Taxpayer industry classification method based on multistage generative model

Abstract

The invention discloses a taxpayer industry classification method based on a multistage generative model, which comprises the following steps: firstly, extracting to-be-mined text and non-text information in taxpayer industry information to carry out text embedding and coding, and carrying out feature processing on the coded information; secondly, converting the taxpayer industry category label with noise into a multi-complementary label; thirdly, constructing a multi-level generation model of label and feature levels under the bidirectional mapping framework; then, training the device based on the encoded features and the generated multiple complementary tags; and finally, taking the prediction of the label level on the real label as the final taxpayer industry category of the test data. According to the method, the noise rate of the tag is reduced by converting the noisy tag into the multi-complementary tag, the feature and tag level bidirectional mapping is introduced to cope with feature-dependent noise existing in the taxpayer industry category tag, and the taxpayer industry classification accuracy can be effectively improved.

Description

technical field [0001] The invention belongs to the field of industry classification, and in particular relates to a taxpayer industry classification method based on a multi-level generation model, which is used to solve the taxpayer industry classification label noise classification problem. Background technique [0002] At present, the industry classification in the tax registration information is mainly judged by the responsible personnel of the tax authority in the tax registration process based on the taxpayer's business scope and actual business operations based on experience, resulting in a discrepancy between the industry classification in the tax registration information and the actual industry classification. That is, there is noise in the labeling of the taxpayer's industry category. Therefore, how to train a noise-robust classifier based on the noisy taxpayer industry category labels, identify and correct the inconsistency between the existing taxpayer's business...

AI Technical Summary

Problems solved by technology

[0010]However, the existing technical solutions described in the above literature mainly have the following problems: when constructing taxpayer industry classification models, the problem of noise in the training data category labels is not considered, resulting in training The resulting model has poor classification accuracy

In addition, the existing technology only considers the one-way mapping from the feature level to the label level, and the noise of industry category labels is often directly related to characteristics such as business scope, and the one-way mapping is difficult to deal with the dependence between label noise and features.

Images