Electric Power Project Correlation Mining and Question Answering: An Unsupervised Method

With rapid growth in the number of project applications in the electric power field, duplicate projects and cross-applications occur repeatedly, wasting financial resources. The traditional duplicate check methods and human review are increasingly unable to handle the problem. Companies need methods to discover project correlation and answer project content-related questions to help them review the declared projects. The emergence of large language models (LLMs) provides new ideas for understanding the projects’ content and mining project relations. However, the current LLM-based methods mainly focus on general knowledge and lack the knowledge of project review. This work analyzes the characteristics of project text in the electric power field and proposes an unsupervised project correlation mining and question-answering method based on LLMs. Then, we test the proposed method for the project review in the electric power field. Experiments show that this method can precisely discover the project correlations and is beneficial to improving the efficiency of project review in the electric power field.