BEATS: Bootstrapping E-commerce Attribute Taxonomies for Search through Iterative Human-AI Collaboration
Researchers have developed BEATS, a framework that uses large language models and human-AI collaboration to create detailed product attribute taxonomies for e-commerce platforms. This iterative system refines attribute generation through quality checks and expert annotation, addressing the lack of structured data in emerging markets. The framework has been deployed at Rakuten Taiwan, enriching millions of products and improving search functionalities like faceted filtering and semantic representation. AI
IMPACT Enhances e-commerce search capabilities by enabling granular filtering and improved semantic understanding of products.