1Z0-184-25 Question 51
Select 2You have created an HNSW vector index on an Oracle Autonomous Database table containing millions of product review embeddings to enable semantic search. You notice that, although the search returns relevant results quickly, some highly relevant reviews are missing. You want to improve recall while still maintaining acceptable query latency. Which two actions should you take to achieve this? (Choose two.)
- A
Increase the 'ef' parameter used at query time to expand the candidate pool.
- B
Decrease the 'M' parameter in the HNSW index configuration to reduce graph connectivity.
- C
Store all embeddings in a BLOB column without the HNSW index to reduce overhead.
- D
Increase the 'M' parameter for your HNSW index to provide a richer neighborhood graph.
- E
Set the 'ef' parameter to 1 to minimize computational complexity.
Show answer and explanation
Correct answers: A, D
Explanation
HNSW indexes rely on parameters such as 'M' (graph connectivity) and 'ef' (candidate search size) to balance recall and latency. By increasing 'M' and 'ef' in a measured way, you enable broader exploration of neighbors during queries, improving recall. For detailed guidance, refer to Oracle Autonomous Database documentation on configuring vector search indexes, which emphasizes testing different parameter values to find the optimum trade-off between speed and accuracy.
- A. Correct.
Correct. Increasing 'ef' (the size of the dynamic candidate list searched during queries) typically improves recall by exploring more potential neighbors. However, it can also slightly increase query latency, so balancing 'ef' is key.
- B. Incorrect.
Incorrect. Decreasing 'M' reduces the number of connections in each layer of the HNSW graph, which may negatively impact both recall and search quality.
- C. Incorrect.
Incorrect. Storing embeddings without an index simply disables the vector search capability and would drastically increase query times as you'd lose approximate nearest neighbor functionality.
- D. Correct.
Correct. Increasing 'M' boosts the connectivity in the HNSW graph, allowing more potential matches to be considered. This generally yields higher recall but involves a trade-off with index size and build time.
- E. Incorrect.
Incorrect. Setting 'ef' to 1 severely limits the candidate pool to explore, which is almost guaranteed to reduce recall to unacceptable levels.