Embeddings Similarity
Dense embedding-based item similarity for semantic recommendation
Embeddings Similarity uses pre-computed dense item embeddings (e.g., from a language model or product encoder) to find semantically similar items. Recommendations are based on nearest neighbor search in the embedding space.
When to use:
- Semantic similarity-based recommendation (articles, products, media)
- When item embeddings from a pre-trained model are available
- Cross-modal recommendations (e.g., text query → image recommendations)
Input: User interaction history or a query item; item embedding vectors from training Output: Ranked list of semantically similar items
Model Settings (set during training, used at inference)
Embedding Columns (set during training) Which columns contain pre-computed item embedding vectors.
Distance Metric (default: cosine)
Similarity metric for nearest neighbor search. cosine is standard for embedding spaces.
N Recommendations (default: 10) Number of nearest neighbors to return.
Inference Settings
No dedicated inference-time settings. Recommendations are retrieved via approximate nearest neighbor search in the embedding space.