# Semantic Product Recommendation

## Definition

Semantic product recommendation recommends products according to the meaning and characteristics of products rather than relying exclusively on observed user behavior.

## Inputs

Possible inputs include:

- product title;
- description;
- attributes;
- category;
- brand;
- technical specifications;
- images;
- product hierarchy.

## Embeddings

Text or image representations can be converted into vectors. Vector similarity can identify products that are close in semantic or visual space.

## Beyond similarity

Similarity is not the same as complementarity. A recommendation engine can also infer whether another product completes the use case or purchase intent of the source product.

## RecoKit

RecoKit uses semantic/catalog information as a foundation for product recommendations and can combine semantic, visual, intent and optional behavioral signals.

## Related

- `embeddings.md`
- `product-similarity.md`
- `product-complementarity.md`
- `../relationships/recokit-semantic-recommendation.md`