Quality Improvement Method for Power Equipment Defect Record

Text quality directly affects the merits of the text mining effect. On the basis of summarizing the quality problems existing in the defective text of power equipment, we propose a defect text quality improvement method. First, through the analysis of a large number of actual defect text, the problems of missing equipment layering and irregular description of defective content in defective texts are concluded. Then, to address the problem of missing equipment layering, we transform the problem of missing description of defect of power equipment into the problem of missing text structures. And the conversion of defect text to structures is realized through the UIE (Universal Information Extraction) framework. Finally, to address the problem of nonstandard defect description, we establish a library of standard statement for device defect description. We use similarity calculation to find the standard sentence with the most similar semantic and correct it, so as to achieve higher quality for poor quality text. Results show that the corrected text has a large improvement in accuracy on the defect level classification task. We provide an effective method for improving the quality of defect text.