Journey Towards Deep Learning for Ranking
summary
GetYourGuide replaces an XGBoost ranker with a PyTorch transformer and deploys it within nine months while keeping high throughput and low latency. The migration tests 50+ offline models and runs 10+ live A/B experiments, yields a ~2.2% ranking metric lift plus bookings and revenue gains from a deep‑and‑cross network feeding a ranking transformer, and relies on scikit‑learn preprocessing, a feature store, and GPU training on Databricks. Engineering headaches include multi‑GPU embedding bugs, CI/docker bloat, and Torch CPU inference tuning.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 0:34Who are you
- 0:53Agenda
- 2:03What is Get Your Guide
- 5:35Learning to Rank
- 6:31Ranking Workflow
- 8:39Evaluating Ranking Models
- 10:17ML Model
- 11:37Why Deep Learning
- 12:39Deep Learning
- 12:54Expectations
- 15:30First Milestone
- 17:38Second Milestone
- 21:06Ranking Model
- 24:53Ranking Transformer
- 28:00Summary
related talks
29mFrom Trees to Transformers: Our Journey Towards Deep Learning for Ranking
38mLearning to Rank journey at GetYourGuide: Our Logbook #bbuzz
29mLearning to rank with the Transformer
25mBetter search relevance using Learning to Rank at mobile.de
29mRanking hotel images using deep learning
33mBuild Better AI with First-Party Data and Fine-Tuning
30mScaling Machine Learning at Discord: Our Journey with Ray and Anyscale
17mLaying the foundation for deep learning