1Z0-184-25 Question 67
Single answerYour e-commerce application is hosted on Oracle Cloud Infrastructure, and you store product embeddings in an Oracle Database table using the Vector data type introduced in Oracle Database 23c. You need to perform near real-time image similarity searches to return the top 10 most similar product images based on a customer's query embedding. Which approach provides the most efficient way to accomplish this task while maintaining query performance at scale?
- A
Create a standard B-tree index on the vector column and use a regular SQL WHERE clause.
- B
Use an approximate nearest neighbor (ANN) index on the vector column to directly perform similarity lookups.
- C
Store embeddings in CSV files on OCI Object Storage and scan them in parallel using batch jobs.
- D
Apply an Oracle Text index to the vector column for similarity search using text-based operators.
Show answer and explanation
Correct answer: B
Explanation
To perform similarity searches on high-dimensional data (like embeddings), an approximate nearest neighbor (ANN) index is the recommended approach. Oracle Database 23c introduces native vector data types and indexing mechanisms that enable fast and efficient similarity queries. By leveraging the ANN index, queries scale better than traditional indexing or file-based approaches. For more details, refer to the Oracle Database 23c documentation on vector data types and indexing best practices in Oracle Cloud Infrastructure.
- A. Incorrect.
Option 1 is incorrect. B-tree indexes work well for exact match or range queries but are not designed for high-dimensional vector similarity searches. This would lead to poor performance with large embedding sets.
- B. Correct.
Option 2 is correct. Creating an approximate nearest neighbor index on the vector column is specifically designed for vector similarity searches. It allows for efficient querying of the top-k most similar embeddings, which is ideal for real-time recommendation scenarios.
- C. Incorrect.
Option 3 is incorrect. Storing embeddings as CSV files and scanning them at runtime is highly inefficient for real-time queries, as it would require reading large unindexed files and performing comparisons in memory. This is too slow for near real-time user interactions.
- D. Incorrect.
Option 4 is incorrect. Oracle Text is optimized for keyword and document analysis rather than vector-based similarity. Text indexes won't improve performance for finding nearest neighbors among numeric embeddings.