Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LaMP-Val: Large Language Models Empower Personalized Valuation in Auction

Jie Sun, Tianyu Zhang, Houcheng Jiang, Kexin Huang, Xiang Shu, Zhibo Zhu, Lintao Ma, Xingyu Lu, Jun Zhou, Junkang Wu, Chi Luo, An Zhang, Junkang Wu, Jiancan Wu, Xiang Wang

TL;DR

LaMP-Val addresses the neglected problem of modeling personalized item valuation in auctions by integrating semantic user preferences via large language models. It introduces a data–learning–evaluation framework: desensitized data augmentation from Epinions, diverse-instruction fine-tuning to obtain a personalized valuation model, and a closed-loop evaluation with a two-stage bidding-and-auction process that yields new metrics PU and PV. Empirical results show that LaMP-Val improves both valuation precision and personalized profits over strong baselines, with robust performance under different budgets and model variants. The work demonstrates the practical value of aligning valuation with individual user goals in auction settings and outlines directions for extending semantic valuation in broader auction mechanisms.

Abstract

Auctions are a vital economic mechanism used to determine the market value of goods or services through competitive bidding within a specific framework. However, much of the current research primarily focuses on the bidding algorithms used within auction mechanisms. This often neglects the potential benefits of incorporating individual users' unique preferences into the valuation process. Our theoretical and empirical analysis demonstrates that valuation errors can significantly impact the overall utility. To bridge this gap, we propose a personalized valuation framework, namely Large \underline{La}nguage \underline{M}odels-powered \underline{P}ersonalized \underline{Val}uation (LaMP-Val), which integrates Large Language Models to incorporate personalized semantic preference into users valuation process. LaMP-Val integrating three components: data, learning, and evaluation. The data component tackles the challenge of building a novel dataset specifically for LLMs fine-tuning in personalized valuation modeling. The learning component introduces a diversity template to enhance LLMs' capacity for modeling fine-grained personal valuation patterns. The evaluation component establishes a closed-loop system where LLM-generated valuations interact with bidding strategies and auction. It proposes two novel metrics to quantify valuation precision and bidding intention accuracy in personalized scenarios. Extensive experiments show that LaMP-Val more accurately captures personalized values and achieves greater profits than baseline approaches.

LaMP-Val: Large Language Models Empower Personalized Valuation in Auction

TL;DR

LaMP-Val addresses the neglected problem of modeling personalized item valuation in auctions by integrating semantic user preferences via large language models. It introduces a data–learning–evaluation framework: desensitized data augmentation from Epinions, diverse-instruction fine-tuning to obtain a personalized valuation model, and a closed-loop evaluation with a two-stage bidding-and-auction process that yields new metrics PU and PV. Empirical results show that LaMP-Val improves both valuation precision and personalized profits over strong baselines, with robust performance under different budgets and model variants. The work demonstrates the practical value of aligning valuation with individual user goals in auction settings and outlines directions for extending semantic valuation in broader auction mechanisms.

Abstract

Auctions are a vital economic mechanism used to determine the market value of goods or services through competitive bidding within a specific framework. However, much of the current research primarily focuses on the bidding algorithms used within auction mechanisms. This often neglects the potential benefits of incorporating individual users' unique preferences into the valuation process. Our theoretical and empirical analysis demonstrates that valuation errors can significantly impact the overall utility. To bridge this gap, we propose a personalized valuation framework, namely Large \underline{La}nguage \underline{M}odels-powered \underline{P}ersonalized \underline{Val}uation (LaMP-Val), which integrates Large Language Models to incorporate personalized semantic preference into users valuation process. LaMP-Val integrating three components: data, learning, and evaluation. The data component tackles the challenge of building a novel dataset specifically for LLMs fine-tuning in personalized valuation modeling. The learning component introduces a diversity template to enhance LLMs' capacity for modeling fine-grained personal valuation patterns. The evaluation component establishes a closed-loop system where LLM-generated valuations interact with bidding strategies and auction. It proposes two novel metrics to quantify valuation precision and bidding intention accuracy in personalized scenarios. Extensive experiments show that LaMP-Val more accurately captures personalized values and achieves greater profits than baseline approaches.

Paper Structure

This paper contains 26 sections, 2 theorems, 17 equations, 5 figures, 8 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Preliminary
  3. Method
  4. Data: Desensitive, Reasonable, Consistent Data Augmentation
  5. Supervised Fine-Tuning with Diversity-Instruction Generation
  6. Evaluation: Personalized Preference Evaluation on Real Data
  7. Experiments
  8. Preliminary Experiment
  9. Experimental Setup
  10. Main Results
  11. Stability Analysis of Profit
  12. Ablation Study
  13. Case Study
  14. Related work
  15. Conclusion
  16. ...and 11 more sections

Key Result

Lemma 1

For two real sequences $\{\alpha_n\}_{n=1}^{\infty}, \{\beta_n\}_{n=1}^{\infty}$, if $\lim_{N\to \infty}\frac{1}{N}\sum_{n=1}^N \alpha_n > 0$ and $\lim_{N\to \infty}\frac{1}{N}\sum_{n=1}^N \beta_n > 0$, then $\lim_{N\to \infty}\frac{1}{N}\sum_{n=1}^N \alpha_n \beta_n > 0$.

Figures (5)

  • Figure 1: (a) Existing works mainly focus on bidding algorithms (from value to bidding price) but neglect the valuation process (from user needs to determine value). (b) Existing works use experts to generate features for predicting values, but are limited to fixed features (e.g., "retro" not in training feature). (c) Using LLMs to analyze semantic information to predict value, accurately capturing user preferences.
  • Figure 2: Overview of our method: (1) Data: Transaction data undergoes desensitization processing, extracting information such as price, item name, rating, and review. LLMs are employed to analyze individual preferences, complete product descriptions, user preferences, valuations, and their justifications. Then, check the consistency and rationality of the generated data. If they are not met, retry. (b) Learning: Utilize diverse instruction templates to template item information and user reviews in the valuation dataset into a fine-tuning dataset. Train the pre-trained LLMs via SFT to form the final model (LLM FINAL). (c) Evaluation: LLM FINAL generates bidding decisions (e.g., preference #YES, valuation of $95) based on item information and reviews. Combined with other users' information, the final bidding result is determined through the bidding algorithm and auction mechanism.
  • Figure 3: A violin plot showing the ratio $v_\text{pred}/v_\text{label}$ from various models. The density of the violin, primarily around the red line, indicates a better valuation.
  • Figure 4: The number of "YES" and "NO" predicted by each model. The lower Relative Error Ratio indicates stronger label ratio alignment and improved preference distribution accuracy.
  • Figure 5: The discrepancy between a bidder's preferences and the auction outcomes, with the budget constraints, utility, value, PU, and PV.

Theorems & Definitions (4)

  • Lemma 1
  • proof
  • Theorem A.1
  • proof