Reproduce “E-CARE: An Efficient LLM-based Commonsense-Augmented Framework for E-Commerce” 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

Ge Zhang, Rohan Deepak Ajwani, Yaochen Hu, Tony Zheng, Hongjian Gu, Wei Guo, Mark Coates, Yingxue Zhang

The E-CARE framework integrates commonsense reasoning from large language models into e-commerce recommendation systems by pre-computing and structuring knowledge into a reasoning factor graph, then leveraging lightweight adapters for efficient real-time inference. This approach enhances search relevance and app recall, yielding up to an 11.1 percentage point gain in Recall@5 and a 3.39% increase in Conversion Rate in online A/B tests.

Key insights

  • Pre-computing and structuring LLM-derived commonsense knowledge into a graph, rather than relying on real-time LLM inference, is crucial for achieving efficiency and scalability in high-throughput e-commerce environments.
  • The multi-stage graph construction pipeline (LLM reasoning, clustering, filtering) effectively condenses diverse LLM outputs into a concise and meaningful set of reasoning factors.
  • Lightweight adapters can successfully learn to map queries and products to the factors in the offline-constructed graph, allowing efficient access to complex commonsense reasoning at inference time.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.