Improved TextRank Method for Streamlined Generation of Audit Work Orders

Audit work orders in the power system are extensive, with an average length of hundreds of Chinese characters and an overwhelming annual volume of tens of millions of orders. This presents a significant challenge that requires addressing. To tackle this problem, we present a novel text summarization model that utilizes an improved TextRank algorithm to simplify audit work orders. Our method involves generating sentence vectors through pre-training models and clustering them while considering the position of sentences and clue words to calculate sentence weight. Additionally, we established a dedicated synonym library for audit work orders and introduced synonym processing to compute similarity. Finally, we merged similar sentences to streamline the abstract content further. Our experimental results show that our system effectively extracts vital information from the audit work order's text, increasing the ROUGE value by approximately 20%. This offers significant support for subsequent automatic audit tasks, and we believe that our proposed model can enhance the accuracy of audit procedures.