1Z0-184-25 Question 80
Select 2You are designing a large-scale recommendation system on Oracle Database 23c in OCI to store product features as vector embeddings. You plan to perform approximate similarity searches to quickly retrieve the most similar products for a given embedding. Which two statements represent best practices for creating and querying approximate vector indexes to achieve both high performance and accurate search results?
- A
You must specify the vector dimension that matches your embedding size and the distance metric (e.g., L2 or cosine) when building the approximate vector index.
- B
Because approximate vector indexes do not support incremental updates, you must drop and recreate the index whenever new embeddings are inserted.
- C
Tuning index creation parameters (e.g., EF Construction) and query parameters (e.g., EF Search) can significantly influence the trade-off between query latency and similarity accuracy.
- D
The approximate vector index supports using standard SQL equality (=) and range (<, >) operators on the vector data for precise filtering.
Show answer and explanation
Correct answers: A, C
Explanation
Oracle Database 23c introduces vector data type and indexing features to enable fast approximate nearest neighbor searches, ideal for embeddings used in recommendation systems or similarity-based applications. The dimension of the stored vectors and the distance metric must be specified during index creation to ensure proper similarity comparisons. Index parameters (such as EF Construction) and query parameters (like EF Search) can be tuned based on the desired trade-off between performance and accuracy. For details, consult the official Oracle Database 23c documentation on vector search.
- A. Correct.
Correct. When creating an approximate vector index, you must specify the dimension that matches the size of your embedding vectors and choose an appropriate distance metric, such as L2 (Euclidean) or cosine similarity. This ensures that the index is built correctly and can effectively measure similarity.
- B. Incorrect.
Incorrect. Oracle� vector search implementation typically supports incremental updates without requiring a full drop and recreate of the index. While reindexing might be advisable after large data ingestions for optimal performance, smaller inserts or updates do not automatically require a complete index rebuild.
- C. Correct.
Correct. Parameters such as EF Construction (which affects index build) and EF Search (which affects query recall) can be tuned to balance query latency and result accuracy. Adjusting these parameters allows you to optimize performance based on your application� needs.
- D. Incorrect.
Incorrect. Approximate vector indexes are specialized for similarity or nearest neighbor queries, rather than exact matches or typical numeric range scans. Standard SQL operators like '=' or '<' and '>' do not apply directly to vector columns in this context.