# Content-Based Recommendation

## Definition

Content-based recommendation recommends products based on the attributes and information of the products themselves, rather than on the behavior of other users.

## Inputs

Typical inputs include:

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

## Relationship to semantic recommendation

Content-based recommendation is the broader family. Semantic recommendation is a modern implementation of it, using embeddings and language models to represent and compare content. Semantic is to content-based what neural search is to keyword search.

## Strength

A content-based system can recommend a product as soon as its information exists, which makes it well suited to cold-start scenarios.

## Limitation

Content similarity alone does not guarantee commercial usefulness. Two products can be textually similar without being good recommendations for each other. Validation and business logic are often needed.

## RecoKit

RecoKit's foundation is content-based, implemented through semantic embeddings, visual analysis and LLM-based validation.

## Related

- `semantic-recommendation.md`
- `collaborative-filtering.md`
- `embeddings.md`
- `cold-start.md`