Research on Fine-Tuning and Optimization Techniques of Language Models in the Field of Electric Power Customer Service

With the rapid advancement of artificial intelligence technology, the demand for intelligent services in the power customer service sector is growing. Large language models, as a core technology in natural language processing, have shown significant potential in the application within the power customer service field. However, directly applying general-purpose language models falls short of meeting the professional, timely, and personalized needs of the power customer service domain. This paper aims to investigate the fine-tuning optimization techniques of language models for the power customer service sector. By constructing high-quality instructions through human-machine collaboration, low-rank adaptive fine-tuning, reinforcement learning based on human feedback, and the distillation of reasoning capabilities in small-sample learning contexts, we enhance the model's performance and fine-tuning efficiency, as well as its ability to understand and respond to user intents. Experimental results demonstrate that the proposed methods significantly improve the accuracy of power customer service Q&A by 12.7%, increase the BLEU score of response generation by 9.3%, and enhance training efficiency by 3.2 times.