An Automated Identification Method for Place Spoofing Toponyms

ABSTRACT Place spoofing toponyms (PSTs) are intentionally misleading toponyms that convey false meanings that are inconsistent with the real scenarios they represent. At present, the negative impact of PSTs on politics and the economy, especially in pursuit of commercial profits in the making of lofty toponyms, is becoming increasingly serious. It is difficult for some government departments to distinguish PSTs efficiently and comprehensively via manual methods. Therefore, there are three aspects in the creation of PSTs that require optimization to facilitate the task: semantic unit segmentation, vector representation, and classification algorithms. This paper proposes an automated method for identifying PSTs. It employs a Transformer model trained on manually labeled semantic units to perform word segmentation, utilizes a pretrained language model to generate word vectors, and employs TextRCNN for short text classification. The automated method was evaluated with 60,000 toponyms from Wuhan, China, and the results show that the weighted F1 is 97.48%, with high precision and recall, which could serve as a reference for toponym management by government departments.