Search Logs + Machine Learning = Auto-Tagged Inventory
summary
Eventbrite builds an auto-tagging pipeline that turns search+click logs into training labels, using a scikit-learn pipeline to vectorize titles/descriptions and a simple neural classifier to predict queries. It clusters related queries with affinity propagation to collapse synonyms and misspellings, reducing hundreds or thousands of queries to exemplars, and trains on the top queries (about 500k unique queries overall, top ~2,000 cover ~50% of traffic) achieving useful tags and ~50% coverage of search traffic.
AI-written summary · about
chapters (31)
- 0:00Intro
- 1:24What is tagging
- 4:27Why do you want tagging
- 7:39The problem with tags
- 9:28The initial approach
- 11:29Getting the data
- 12:59Converting the data
- 14:14Scikitlearn pipeline
- 17:09Results
- 21:29Test Set
- 24:04Improved Results
- 25:29Improved Results Notebook 2
- 26:14Job Fair
- 27:19Normalized Edge
- 27:59Adjacency Matrix
- 29:04Big Mistake
- 29:19Spelling Correction
- 29:44Clustering
- 30:34Ingredient Data
- 31:24Data Set
- 31:34Query Results
- 32:29Matrix Multiplication
- 33:19Query Terms
- 34:29Drumroll
- 34:59Afro Caribbean Wine Fest
- 35:59Open Mic Marketing
- 36:44Improvements
- 37:44Drawbacks
- 38:49Tag related applications
- 39:49Better coverage
- 40:44Summary
related talks
22mMachine Learning: The Engine Behind Modern E-commerce Search
34mClassifying Search Queries Without User Click Data
55mSimple Machine Learning
39mAI-Powered Search in Modern E-Commerce
41mText to Context: How We Introduced Hybrid Search
35mIndustrial Applications of Machine Learning in Search
31mSponsored Session: From Clicks to Rankings: Building a Learning-to-Rank Pipeline with…
01h 30mBecome a data-driven Organization with Machine Learning