Application and Exploration of Large Model-Based Intelligent Service Ticket Generation Technology in Large-Scale Power Customer Service

Addressing the complexities of conversational interactions, high information density, and stringent business logic in large-scale power customer service scenarios, this study investigates the application of large language model (LLM) agents in automated service ticket generation. We propose an intelligent ticket-generation architecture featuring "phased processing with multi-path validation," which decouples complex dialogue understanding into two sequential stages: critical information extraction and structured summary generation. This design enhances system accuracy and controllability. Based on real-world power customer service data, we systematically evaluate the performance of three LLMs-ERNIE 3.5-8K, Qwen2.5-32B, and Qwen3-32B-in ticket generation tasks. Experimental results demonstrate that the proposed method significantly improves the quality of information extraction and summary generation. Notably, Qwen3-32B achieves an overall accuracy of 96.33%, excelling particularly in complex semantic reasoning tasks such as customer request comprehension and emotion recognition. All evaluated models meet real-time operational efficiency requirements. This research validates the practical potential of LLM agents in industries with stringent requirements like power systems and provides actionable methodologies for constructing and optimizing intelligent customer service systems.