1Z0-184-25 Question 91
Select 2You are building an advanced customer support system on Oracle Cloud Infrastructure (OCI). The system ingests thousands of product reviews and support tickets across multiple domains (software, hardware, etc.) and in several languages. You use multi-vector similarity to incorporate both domain-specific and language-based embeddings for more accurate retrieval. The data is stored in Oracle NoSQL Database with vector indexing, and you want to retrieve the most relevant documents in near real-time. Which two actions should you take to ensure an effective multi-vector similarity search that scales with your data?
- A
Use a single fixed dimension for all domain-specific embeddings, ignoring language differences.
- B
Leverage Oracle NoSQL Database� vector indexing to store and query multiple embedding dimensions for each document.
- C
Migrate your entire corpus to Oracle Autonomous Data Warehouse for additional relational indexing.
- D
Include domain or language metadata in the query filters to improve search relevance.
- E
Duplicate each document per domain to ensure coverage across different embedding vectors.
Show answer and explanation
Correct answers: B, D
Explanation
A multi-vector similarity approach benefits from storing multiple embeddings (e.g., domain-specific, language-based) in a vector-friendly database like Oracle NoSQL Database. By leveraging its vector indexing capabilities, you can efficiently query high-dimensional embeddings. Adding metadata such as domain or language tags further improves recall and precision by filtering out irrelevant documents before or during the similarity search. For more details, consult Oracle NoSQL Database documentation regarding vector indexing and Oracle Cloud Architecture Center guidelines on designing scalable multi-vector search solutions.
- A. Incorrect.
Incorrect. Using a single fixed dimension for all embeddings ignores the differences in domains and languages. This can reduce the accuracy of semantic matching since embeddings for different contexts often have different dimensions or underlying distributions.
- B. Correct.
Correct. Oracle NoSQL Database supports vector indexing, allowing you to store and query multi-dimensional embeddings. This is essential for multi-vector similarity search, enabling the retrieval of documents based on several embedding vectors (e.g., domain-specific and language-based) without losing performance.
- C. Incorrect.
Incorrect. Merely migrating data to Oracle Autonomous Data Warehouse for relational indexing does not address the need for semantic or vector-based similarity. While ADW might help for analytic queries, it� not specifically designed for multi-vector similarity search on embedded data.
- D. Correct.
Correct. Incorporating metadata filters (such as domain or language tags) refines the similarity search results, ensuring only the most relevant documents are returned. This also helps reduce search space and improves performance in multi-vector retrieval scenarios.
- E. Incorrect.
Incorrect. Duplicating documents per domain increases storage costs and operational overhead. With vector indexing, you can store different embeddings or metadata in a single location and still achieve effective similarity search, avoiding unnecessary data redundancy.