Power Text Hierarchical Multi-label Classification Based On Deep Learning

With the continuous strengthening of information technology and the rapid improvement of technology level of State Grid Corporation of China,information system has been integrated into all parts of power grid operation.Electric power information communication customer service system(ICT customer service system)has accumulated a large number of work order failure records.Although the informatization level of the ICT customer service system has been improved year by year,the traditional operation and maintenance method that relies on the experience of customer service agents to locate fault types has been used for many years.Since the information system failures involved in the ICT customer service system failures do not have a clear failure category system and involve a wide range of areas,the knowledge of customer service personnel is limited.Therefore,there is a problem of fuzzy classification or even misclassification of the fault in the fault judgment and location,which causes the ICT customer service system to be unable to guarantee its specific timeliness and accuracy when processing the fault.Ensuring the safe and stable operation of the ICT customer service system,providing high-quality services,efficiently and accurately locating the type of failure based on the user's fault description,and proposing targeted solutions is the ultimate goal of the ICT customer service system.Therefore.it is extremely urgent to realize the automatic identification and location of fault types in the ICT customer service system.During the research,it was discovered that the types of faults involved in the work order fault records of the electric power ICT customer service system were not independent of each other,but had a hierarchical relationship.In the era of Internet big data,text classification technology is a basic task for us to use big data technology for information processing.The purpose is to sort and classify text resources,which can be effectively applied to ICT customer service system fault type classification.In view of the characteristics of the work order fault record of the electric power ICT customer service system,on the one hand,this paper automatically constructed hierarchical power failure labels of ICT customer service system,and on the other hand,took into consideration the taxonomy relation between fault types(categories),proposed the HDPCNN and HDRNN methods,and adopted the method of hierarchical category embedding to classify work order failure records of ICT customer service system layer by layer.In the process of classification,each layer will introduce the category embedded information of the classification result of the upper layer to realize the hierarchical text classification.The hierarchical text classification algorithm HDPCNN in this paper is better than the traditional multi-label classification algorithm when it is applied to the fault classification of ICT systems,and can intelligently and accurately locate the fault categories of the ICT customer service system.To a large extent,the timeliness and accuracy of ICT customer service system in dealing with faults are improved.Moreover,the hierarchical structure of ICT system power fault label constructed in this paper lays a foundation for the subsequent hierarchical text classification in the field of power.