1Z0-184-25 Question 88
Single answerYou are designing a multi-document retrieval system on Oracle Cloud Infrastructure (OCI) to handle diverse documents stored in an Autonomous Database. Your team wants to perform multi-vector similarity searches across different embeddings (e.g., text-based and metadata-based) for more accurate results. Which approach best ensures efficient multi-vector similarity searches when indexing and querying multiple document embeddings?
- A
Store each document� text-based and metadata-based embeddings separately in dedicated vector columns and perform combined queries using vector similarity functions.
- B
Create a single merged embedding by concatenating text-based and metadata-based vectors into a single vector column for simplified storage.
- C
Use a traditional B-Tree index on a combined text and metadata field, then manually filter out dissimilar results in application code.
- D
Rely solely on text-based keyword indexing in the Autonomous Database, because metadata-based embeddings often reduce indexing performance.
Show answer and explanation
Correct answer: A
Explanation
When planning multi-document search with multi-vector embeddings, treating each embedding separately and combining their respective similarity scores is considered a best practice. In OCI, Autonomous Database supports vector data types and similarity queries, allowing multi-vector approaches to be handled effectively. For further details, refer to Oracle documentation on vector support and best practices for semantic search in Autonomous Database.
- A. Correct.
Correct. Storing each embedding in its own vector column and using a multi-vector query ensures you can calculate similarity across different types of embeddings. Autonomous Database supports vector storage and queries, enabling you to combine similarity scores from multiple embeddings efficiently.
- B. Incorrect.
Incorrect. While merging embeddings can simplify storage, it can also dilute the unique characteristics of each embedding type, lowering the quality of similarity results and making maintenance more complex.
- C. Incorrect.
Incorrect. A traditional B-Tree index is not designed for semantic similarity queries on high-dimensional vectors, leading to inefficient retrieval and high application-side overhead.
- D. Incorrect.
Incorrect. Relying only on text-based indexing is a common misconception. Excluding metadata-based embeddings can reduce overall search accuracy, and modern vector indexes can handle embeddings without significantly harming performance.