E-commerce product attribute extraction made accurate and efficient
Scaling E-Commerce Attribute Extraction with Parallel Decoding
Computation and Language
Summary
Online product catalogs are often messy, making it hard to find the important details shoppers care about. The authors created a two-step process using language models to first find the most useful product features and then quickly pull out those details from product descriptions. Their method matches the accuracy of larger models but uses much less computing power. This helps build clear, organized product info that can improve shopping and product searching.
What this means in practice
- •For e-commerce platform engineers: Create automated pipelines that find key product attributes and extract them accurately while drastically reducing processing costs.
- •For online retail product managers: Enrich product listings with consistent, comparable attribute information to improve product discovery and customer decision making.
Authors
Nikhita Vedula, Dushyanta Dhyani, Bryan Wang, Shervin Malmasi
Abstract
Customers rely on specific product attributes to compare products and make purchasing decisions, but e-commerce catalogs are messy and unstructured, making it difficult to identify which attributes matter most and extract them at scale. Standard Attribute Value Extraction (AVE) systems treat all attributes equally, producing large, inconsistent attribute sets that do not reflect the factors consumers use to differentiate products. We introduce a two-stage LLM pipeline that first discovers a compact, ranked schema of purchase-discriminative attributes for each product category, then extracts their values from catalog text using a fine-tuned compact LLM (Qwen3-4B) with Hyper-Parallel Decoding (HPD). This pipeline achieves 85% extraction accuracy, on par with the foundational LLM it was distilled from, while reducing inference costs by 92% over foundational LLMs, enabling production-scale use for product discovery and catalog enrichment. The resulting category-level structured representations effectively constitute automatically constructed product knowledge bases, providing consistent, comparable attributes across varied product categories that can ground downstream knowledge-intensive applications.