Interactive AI NPCs Powered by LLMs: Technical Report for the CPDC Challenge 2025

2025-11-25

AI Summary

This work addresses the weak commonsense reasoning and poor role consistency exhibited by large language model (LLM)-driven AI non-player characters (NPCs) in interactive dialogue. To this end, we propose an interactive role modeling framework that synergistically integrates context engineering with reinforcement learning. Methodologically, we introduce dynamic tool pruning and persona-aware feature trimming to achieve efficient context compression; replace supervised fine-tuning with reward-guided Generalized Reinforcement Policy Optimization (GRPO) to enhance task-oriented behavior while mitigating few-shot overfitting. The approach incorporates systematic engineering improvements—including prompt optimization, streamlined tool invocation, parameter normalization, and function merging. Evaluated on the CPDC 2025 Challenge, our solution ranked first in Task 2 (API), second in Task 1 (API), and third in both Task 3 (API) and the GPU track, demonstrating substantial gains in commonsense reasoning, role consistency, and cross-task generalization.

Technology Category

• Natural Language Processing: (Large) Language Models
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Machine Learning: Reinforcement Learning

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

This report presents the solution and results of our team MSRA_SC in the Commonsense Persona-Grounded Dialogue Challenge (CPDC 2025). We propose a simple yet effective framework that unifies improvements across both GPU Track and API Track. Our method centers on two key components. First, Context Engineering applies dynamic tool pruning and persona clipping for input compression, combined with post-processing techniques such as parameter normalization and function merging. Together with manually refined prompts, this design improves tool call stability, execution reliability, and role-playing guidance. Second, in the GPU Track, we further adopt GRPO training, replacing supervised fine-tuning with reinforcement learning directly optimized by reward signals. This mitigates small-sample overfitting and significantly enhances task-oriented dialogue performance. In the final evaluation, our team ranks 1st in Task 2 API, 2nd in Task 1 API, and 3rd in both Task 3 API and GPU track, demonstrating the effectiveness of our approach. Our code is publicly available at https://gitlab.aicrowd.com/nikoo_yu/cpdc-2025-winning-solution

Problem

• Improving tool call stability and execution reliability through context engineering
• Enhancing task-oriented dialogue performance with GRPO reinforcement learning
• Addressing small-sample overfitting in persona-grounded dialogue systems

Innovation

• Context Engineering with dynamic pruning and persona clipping
• GRPO training replaces supervised fine-tuning with reinforcement learning
• Parameter normalization and function merging enhance execution reliability