Chinese Address Similarity Calculation Based on Auto Geological Level Tagging

摘要

How to quickly measure the similarity of addresses has become an urgent need in various fields including financial anti-fraud. Traditional string-based similarity calculation methods have not completed this task perfectly. Taking into account the hierarchical nature of addresses, we constructed a framework for calculating the similarity of Chinese addresses. First, the whole address strings are split and annotated with proper level by a LM-LSTM-CRF model, and then sub-string level similarities are calculated. Last, similarity scores are combining by BP neural networks. This framework has achieved good results in practice for financial anti-fraud tasks.