Semantic Search Personalisation — Results Ranked by Preferences

I will build a personalised semantic search system — combining semantic relevance with user preference signals (purchase history, browsing behaviour, saved items, and explicit preferences) to rank search results that are both semantically relevant and personally appropriate for each user. Personalised search delivers results that would both satisfy the semantic query and align with the individual user's preferences — a search for "running shoes" returns trail running shoes for a user whose history shows trail running interest and road running shoes for a road runner, both semantically relevant but personally calibrated.

The system covers user preference embedding generation from behavioural signals, hybrid scoring combining semantic similarity and preference alignment, cold-start handling for new users, preference drift management as user interests evolve, and A/B testing for personalisation impact measurement.