LLP: LLM-based Product Pricing in E-commerce
Hairu Wang, Sheng You, Qiheng Zhang, Xike Xie, Shuguang Han, Yuchen Wu, Fei Huang, Jufeng Chen
TL;DR
This paper tackles price estimation for second-hand products on C2C platforms by introducing LLP, a retrieval-then-reasoning framework that uses LLMs to generate price suggestions grounded in real-time market references. It combines a dynamic Similar Products Retrieval module (GSID-based multimodal representations and Proxima ANNS) with an LLM-based reasoning module trained via supervised fine-tuning on a bidirectional reasoning dataset and guided by Group Relative Policy Optimization, plus a confidence-based filtering mechanism. Key contributions include the first LLM-based generative pricing system for second-hand goods, a retrieval-aware two-stage training regime, and a deployment on Xianyu with substantial improvements in RMSLE, MALE, SAR, and DAR, especially across unseen categories. The approach demonstrates strong generalization, robust performance under varying coverage, and practical viability for industrial pricing, highlighting the potential of retrieval-augmented LLMs in dynamic e-commerce pricing tasks. $RMSLE$, $MALE$, $SAR$, and $DAR$ are used to quantify accuracy and adoption, with $\text{RMSLE} = \sqrt{\frac{1}{M} \sum_{i=1}^{M} (\log(\hat{p}_i) - \log(p_i))^2}$, $\text{MALE} = \frac{1}{M} \sum_{i=1}^{M} |\log(\hat{p}_i) - \log(p_i)|$, $\text{SAR} = \frac{1}{M} \sum_{i=1}^{M} \mathbb{I}(\frac{|\hat{p}_i - p_i|}{p_i} \le \tau)$ with $\tau=0.2$, and $\text{DAR} = \frac{1}{M} \sum_{i=1}^{M} \mathbb{I}(\frac{|\hat{p}_i - p_i|}{p_i} \le \frac{a}{\ln(p_i+b)})$ for chosen $a,b$.
Abstract
Unlike Business-to-Consumer e-commerce platforms (e.g., Amazon), inexperienced individual sellers on Consumer-to-Consumer platforms (e.g., eBay) often face significant challenges in setting prices for their second-hand products efficiently. Therefore, numerous studies have been proposed for automating price prediction. However, most of them are based on static regression models, which suffer from poor generalization performance and fail to capture market dynamics (e.g., the price of a used iPhone decreases over time). Inspired by recent breakthroughs in Large Language Models (LLMs), we introduce LLP, the first LLM-based generative framework for second-hand product pricing. LLP first retrieves similar products to better align with the dynamic market change. Afterwards, it leverages the LLMs' nuanced understanding of key pricing information in free-form text to generate accurate price suggestions. To strengthen the LLMs' domain reasoning over retrieved products, we apply a two-stage optimization, supervised fine-tuning (SFT) followed by group relative policy optimization (GRPO), on a dataset built via bidirectional reasoning. Moreover, LLP employs a confidence-based filtering mechanism to reject unreliable price suggestions. Extensive experiments demonstrate that LLP substantially surpasses existing methods while generalizing well to unseen categories. We have successfully deployed LLP on Xianyu\footnote\{Xianyu is China's largest second-hand e-commerce platform.\}, significantly outperforming the previous pricing method. Under the same 30\% product coverage, it raises the static adoption rate (SAR) from 40\% to 72\%, and maintains a strong SAR of 47\% even at 90\% recall.