# Semantic Similarity

# Semantic Similarity

Semantic similarity answers “how related are these two items by meaning?” even when they do not share keywords.

In Curiosity Workspace, semantic similarity is usually powered by:

• embeddings (vector representations)
• vector indexes for fast nearest-neighbor lookup

# Common use cases

• “Show similar cases/documents”
• “Find duplicates or near-duplicates”
• “Recommend related entities”
• “Cluster items into themes”

# Similarity in context

Similarity is most useful when constrained by context, such as:

• only compare within a given customer or project
• only compare within a product line
• only compare within a time window

You can enforce context via:

• facets (property filters)
• graph traversal (related-to constraints)

# Practical tuning tips

• Choose the right field(s) to embed (often Summary + first message, or Title + body).
• Use chunking for long fields.
• Tune cutoffs so “similar” is meaningful for your domain.
• Evaluate with known pairs (positive and negative examples).

# Next steps

• Configure embeddings: NLP → Embeddings
• Apply similarity to retrieval: Search → Vector Search