Classification and Implementation of Relevance Data Based on Address Text

Tasks based on address text are commonly seen in real life, such as location-based services based on geographic information systems, rapid localization and positioning systems for emergencies, and alignment of different address information systems. This article introduces the solutions to the classification problems of address text by comparing different models of different parameters, such as pre-training + nezha + rdropout and pre-training + nezha + swa + rdropout, in the classification of address information, obtaining different model effects. After using rdropout, it also shows a faster training speed and convergence speed.