A Generative AI Engineer is deciding between using LSH (Locality Sensitive Hashing) and HNSW (Hierarchical Navigable Small World) for indexing their vector database. Their top priority is semantic…

Question

Options

• ACompare the cosine similarities of the embeddings of returned results against those of a
• BCompare the Bilingual Evaluation Understudy (BLEU) scores of returned results for a
• CCompare the Recall-Oriented-Understudy for Gisting Evaluation (ROUGE) scores of returned
• DCompare the Levenshtein distances of returned results against a representative sample of test

• A
  77% (20)
• B
  4% (1)
• C
  8% (2)
• D
  12% (3)

Explanation

Semantic accuracy in a vector database context measures how semantically similar the retrieved documents are to the query. Cosine similarity between embeddings is the canonical metric for this: it quantifies the angular closeness of two vectors in the embedding space, directly reflecting semantic relatedness. Comparing cosine similarities of results from LSH vs. HNSW against a ground-truth set reveals which algorithm retrieves more semantically relevant documents. BLEU (B) measures n-gram overlap for machine translation quality. ROUGE (C) measures recall-oriented n-gram overlap for summarization quality. Levenshtein distance (D) measures character-level edit distance between strings, which is a surface-level string metric with no connection to semantic meaning.

No community discussion yet for this question.