Address matching algorithm based on interest point knowledge graph pre-training

Abstract

The invention discloses an address matching algorithm based on interest point knowledge graph pre-training, comprising the following steps of: obtaining interest point addresses, distinguishing administrative regions with different granularities, and obtaining marked interest point addresses; randomly covering a part of administrative regions with the marked interest point addresses, inputting themarked interest point addresses into a language model, outputting predicted interest point addresses, calculating a loss function by utilizing the interest point addresses and the predicted interestpoint addresses, and obtaining the language model for outputting accurate interest point addresses after multiple iterations; connecting a full connection layer behind the language model, and performing overall parameter fine adjustment on the model and the full connection layer by using the marked address matching task data set to obtain a fine-adjusted language model and a fine-adjusted full connection layer; and inputting the marked to-be-predicted original interest point address into the finely tuned language model and the full connection layer to obtain a predicted address of the to-be-predicted interest point, and performing similarity calculation on the to-be-predicted original interest point address and the predicted address of the to-be-predicted interest point to complete addressmatching.

Description

technical field [0001] The present invention relates to the fields of knowledge graph and natural language processing, in particular to an address matching algorithm based on the pre-training of interest point knowledge graph. Background technique [0002] Natural language processing tasks in traditional network training models require a large amount of labeled data. The labeling of these data requires a lot of manpower. ability. [0003] Semantic matching of text refers to judging whether two pieces of natural language express the same meaning. The problem with the traditional bag-of-words model is that it cannot handle the ambiguity of natural language well. There are many titles and expressions for the same meaning, but the same The expression may have multiple meanings in different contexts. Classic semantic matching models include traditional bag-of-words-based TF-IDF, BM25 algorithm, deep learning-based DSSM, Match Pyramid and other models. [0004] As a natural lang...

AI Technical Summary

Problems solved by technology

Due to the differences in the expression of addresses and the complexity of geographic information, the exact character matching based on the traditional bag-of-words model cannot achieve good results; the problem with the deep learning model is that it requires a large amount of training corpus and is prone to generalization. Poor chemical performance and other characteristics

Images