the number of embedding dimensions adds no direct performance benefit for semantic searches.