Reproduce “PAE: LLM-based Product Attribute Extraction for E-Commerce Fashion Trends” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Apurva Sinha, Ekta Gujral
Walmart Global Technology developed PAE, an LLM-based framework that extracts product attributes and hashtags from unstructured, multi-modal PDF fashion trend reports. The system achieves 92.5% F1-score for text and 100% F1-score for image attribute extraction, enabling proactive assortment planning for e-commerce retailers.
Key insights
- The specificity and temperature of LLM prompts significantly impact attribute extraction accuracy, with a specific prompt and lower temperature (0.2) yielding optimal results.
- Extracting attributes from images can achieve higher accuracy (100% F1-score) compared to text-based extraction for future trend reports, due to clearer visual cues.
- A modular, multi-step framework combining PDF data extraction, LLM-based text and image attribute extraction, merging, and catalog matching enables effective processing of unstructured trend reports for proactive assortment planning.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.