1Z0-184-25 Question 92
Single answerYou are designing a semantic search feature to find similar articles in a large knowledge base stored in an Oracle Autonomous Database on Oracle Cloud Infrastructure. The system must handle frequent updates as new articles arrive and still deliver low-latency queries on vector embeddings. Which approach should you use to ensure high performance and maintain accurate similarity results over time?
- A
Store embeddings as plain VARCHAR columns and use a standard B-Tree index.
- B
Use Oracle Database 23c� vector data type alongside a specialized vector index.
- C
Save all embeddings as JSON documents on OCI Object Storage and retrieve them through batch jobs.
- D
Keep embeddings in a relational table with primary keys only, relying on row-by-row lookups.
Show answer and explanation
Correct answer: B
Explanation
Oracle Database 23c supports a dedicated vector data type and specialized indexes to handle semantic search requirements (nearest-neighbor lookups, similarity queries) at scale. By using these native capabilities within Oracle Autonomous Database on OCI, you can maintain low query latency even as new articles and their corresponding embeddings are frequently added. Refer to Oracle� official Database 23c documentation for details on vector indexing and best practices for implementing semantic search solutions.
- A. Incorrect.
Plain text (VARCHAR) with a B-Tree index does not support vector similarity operations. This approach makes embedding distance computations cumbersome and negatively affects query performance.
- B. Correct.
Oracle Database 23c introduces a native vector data type and indexing capabilities designed for efficient similarity (e.g., nearest-neighbor) queries on embeddings. This is the recommended approach for semantic search solutions that rely on vector operations.
- C. Incorrect.
Storing embeddings as JSON in Object Storage and processing them via batch jobs will not meet low-latency requirements because it introduces significant overhead with no native vector indexing. This is more suited for offline analytics rather than real-time semantic search.
- D. Incorrect.
Relying solely on a primary key approach would not allow for similarity or distance-based queries on embeddings. It forces serial lookups and cannot leverage specialized vector operations.