Enabling Explainable Recommendation in E-commerce with LLM-powered Product Knowledge Graph

TL;DR

The paper addresses explainable recommendations in e-commerce by combining LLMs with a product knowledge graph (PKG). It proposes LLM-PKG, which distills LLM knowledge into a PKG via offline construction, prompt-driven KG generation, validation and refinement, and internal product mapping, followed by online serving for item- and user-based recalls with explainable edges. The approach demonstrates significant improvements in user engagement and transactions in an online A/B test on sneaker data, validating the practicality of LLM-powered explainability at web scale. The work contributes a concrete pipeline for building and maintaining an LLM-driven PKG and shows how to integrate it with enterprise inventories for real-time recommendations.

Abstract

How to leverage large language model's superior capability in e-commerce recommendation has been a hot topic. In this paper, we propose LLM-PKG, an efficient approach that distills the knowledge of LLMs into product knowledge graph (PKG) and then applies PKG to provide explainable recommendations. Specifically, we first build PKG by feeding curated prompts to LLM, and then map LLM response to real enterprise products. To mitigate the risks associated with LLM hallucination, we employ rigorous evaluation and pruning methods to ensure the reliability and availability of the KG. Through an A/B test conducted on an e-commerce website, we demonstrate the effectiveness of LLM-PKG in driving user engagements and transactions significantly.

Figures (4)

• Figure 1: Overall Framework
• Figure 2: Effects of One-shot learning.
• Figure 3: Sneaker LLM-PKG Visualization (Sampled)
• Figure 4: An example of LLM-PKG based recommendation in experiments