1Z0-184-25 Question 86
Select 2You are designing a knowledge management system on Oracle Cloud Infrastructure that ingests large collections of text documents from multiple business units. Each document is split into sections (title, abstract, body) for finer granularity, and you want to store multiple vector embeddings per section (e.g., one embedding for key terms, another for semantic context) so that searches can leverage different facets of the text. You decide to use Oracle Database 23c� vector data type for multi-vector similarity search. Which two steps should you implement to properly enable multi-vector similarity search across all documents and sections?
- A
A. Store each embedding in a dedicated vector column, one column per embedding type, and configure an index on each vector column.
- B
B. Convert all embeddings into a single concatenated text string and store them in a VARCHAR2 column for direct text matching.
- C
C. Enable approximate nearest neighbor (ANN) indexing on each embedding vector column to optimize similarity queries.
- D
D. Use a single numeric column to store all embeddings and rely on manual partitioning queries to find the closest vectors.
Show answer and explanation
Correct answers: A, C
Explanation
To implement multi-vector similarity search in Oracle Database 23c, each type of embedding (e.g., one for keywords, another for semantic context) should have its own vector column. You then enable approximate nearest neighbor (ANN) indexing on each vector column so that the database can efficiently find similar vectors at query time. Refer to Oracle Database 23c documentation and best practices for working with the vector data type and ANN indexes.
- A. Correct.
A. Correct: Each embedding type should be stored in its own vector column so that you can independently index and query them. Oracle Database 23c supports a vector data type that allows storing and indexing vectors natively.
- B. Incorrect.
B. Incorrect: Simply concatenating embeddings into a text field loses the structure and numeric properties that are essential for vector operations like cosine similarity. This would prevent efficient multi-vector similarity search.
- C. Correct.
C. Correct: After storing the embeddings in vector columns, enabling approximate nearest neighbor (ANN) indexing is essential for real-time similarity queries. This drastically speeds up lookups for the most similar vectors.
- D. Incorrect.
D. Incorrect: Storing all embeddings in a single numeric column and manually partitioning them is inefficient and prone to errors. Oracle Database 23c provides native support for vector indexing, which is far more efficient than manual partitioning.