Knowledge Graphs Enhanced Large Language Model Prompt for Electric Power Question Answering

With the continuous development and digital transformation in the field of electric power, the application of large language models in the electric power industry has become a remarkable trend. The electric power industry is an information-intensive domain involving extensive data processing, predictive analysis, and decision-making. Therefore, the application of large language models in the electric power sector is of great significance. Current large language models such as GPT3.5 and GLM can perform well in tasks such as question answering dialogues. However, these models still face challenges such as answer hallucination and inaccurate responses. This paper proposes a method to enhance question answering in large language models using knowledge graphs, aiming to improve the accuracy and reliability of these models in question answering tasks in the electric power domain.The proposed method first utilizes local electric power data to extract triplets and generate a question answering dataset specific to the electric power domain using a large language model. Then, the relationships of the knowledge graph triplets are incorporated into the question prompt to enhance the quality of the model's answers. Furthermore, we fine-tune the large language model using the expanded question set derived from the triplets as knowledge enhanced data. Subsequently, we conduct experiments on both an electric power question answering dataset and a knowledge graph question answering dataset. The experimental results demonstrate that our method significantly improves various metrics of the large language model in the electric power question answering task. This research provides new insights and approaches to enhance the effectiveness of question answering systems in the electric power domain. Future studies can further explore and optimize this prompt expansion method for application in broader domains and tasks.