1Z0-184-25 Question 90
Single answerYour marketing team plans to categorize and search through thousands of product-related documents stored in an Oracle Autonomous Database. They generate multiple vector embeddings per document segment (e.g., capturing different semantic aspects like topics, sentiments, and product features) and need to query across these vectors simultaneously for multi-document similarity search. Which approach best ensures efficient multi-vector similarity lookups in this scenario?
- A
Store all vectors in a single column as a concatenated array and query them using a single dimensional index.
- B
Place distinct embeddings in separate columns and create a multi-vector index that references each embedding column for similarity search.
- C
Organize embeddings for each document into multiple rows�one row per embedding�and utilize standard B-tree indexes for each row.
- D
Retain basic text indexing for all documents and run string-based queries to approximate semantic similarity.
Show answer and explanation
Correct answer: B
Explanation
For multi-vector similarity search in Oracle Autonomous Database (and newer Oracle Database releases offering built-in vector datatypes), it is a best practice to store each embedding in its own column and employ a specialized multi-vector index. This approach enables efficient similarity calculations across multiple embedding vectors. Refer to Oracle's documentation on Vector Search in Oracle Database (23c and later) to learn about defining vector columns, enabling advanced indexing, and performing multi-vector similarity queries.
- A. Incorrect.
Storing all vectors as a single concatenated array usually leads to less precise or more complex queries, because the database cannot independently compare each embedding dimension. It complicates multi-vector lookups and often reduces performance.
- B. Correct.
This is correct. Creating separate columns for each embedding (reflecting different semantic aspects) and applying a multi-vector index strategy allows the database's vector search engine to efficiently handle simultaneous similarity comparisons. It leverages native vector indexing to handle multiple embeddings per document segment.
- C. Incorrect.
Storing each embedding as a separate row would fragment the data, making the query logic more complex. B-tree indexes are not ideal for vector similarity operations because vectors are inherently multidimensional, and B-trees do not efficiently support multi-dimensional distance calculations.
- D. Incorrect.
Text-only indexes focus on keyword-based matching rather than semantic matching of embeddings. They are insufficient for accurate multi-vector similarity searches, which rely on measuring distances or similarities across vector spaces.