Improving product discovery via Hierarchical Recommendations!
summary
Describes Flipkart’s move from recommending individual products to recommending product collections for fast-moving Lifestyle categories, using a hybrid relevance algorithm that combines collaborative filtering and content-based methods. Explains how this hierarchical recommendation approach addresses cold-start and discovery problems, increases basket size and cross-category discovery, and how it is implemented to scale against a ~400M-product catalogue.
AI-written summary · about
chapters (16)
- 0:00Intro
- 0:31What should be recommended ?
- 2:23Overview
- 3:30Content & Collaborative Filtering - The Standard Approach
- 4:53Motivation
- 8:00Taxonomy Structure
- 8:54Computing Relevance via Cosine Similarity
- 10:02Content Matching - A Product Level Approach
- 11:08Entity Matching - A Category Based Approach
- 12:23Cluster Matching - Done at a Cluster Level
- 13:51Cluster Generation
- 16:43Design Considerations
- 19:06Metrics to measure Improvements
- 21:16Approaches Recap
- 21:45Comparison of Approaches - Sparsity vs Coverage
- 22:33Summary - Problem, Approaches Adopted, Metrics
related talks
41mE-Commerce search at scale on Apache Lucene
20mDesigning for Big Data
21mBuilding a Progressive Web App
18mExpanding Omnichannel Retail Horizon: Headless Commerce
01h 15mSocial Media Recommender Systems Should Deliver Value, not “User Engagement"...
47mRust At The Edge
21mBecoming the Multi-armed Bandit
51mHow AWS Helps You Improve the Customer Shopping Experience