AutoKnow: Self-Driving Knowledge Collection for Products of Thousands of Types (Amazon, 2020)
Overview
AutoKnow: self-driving knowledge collection for products of thousands of types is an Amazon publication featured on Amazon Science in 2020. It describes an automated, self-driving pipeline for collecting product knowledge at e-commerce scale — covering thousands of product types — so that product attributes, relations, and taxonomy structures can be curated with reduced manual effort.
> Note: This forum post is a reading-note entry. For the paper's full methodology, system design, and quantitative results, consult the original publication page.
Why It Matters
Product knowledge graphs underpin search, recommendation, and personalization in large e-commerce platforms. Key challenges include:
- Scale and diversity: catalogs span thousands of product types, each with distinct attribute schemas.
- Noisy, heterogeneous sources: seller inputs, titles, descriptions, and images require automated extraction and cleaning.
- Maintenance cost: manual curation does not scale, motivating "self-driving" (automated, continuously learning) collection pipelines.
- Structured product knowledge complements retrieval-augmented generation (RAG): knowledge graphs and ontologies improve recall precision for attribute-driven and multi-hop queries.
- The work predates the LLM era but is frequently cited in discussions of GraphRAG and industrial knowledge graph construction.
- Related entries in this collection include surveys on graph RAG, agentic RAG, and domain-specific RAG (e.g., RAFT).
- Original: AutoKnow — Amazon Science
- A Survey of Graph Retrieval-Augmented Generation for Customized Large Language Models
- A Survey on Retrieval-Augmented Text Generation for Large Language Models
- Agentic Retrieval-Augmented Generation: A Survey on Agentic RAG
- RAFT: Adapting Language Model to Domain Specific RAG
- RAG vs. GraphRAG: A Systematic Evaluation and Key Insights
- Algolia's Knowledge Graphs and Ontologies — Adding Knowledge to Keyword Search
AutoKnow addresses these by automating knowledge collection across product categories, reducing human-in-the-loop effort for catalog structuring.
Context in the IR / RAG Landscape
Engineering Checklist (General Takeaways)
| Area | Consideration | |------|---------------| | Data | Source cleaning, PII handling, versioned indexes/embeddings | | Latency | Cascade retrieval with early stopping; cache frequent queries | | Quality | Offline metrics vs. online satisfaction; human audits | | Safety | Source allowlists, poisoning detection, output filtering | | Cost | Model routing, distillation, hybrid sparse+dense retrieval |
Glossary
| Term | Meaning | |------|---------| | IR | Information Retrieval | | RAG | Retrieval-Augmented Generation | | Knowledge Graph | Structured representation of entities and relations | | nDCG | Normalized Discounted Cumulative Gain, a ranking metric | | Agentic Search | Search modeled as sequential decisions and tool calls |