Geographic entity extraction and correction methods based on the 12345 public service hotline

The 12345 public hotline service is one of the primary channels for residents to provide feedback on urban issues and suggestions. Extracting spatial information from this feedback is crucial for government departments to accurately locate various events and explore their spatial characteristics. Traditional methods, such as those based on regular expressions, often struggle with inaccurate extraction of geographic entity features and low recognition efficiency. To address these issues and improve fine-grained geographic entity recognition, a method for extracting and correcting geographic information based on 12345 text data is proposed in this paper. Initially, official data from the 12345 hotline in Sanya City were used as a corpus, and the geographic entities within it were annotated and classified. Subsequently, three geographic entity extraction models were selected for testing, and relevant evaluation metrics were calculated. Finally, a spatial correction method for geographic entities based on semantic matching was proposed. This method utilizes the Jaccard similarity function and spatial correction buffers to establish correction rules, further optimizing the extraction results. The results show that the geographic entity recognition method based on BERT-BiLSTM-CRF achieves a precision of 89.21%, recall of 92.59%, and F1 score of 90.86%. Additionally, after spatial correction, all geographic entities fall within the POI and road network buffer zones, making the recognition results more accurate and practically relevant.