1Z0-184-25 Question 69
Select 2You are helping a retail company implement a similarity search feature in Oracle Cloud Infrastructure 2025. They plan to store product description embeddings in an Autonomous Database and want efficient approximate nearest neighbor (ANN) queries at scale. Which two approaches should be part of your design to achieve optimal similarity search performance?
- A
Create a dedicated vector data type column and build a vector index on it to enable approximate nearest neighbor queries.
- B
Store the embeddings as text strings and rely solely on a full-text (CTXSYS) index to retrieve similar items.
- C
Enable dimension-based or range-based partitioning on large embedding tables to reduce query overhead.
- D
Use standard B-tree indexes on the numeric columns storing the embeddings to reduce storage complexity.
Show answer and explanation
Correct answers: A, C
Explanation
When building a similarity search solution on Oracle Cloud Infrastructure 2025, storing embeddings in a vector data type and using a specialized vector index is crucial for running efficient approximate nearest neighbor queries. B-tree and full-text indexes are not designed for multi-dimensional vector lookups. Additionally, partitioning large tables containing vector data can further optimize query performance by minimizing the search space. Refer to Oracle� documentation on Autonomous Database vector capabilities and partitioning strategies for best practices in high-dimensional similarity search.
- A. Correct.
Correct. Oracle Database (including Autonomous Database) provides vector data type support and specialized vector indexing (often called an ANN index). This approach is specifically designed for similarity search on high-dimensional data such as embeddings, offering significant performance improvements compared to traditional indexes.
- B. Incorrect.
Incorrect. Storing embeddings as plain text and relying on full-text or keyword-based indexes doesn�t properly leverage vector operations. Full-text indexes are designed for textual relevance, not for scoring multi-dimensional vector similarity.
- C. Correct.
Correct. Partitioning a large embeddings table can help reduce query overhead. By organizing data into more manageable segments, the database can quickly exclude irrelevant partitions, thus speeding up approximate nearest neighbor searches over very large vectors.
- D. Incorrect.
Incorrect. B-tree indexes are efficient for equality or range queries on numeric columns, but they do not handle high-dimensional vector similarity lookups effectively. A specialized vector index is needed for performance in similarity search use cases.