Fraud risk identification method and device based on relational network

Abstract

The invention discloses a fraud risk identification method and device based on a relational network. The identification method comprises the steps of obtaining application user data; constructing a user relationship network based on the application user data; calculating a network structure index of the user relationship network; and based on the network structure index, identifying a fraud risk of the application user. According to the invention, isolated user nodes are associated and intercommunicated; the relational network system is constructed, a central node in the relational network isdiscovered; the discovered center nodes are used for judging the intermediary fraud risk in a business scene, potential fraud risk nodes are mined through high-risk label propagation between the nodes, the barrier of user isolation evaluation is broken through, the user application is evaluated more comprehensively and deeply, and thus the purposes of improving the fraud recognition rate and predicting the risk are achieved.

Description

technical field [0001] The invention relates to the field of computers, in particular to a method and device for identifying fraud risks based on a relational network. Background technique [0002] In recent years, with the vigorous development of Internet consumer finance, more and more people use online platforms to apply for business. The fast operation and simple process make mobile phones as the main application tool become the link between users and platforms, and between users and users. important medium. Fraud risk is one of the main risks faced by the Internet finance industry and an important part of credit risk management. In the process of identifying fraud risks, most traditional risk identification methods are based on the application information of a single user, and evaluate the user's risk situation in isolation, ignoring many other risks related to the user. Risks need to be better assessed and identified. The convenience of the application process also ...

AI Technical Summary

Problems solved by technology

The disadvantages of this method are: (1) Many current classification models, such as logistic regression and random forest, belong to supervised machine learning, which requires enough learning samples, and it takes a lot of time and cost to accumulate these samples; ( 2) Due to the need to extract new user features, if the features are related to third-party data, the model operation will require a lot of cost. If the new user has no relevant data and the model input items are missing, the accuracy will be greatly reduced

Images