Personalization of Search Ranking in E-commerce
Search engine without personalization shows same results to all users. ML ranking accounts for view history, purchases, returns and session context — and rearranges results individually. Win: +8-15% to search conversion.
Learning-to-Rank Architecture
LambdaMART (LightGBM ranker) for personalized search. Trains on implicit feedback: clicks, purchases, view time. Personalized search especially effective for head queries (top-20% queries give 80% traffic). For tail queries (rare), semantic search via vector index more important than personalization.
Typical metric wins: CTR +12%, Conversion Rate +8%, Revenue per Search +10% with correct feature engineering.