Analysis and Research on Model of Forecasting User Complaint Based on Hierarchical Labels

This paper mainly expounds the research status of user complaints in power industry, as well as the analysis and research of current user complaints, and establishes a new model of predicting user complaints based on hierarchical labels. Firstly, the hierarchical label generation method is used to refine and subdivide various behaviors of power users, so as to realize the classification of complaint behavior labels for power users, and build a portrait of complaint behavior from power users by establishing a user label system; Then, based on the complaint behavior of power users, a joint graph convolution network is formed by integrating graph convolution, time-gated convolution and non-time gated convolution, and finally the prediction of user complaint behavior is realized. The joint graph convolution network is used to predict the complaint behavior of power users, so as to realize the rapid and active response to user complaints, thereby improve the satisfaction of power users.