What’s a good recommendation engine for new items and low traffic?

For new items + low traffic, you usually want an engine that works well with little or no interaction history. The best choice is typically a content-based or hybrid recommender, not a pure collaborative filtering system.

Good options

1) Content-based recommendation

Recommend items similar to what the user has viewed/bought using item features like:

• category
• brand
• price
• text description
• embeddings from images or text

Best for: cold-start items and sparse traffic
Pros: works immediately for new items
Cons: can feel repetitive / less “discovery” focused

---

2) Hybrid recommendation

Combine:

• content-based signals
• popularity/trending
• collaborative filtering when enough data exists

Best for: low traffic marketplaces or catalogs with many new items
Pros: handles cold start better than pure CF
Cons: more engineering complexity

A common approach is:

• new items → content similarity + editorial/popularity boost
• active items → collaborative filtering / matrix factorization
• fallback → trending, popular in category, or personalized by demographics/context

---

3) Retrieval + ranking with embeddings

Use vector embeddings for users/items and retrieve nearest neighbors.

Best for: catalogs with rich text/images and not much interaction data
Pros: strong cold-start performance if item metadata is good
Cons: needs good feature generation and tuning

---

Practical recommendation

If your traffic is low and items are frequently new, I’d recommend:

Start with a hybrid system:

1. Build item embeddings from metadata/text/images
2. Use content similarity for new items
3. Add popularity/trending as a fallback
4. Once you have enough interactions, layer in collaborative filtering
5. Rank results with business rules and freshness boosts

---

If you want a simple stack

• Small/simple: content-based + popularity
• Medium: hybrid with embeddings + rules
• Advanced: two-stage recommender (candidate retrieval + learning-to-rank)

If you tell me your domain (e-commerce, jobs, content, marketplace, etc.) and what data you have, I can suggest a more specific architecture.