Reproduce “AIGP: An LLM-Based Framework for Long-Term Value Alignment in E-Commerce Pricing” 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

Chennan Ma, Yanning Zhang, Siqi Hong, Xiuchong Wang, Fei Xiao, Keping Yang

The AIGP framework from Alibaba integrates a 30B parameter Large Language Model with an offline Reinforcement Learning-based Long-Term Value Estimator to generate interpretable and long-term value-aligned pricing decisions for e-commerce. Online A/B tests demonstrated sustained improvements, including a 13.21% increase in cumulative Gross Merchandise Value (GMV) and an 8.20% increase in milestone achievement rate over a 14-day horizon.

Key insights

  • Offline RL-trained LTVE robustly estimates long-term business value, showing strong correlation with realized GMV, ROI, and milestone achievement over 7- and 14-day horizons (Spearman correlation > 0.93).
  • Applying DPO specifically to the `[ACT_ONLY]` pricing action tokens is crucial for effective long-term value alignment, preventing dilution of the optimization signal by reasoning text.
  • Leveraging unstructured contextual information and an internal knowledge base significantly enhances the framework's robustness for cold-start products and challenging out-of-distribution scenarios.

Summary by alphaXiv.

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