1Z0-184-25 Question 78
Single answerAn online retailer wants to add a 'similar items' feature to its product catalog. They have stored product image embeddings as vectors in an Oracle Autonomous Database (supporting vector columns) within OCI. The goal is to achieve sub-second response times for finding the top nearest neighbors based on visual similarity. Which approach best meets their needs for performing an approximate similarity search using vector indexes?
- A
Build a B-tree index on the vector columns for fast similarity lookups.
- B
Use Oracle Text indexes to search for similar embeddings by matching keywords.
- C
Implement an approximate nearest neighbor (ANN) vector index (e.g., HNSW) on the vector columns.
- D
Execute a custom external MapReduce job to calculate similarities in batches.
Show answer and explanation
Correct answer: C
Explanation
To support a real-time 'similar items' feature using product embeddings, you need an index type optimized for vector similarity, often referred to as an approximate nearest neighbor (ANN) index. In Oracle Cloud Infrastructure, an ANN algorithm like Hierarchical Navigable Small World (HNSW) can be used, and you can create a vector index on the stored embeddings. This correlation greatly improves query performance compared to traditional indexes or batch-based processing. For official guidance, refer to Oracle's documentation on performing vector searches in the Oracle Database (23c or later) and Autonomous Database.
- A. Incorrect.
B-tree indexes are not designed for high-dimensional proximity searches and would not efficiently handle vector similarity queries.
- B. Incorrect.
Oracle Text focuses on linguistic and keyword-based matching, which is not suitable for vector-based similarity.
- C. Correct.
Approximate nearest neighbor (ANN) vector indexes such as HNSW are specifically optimized for large-scale vector similarity queries, balancing accuracy and performance.
- D. Incorrect.
A custom MapReduce approach would be inefficient for real-time queries, as it focuses on batch processing rather than low-latency similarity lookups.