What recommender should I use for a marketplace with sparse data?

For a marketplace with sparse data, the best recommender is usually a hybrid recommender, not a pure collaborative-filtering model.

Good default choice

Hybrid = content-based + collaborative + popularity/freshness fallback

Why:

• Collaborative filtering struggles when users/items have too few interactions.
• Content-based works better for new or sparse items because it uses item attributes.
• Popularity / trending / recent activity gives reliable recommendations when personalization is weak.

Practical recommendation stack

1. Cold-start / sparse users

   • Recommend:
     • popular items in the user’s category/region
     • recently active listings
     • items matching profile signals or session behavior

2. Sparse items

   • Use item metadata:
     • title, description, category, brand, price, location, tags
   • Build embeddings from text + structured features

3. Once you have enough interaction data

   • Add collaborative filtering or a two-tower model
   • A two-tower retrieval model is often strong for marketplaces at scale
   • For ranking, use a learning-to-rank model on top

If you want a simple starting point

Start with:

• content-based retrieval
• plus popularity/trending
• plus business rules
• then gradually add collaborative signals

If your marketplace is very sparse

Use:

• matrix factorization with side information
• factorization machines / field-aware factorization machines
• two-tower model with metadata
• hybrid ranker

Rule of thumb

• Very sparse / new marketplace: content-based + popularity
• Moderately sparse: hybrid model
• Lots of interaction data: collaborative filtering or two-tower + ranking

If you want, I can recommend a model based on your exact marketplace type, like e-commerce, jobs, rentals, services, or classifieds.