Machine Learning: The Engine Behind Modern E-commerce Search
summary
E‑commerce search moves from brittle keyword matching to embeddings and semantic retrieval that recover synonyms and context, with query classification into direct versus exploratory pipelines and entity extraction (material, style, price) to apply filters. LLMs are used for LLM embellishment of poor product text and images, producing a reported ~30% search lift. Results then use a cross-encoder reranker plus engagement signals for personalization; two‑tower or numeric feature models handle many near‑identical SKUs. Smartcat work with vintage Discovery supports these claims; metrics cited: ~60% land on search and ~80% drop without fixes.
AI-written summary · about
chapters (10)
- 0:00ML-powered e-commerce search
- 1:59Bad search loses customers
- 3:03Fixing retrieval with semantic search
- 5:57Improving product descriptions with LLMs
- 7:44Query classification for pipelines
- 10:33Breaking down queries with entity extraction
- 12:23Reranking results for relevance
- 14:56Personalizing search with signals
- 16:45Key takeaways for ML integration
- 18:57Q&A
related talks
20mBeyond Relevance: Using LLMs to Transform E-commerce Search into Product Discovery Engines
39mAI-Powered Search in Modern E-Commerce
33mAI-Powered Search in Modern E-Commerce
56mMachine Learning in E‑commerce
31mAdvanced search for your e-commerce website
22mAI-Powered Personalized Discovery with Graph Search and VLM-Enhanced Embeddings
30mSearch on Steroids — A Story of Semantics
47mModern Approaches to Product Matching Using LLMs