Personalized Recommendations

Personalized recommendations surface products, videos, listings, or content that match a user's past behavior and stated or inferred preferences.

Business impact

• Share of homepage video clicks attributed to recommendations — YouTube reports recommendations account for about 60% of all video clicks from the homepage.
• Consumer engagement — Trulia reports a double-digit increase in consumer engagement.
• Click-through-rate on recommended similar property listings — Airbnb reports a 21% increase in click-through-rate on recommended similar property listings.
• Guests who discovered their listings — Airbnb reports 4.9% more guests discovered their listings.
• Customer base — Gousto reports its customer base grew to 7 times its previous size while using personalization and forecasting.

Data requirements

• Past user activity (tabular) — Used to infer preferences from watched, liked, rated, favorited, purchased, or searched items.
• Related-item signals (tabular) — Used to connect items that tend to be viewed or bought together.
• Search behavior (tabular) — Used by property and shopping recommenders to tailor results to intent.
• Item metadata (tabular) — Gousto uses recipe attributes such as cuisine, cooking method, cooking time, and ingredients.
• Listing or product images (image) — Trulia uses images of listings and their content to support recommendation and photo selection.

AI methods and techniques

• Recommendation systems — Used to rank and surface items for each user based on preference and behavior signals.
• Collaborative filtering — Used by Amazon and Trulia to match users to similar items or listings.
• Association rules — YouTube combines related-videos association rules with user activity to compute recommendations.
• Machine learning — Used across the cases to infer preferences, rank items, and update suggestions from behavior data.
• Embedding-based modeling — Airbnb uses an embedding-based approach to represent listings and improve personalized search results.
• Random forest classification — Gousto uses a random forest model to predict recipe uptake.

AI models and model families

HyperCLOVA

Industries

Real-world evidence

8 documented case studies on record.

Companies using this: Naver, YouTube, TGI Fridays, Amazon, Trulia, Woven (fka Everyday Agents), Airbnb, Gousto.