1Z0-184-25 Question 84
Single answerYou work for a large e-commerce company that stores millions of product embeddings in an Oracle Database on OCI. Your team needs to quickly retrieve the most similar products for any incoming query vector. Which action best ensures that you can perform approximate similarity searches efficiently on this large dataset?
- A
Create a vector index on the embedding column using approximate similarity search parameters to optimize query speed.
- B
Use a standard B-tree index on the embedding column to handle approximate search queries.
- C
Distribute the embeddings across multiple tables based on product categories and rely only on row-level scans for similarity matches.
- D
Partition the table by product ID ranges to accelerate nearest neighbor queries without creating any specialized indexes.
Show answer and explanation
Correct answer: A
Explanation
To perform approximate similarity searches at scale, Oracle recommends creating specialized vector indexes on the columns storing embeddings. This index type uses data structures and algorithms (e.g., approximate nearest neighbor) optimized for high-dimensional vectors. Refer to Oracle Database documentation on Vector Search and Approximate Nearest Neighbor Indexing for detailed guidance on creating and querying vector indexes with minimal performance overhead.
- A. Correct.
Option 1 is correct. An approximate vector index is specifically optimized for vector data and similarity queries, significantly speeding up nearest neighbor searches on large datasets. By configuring approximate parameters during index creation, you reduce the computational overhead when retrieving similar vectors.
- B. Incorrect.
Option 2 is incorrect. A B-tree index is designed for ordered data (e.g., numeric or lexicographic). B-trees are not suitable for high-dimensional vector data and cannot efficiently handle approximate similarity searches.
- C. Incorrect.
Option 3 is incorrect. Simply distributing embeddings into multiple tables by category and relying on row-level scans does not leverage Oracle's specialized indexing for vector data. This approach is less efficient for large-scale similarity searches.
- D. Incorrect.
Option 4 is incorrect. Partitioning by product ID does not specifically address multi-dimensional similarity queries. While partitioning can help with manageability and certain performance aspects, it does not replace the need for a specialized vector index.