Idea
Platform generating detailed product attribute taxonomies to enhance e-commerce search relevance and filtering capabilities.
Research Paper
Core Innovation
This paper introduces BEATS, a multi-stage human-in-the-loop LLM framework that bootstraps product attribute taxonomies from scratch. It uniquely integrates proactive quality checks by developers and domain-expert annotation in iterative cycles to progressively improve attribute accuracy and coverage, enabling large-scale attribute tagging that enhances search system components.
Why It Matters
E-commerce platforms in emerging markets often lack structured product attributes, limiting search precision and user experience. By generating and validating rich attribute taxonomies, BEATS enables granular filtering and better semantic search, improving product discoverability and customer satisfaction. This scalable approach transforms underdeveloped catalogs into enriched data assets that boost search effectiveness across large inventories.
Market Size (TAM)
$20–50B TAM for e-commerce search and catalog enrichment; $2–10B SAM from emerging market platforms and online marketplaces. Driven by growing e-commerce adoption and demand for improved search relevance.
Potential Customers & Pain Points
- E-commerce platforms – Lack detailed product attributes limiting search quality
- Online marketplaces – Need scalable attribute generation for diverse catalogs
- Retailers – Require improved product discoverability and filtering
- Search technology providers – Need enriched data for ranking and retrieval models.
Business Model
Subscription-based SaaS platform offering attribute taxonomy generation and tagging services with tiered pricing based on catalog size and feature set; potential for custom integration and consulting fees.
Competitive Landscape
- Algolia
- Salsify
- Coveo
- Clerk.io
Implementation Challenges
- High dependency on domain expert availability for annotation
- Integration complexity with existing e-commerce platforms
- Maintaining attribute quality across diverse and evolving product catalogs
Validation Strategy
- Pilot deployment with select e-commerce partners to measure search relevance improvements
- A/B testing of attribute-enriched search versus baseline catalogs
- Collecting user engagement and conversion metrics post-integration
- Iterative feedback loops with domain experts to refine taxonomy quality
Research Paper Overview
BEATS: Bootstrapping E-commerce Attribute Taxonomies for Search through Iterative Human-AI Collaboration
Summary
BEATS is a human-in-the-loop LLM framework that creates detailed product attribute taxonomies from scratch for e-commerce platforms with underdeveloped catalogs. It iteratively refines attribute generation through quality checks and expert annotation, enabling enriched product tagging that improves search filtering, ranking, and retrieval. Deployed at Rakuten Taiwan, it has enhanced millions of products across thousands of categories with tens of thousands of generated attributes.