1Z0-184-25 Question 17
Single answerYou are using Oracle Cloud Infrastructure 2025� integrated vector search capabilities to build a text-based recommendation engine. Your embeddings primarily capture the direction of the language features rather than their magnitudes. Which vector distance metric is typically most suitable for ranking semantically similar documents in this scenario?
- A
Cosine distance
- B
Euclidean distance
- C
Jaccard distance
- D
Gaussian kernel distance
Show answer and explanation
Correct answer: A
Explanation
In Oracle Cloud Infrastructure 2025 Architect Professional scenarios involving semantic vector search (especially for text embeddings), Cosine distance tends to be the metric of choice. Cosine distance emphasizes how similar the directions of the embeddings are, disregarding differences in magnitude�which generally aligns better with how language embeddings capture semantic meaning. For more details, refer to Oracle� AI and ML best practices for vector-based semantic search in the Autonomous Database documentation.
- A. Correct.
Cosine distance is correct because it focuses on the angle between vectors, making it well-suited for semantic embeddings where direction is more important than magnitude. This is especially common in language models where similar terms tend to be close in direction but may vary in length or magnitude.
- B. Incorrect.
Euclidean distance is a common metric but often less effective for high-dimensional text embeddings, because magnitude differences can overshadow subtle directional similarities in semantic space.
- C. Incorrect.
Jaccard distance measures the similarity between finite sets, making it more appropriate for sets of discrete tokens rather than continuous vector embeddings. It is not typically used for semantic vector similarity.
- D. Incorrect.
Gaussian kernel distance is more of a kernel function approach utilized in certain machine learning methods (e.g., kernel SVMs). It� not the standard metric for semantic similarity in vector search and is more computationally complex to implement in a large-scale semantic search scenario.