The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Building the Product Knowledge Graph at Amazon with Luna Dong - #457
27 snips
Feb 18, 2021 In this engaging discussion, Luna Dong, Sr. Principal Scientist at Amazon, shares her expertise in product knowledge graphs and their role in enhancing search and recommendation systems. She explores the integration of machine learning and discusses the challenges of data curation. Luna delves into the unique dynamics of media versus retail knowledge graphs and the necessity of human intervention for data quality. The conversation also highlights practical applications of knowledge graphs and the importance of standardization across the industry.
AI Snips
Chapters
Transcript
Episode notes
Project Initiation
- Luna Dong joined Amazon and initiated the product knowledge graph project.
- This project covers media (books, music, movies) and retail products, along with web knowledge extraction.
Media vs. Retail Data
- Media product data is readily available from publishers as structured metadata.
- Retail product data requires information extraction from unstructured text and images.
Model Training
- Knowledge extraction models are trained using manual labels and existing "seed knowledge."
- Weak learning helps generate training data, mirroring human learning.