All You Need to Know to Build a Product Knowledge Graph (KDD 2021 Tutorial)
knowledge-graphproductecommerceinformation-extractionontologyamazon
Abstraction: Industry best practices for building scalable e-commerce product knowledge graphs
Key points:
- Product KGs are harder than generic KGs due to highly specific domains, sparse training data, and constantly evolving taxonomies
- Pipeline covers: ontology/taxonomy construction, knowledge extraction (text + multimodal/image), knowledge cleaning and quality control, and applications
- Presenters from Amazon Product Graph team; Xin Luna Dong previously led Google Knowledge Vault and Amazon Product Graph efforts
- Information extraction uses both text-based NLP and multimodal signals (product images) to extract structured attributes
- Industry deployments require very high precision with minimal recall loss — data cleaning and quality control are emphasized
- KG supports downstream search, recommendation, and question answering in e-commerce platforms
Connections: Amazon · Knowledge Graphs · Information Extraction · Ontology
Source: https://naixlee.github.io/Product_Knowledge_Graph_Tutorial_KDD2021/