PU LEARNING BASED CROSS-REGIONAL ENTERPRISE TAX EVASION RECOGNITION METHOD AND SYSTEM

摘要

Disclosed are a positive-unlabeled learning (PU learning) based cross-regional enterprise tax evasion recognition method and system, comprising: first, on the basis of PU learning, generating a tag of a tag-free sample in a source domain by using a small amount of tagged positive samples in the source domain; second, constructing a cross-regional enterprise tax evasion recognition model by means of a domain adaption method of maximizing a classifier difference; then, training the model by using a tagged enterprise sample to adjust a parameter of the model; and finally, performing tax evasion recognition on the enterprise sample in a target domain by using the trained tax evasion recognition model. The present invention realizes the purpose of establishing the tax evasion recognition model for the target domain having no sample tag in the case that the enterprise samples in the source domain merely comprise positive samples and a large amount of unlabeled samples.