Learning to Rank: where search meets machine learning #bbuzz
summary
Introduces Ranking SVM applied to web search: trains on pairwise training examples derived from click logs, learns feature weights (tfidf, metadata, CNN image embeddings) and scores items by a weighted sum. Recommends engineering options: offline batch scoring in Hadoop/Spark or caching model weights and doing a top-K rerank at query time. Covers practicalities—position bias, feedback loops, NDGC evaluation—and notes the method scales and is a good starting point.
AI-written summary · about
chapters (9)
- 0:00Introduction to Learning to Rank
- 3:34Traditional search relevance ranking
- 8:09Feature engineering for LTR
- 12:48Query-specific models
- 13:09Training with SVMs
- 20:57Pairwise preference training
- 21:19Production deployment strategies
- 25:34Avoiding feedback loops
- 29:25Further resources and Q&A
related talks
55mLearning to Rank, a Little
24mLearning "Learning to Rank"
31mSponsored Session: From Clicks to Rankings: Building a Learning-to-Rank Pipeline with…
26mTeaching Ruby to Rank: A Search Pipeline Story
47mLearning to rank search results
37mSearch and Recommendations: 3 Sides of the Same Coin #bbuzz
43mLearning to Rank for Reddit Search: A Project Retro
39mRanking Listings with Machine Learning and ElasticSearch