Address matching method based on multi-task federated learning and address hierarchy knowledge

Abstract

The invention relates to an address matching method based on multi-task joint learning and address hierarchical structure knowledge, specifically comprising the following steps: S1: input address pair; S2: address pair enters shared address feature extraction network; address pair enters address element labeling network, After tagging, the first word segmentation result is obtained, and the first word segmentation result enters the address element recognition network; S3: The shared address feature enters the address element recognition network, and after word segmentation, the second word segmentation result is obtained, and the second word segmentation result and the first word segmentation result Comparison; used to judge whether the second word segmentation result conforms to the first word segmentation result and the hierarchical order of address elements, if not, readjust the shared address feature; S4: The adjusted shared address feature enters the address matching network and is used to judge the input Whether the address pair matches; S5: output the matching result. The invention combines the address level element recognition task and the address matching task to learn, thereby improving the performance of the model.

Description

technical field [0001] The invention relates to an address matching method based on multi-task joint learning and address hierarchical structure knowledge. Background technique [0002] Address matching is to match the unstructured addresses that need to be queried with the standard addresses in the database, so as to convert random addresses into standard geographic coordinates, so as to locate them on the map. The key issue is to determine whether the two address texts match , involving the comparison of corresponding address hierarchy elements. Address elements refer to the names of address entities such as provinces, cities, districts, and streets (such as Shenzhen). Previous methods mainly focus on complex rule-based string matching and shallow semantic matching based on machine learning or deep learning models, ignoring address-specific address hierarchy information. [0003] Traditional methods are usually based on character by character to determine the similarity ...

AI Technical Summary

Problems solved by technology

For example, in 2020, Lin Yue and others used the enhanced sequential inference deep learning model ESIM (Enhanced LSTM for Language Inference, enhanced LSTM for language inference) to infer whether the address matches from the local and the whole, and proved through experiments that this is a way to judge the address An effective method for matching, however they ignore how to make the model learn the address level information

Images