Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Leveraging Large Language Models for Active Merchant Non-player Characters

Byungjun Kim, Minju Kim, Dayeon Seo, Bugeun Kim

TL;DR

This work tackles the passive pricing and scripted communication of merchant NPCs by introducing MART, a two-module framework with an appraiser that estimates item value and a negotiator that conducts price negotiations using LLMs. The study systematically compares insertion methods, including in-context learning and supervised fine-tuning for appraisal and zero-shot prompting versus knowledge distillation for negotiation, across a WoW Classic item dataset. Key findings show that large LLMs via prompting deliver top pricing accuracy but can be unreliable, while smaller LLMs trained with SFT or KD offer robust, efficient alternatives with solid persuasive capability and controllable personality. The framework demonstrates the potential to enable active, immersive merchant interactions in open-world games, while acknowledging limitations and offering practical mitigation strategies for deployment.

Abstract

We highlight two significant issues leading to the passivity of current merchant non-player characters (NPCs): pricing and communication. While immersive interactions with active NPCs have been a focus, price negotiations between merchant NPCs and players remain underexplored. First, passive pricing refers to the limited ability of merchants to modify predefined item prices. Second, passive communication means that merchants can only interact with players in a scripted manner. To tackle these issues and create an active merchant NPC, we propose a merchant framework based on large language models (LLMs), called MART, which consists of an appraiser module and a negotiator module. We conducted two experiments to explore various implementation options under different training methods and LLM sizes, considering a range of possible game environments. Our findings indicate that finetuning methods, such as supervised finetuning (SFT) and knowledge distillation (KD), are effective in using smaller LLMs to implement active merchant NPCs. Additionally, we found three irregular cases arising from the responses of LLMs.

Leveraging Large Language Models for Active Merchant Non-player Characters

TL;DR

This work tackles the passive pricing and scripted communication of merchant NPCs by introducing MART, a two-module framework with an appraiser that estimates item value and a negotiator that conducts price negotiations using LLMs. The study systematically compares insertion methods, including in-context learning and supervised fine-tuning for appraisal and zero-shot prompting versus knowledge distillation for negotiation, across a WoW Classic item dataset. Key findings show that large LLMs via prompting deliver top pricing accuracy but can be unreliable, while smaller LLMs trained with SFT or KD offer robust, efficient alternatives with solid persuasive capability and controllable personality. The framework demonstrates the potential to enable active, immersive merchant interactions in open-world games, while acknowledging limitations and offering practical mitigation strategies for deployment.

Abstract

We highlight two significant issues leading to the passivity of current merchant non-player characters (NPCs): pricing and communication. While immersive interactions with active NPCs have been a focus, price negotiations between merchant NPCs and players remain underexplored. First, passive pricing refers to the limited ability of merchants to modify predefined item prices. Second, passive communication means that merchants can only interact with players in a scripted manner. To tackle these issues and create an active merchant NPC, we propose a merchant framework based on large language models (LLMs), called MART, which consists of an appraiser module and a negotiator module. We conducted two experiments to explore various implementation options under different training methods and LLM sizes, considering a range of possible game environments. Our findings indicate that finetuning methods, such as supervised finetuning (SFT) and knowledge distillation (KD), are effective in using smaller LLMs to implement active merchant NPCs. Additionally, we found three irregular cases arising from the responses of LLMs.

Paper Structure

This paper contains 19 sections, 3 equations, 1 figure, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Using LLMs to communicate with NPCs
  4. Using LLMs to estimate values
  5. MART Framework
  6. Appraiser Module
  7. Two tested approaches
  8. Used Dataset
  9. Evaluation metrics
  10. Results
  11. Discussion
  12. Negotiator Module
  13. Two tested approaches
  14. Dataset
  15. Evaluation metrics
  16. ...and 4 more sections

Figures (1)

  • Figure 1: Comparison of current interaction and our mart framework: (a) a screenshot of trader in WoW, borrowed from https://www.youtube.com/watch?v=VqVgXp-7h8A&t=706s (b) the proposed Appraiser module, and (c) the proposed Negotiator module.