An automatic inspection system of work orders based on text classification

The number of rectification work orders handled by electric power system departments each year is as high as tens of millions, with each work order consisting of hundreds of Chinese characters on average. Inspection work is both heavy and difficult, so it is crucial and urgent to research automated inspection of rectification work orders. In this paper, we utilize the Transformer model as the foundation for building an automatic inspection system for rectification work orders. By analyzing the characteristics of rectification work orders, we propose using comprehensive position encoding and sparse attention in the system. Comprehensive position encoding allows the model to obtain both relative and absolute position information between tokens, while sparse attention enables the model to focus on key tokens and avoid irrelevant information. Our system first processes the text using a self-designed stop words list and data enhancement method. We then use an improved model to extract semantic features, followed by classifying the rectification work orders into two categories to obtain inspection results. The experimental results show that our system can automatically inspect rectification work orders with an accuracy rate of 95.41%.