CN-115809701-B - Graph feature search method and system for risk transaction capture

Abstract

The present disclosure proposes a graph feature search method for risk transaction capture. The method includes constructing graph features and an initial search space thereof based on transaction data, acquiring a plurality of candidate graph features from the initial search space, wherein each candidate graph feature comprises an information aggregation representation, determining a variation position and a variation value by utilizing a reinforcement learning strategy to obtain feature feedback, reducing the initial search space based on the feature feedback to obtain a target search space, acquiring target graph features from the target search space, and capturing risk transactions in the transaction data by utilizing the target graph features.

Inventors

• DAN JIAWANG
• ZHU LIANG
• TIAN SHENG

Assignees

• 支付宝(杭州)信息技术有限公司

Dates

Publication Date

20260508

Application Date

20221222

Claims (10)

1. 1. A graph feature search method for risk transaction capture, comprising: Constructing graph features and initial search spaces thereof based on the transaction data; obtaining a plurality of candidate graph features from the initial search space, wherein each candidate graph feature comprises an information aggregation representation; Recommending a variation location and a variation value using a reinforcement learning strategy, the variation location being one of a plurality of candidate variation locations of the graph feature and the variation value being one of a plurality of candidate feature values of the graph feature; applying the mutation locations and the mutation values to the plurality of candidate map features to obtain a plurality of mutation map features; determining and evaluating a feature value of each variation graph feature to obtain feature feedback; Downscaling the initial search space based on the feature feedback to obtain a target search space; Obtaining target graph characteristics from the target search space; The target graph features are utilized to capture risk transactions in the transaction data.
2. 2. The method of claim 1, each candidate graph feature comprising a subject, an object, a time window, a filtering condition, and a graph structure.
3. 3. The method of claim 1, the information aggregation representation comprising aggregation of two degrees of information and/or more.
4. 4. The method of claim 1, each candidate graph feature being a Graphical Query Language (GQL) feature.
5. 5. The method of claim 1, determining and evaluating feature values for each variogram feature to obtain the feature feedback further comprising: Determining a measure of the feature value of each variation graph feature; Screening out the variation graph characteristics of which the measurement does not meet the condition; And determining the importance degree of the reserved variation graph characteristics as the characteristic feedback.
6. 6. The method of claim 5, the metric comprising at least one of variance, missing value ratio, chi-square value, IV value, PSI value.
7. 7. The method of claim 5, wherein determining the importance of the retained variogram feature further comprises inputting the retained variogram feature into a tree model to determine the importance.
8. 8. The method of claim 1, reducing the initial search space based on the feature feedback to obtain a target search space further comprising iteratively performing the following steps until the feature feedback satisfies a preset condition: Adjusting a reinforcement learning strategy based on the feature feedback; Determining new mutation locations and mutation values using the adjusted reinforcement learning strategy to obtain a plurality of new mutation map features; determining and evaluating feature values of each new variogram feature to obtain new feature feedback; the initial search space is reduced based on the new feature feedback.
9. 9. A graph feature search system for risk transaction capture, comprising: the search space construction module constructs graph features and initial search spaces based on the transaction data; A candidate graph feature module that obtains a plurality of candidate graph features from the initial search space, wherein each candidate graph feature includes an information aggregation representation; A graph feature mutation module configured to: Recommending a variation location and a variation value using a reinforcement learning strategy, the variation location being one of a plurality of candidate variation locations of the graph feature and the variation value being one of a plurality of candidate feature values of the graph feature; applying the mutation locations and the mutation values to the plurality of candidate map features to obtain a plurality of mutation map features; determining and evaluating a feature value of each variation graph feature to obtain feature feedback; a search space reduction module that reduces the initial search space based on the feature feedback to obtain a target search space; The target graph feature module acquires target graph features from the target search space; And a capturing module for capturing risk transactions in the transaction data by utilizing the target graph characteristics.
10. 10. A computer readable storage medium storing a computer program executable by a processor to perform the method of any one of claims 1-8.

Description

Graph feature search method and system for risk transaction capture Technical Field The present disclosure relates generally to the field of machine learning, and more particularly to a graph feature search method and system for risk transaction capture. Background In the real world, the links between transactions do not always occur on a single object, but may exist on a common link between objects. For example, in a transaction network, by accumulating transaction amounts of all accounts of a user with a card over a period of time, it is possible to determine whether a transaction of the user is risky. Such accumulated information may be characterized by a graph. The graph features can effectively reveal the relation among things and promote modeling effect. However, to obtain graph features, case analysis is continuously performed by using existing data, and the manually refined patterns are organized into feature generation logic, which occupies a great deal of human resources. The feature interpretation obtained by the feature learning method typified by deep feature learning is poor, and sometimes cannot be used even directly. Meanwhile, the features obtained by the method in the prior art are often aggregation of first-degree information, and the features cannot be drawn by means of information etching of two degrees or more, so that efficient utilization of the drawing features is lacking. In view of this, it is desirable to provide an improved graph feature searching method and system, which can automatically obtain effective graph features, reduce manpower input, reduce time consumption of feature engineering, and promote modeling effects. Disclosure of Invention The following presents a simplified summary of one or more aspects in order to provide a basic understanding of such aspects. This summary is not an extensive overview of all contemplated aspects, and is intended to neither identify key or critical elements of all aspects nor delineate the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form as a prelude to the more detailed description that is presented later. The disclosure provides a graph feature searching method for risk transaction capture, which comprises the steps of constructing graph features and an initial searching space thereof based on transaction data, acquiring a plurality of candidate graph features from the initial searching space, wherein each candidate graph feature comprises an information aggregation representation, determining a variation position and a variation value by utilizing a reinforcement learning strategy to obtain feature feedback, reducing the initial searching space based on the feature feedback to obtain a target searching space, acquiring target graph features from the target searching space, and capturing risk transaction in the transaction data by utilizing the target graph features. In an embodiment of the present disclosure, each candidate graph feature includes a subject, an object, a time window, a filter condition, and a graph structure. In an embodiment of the present disclosure, the information aggregation represents an aggregation that includes two degrees of information and/or more than two degrees of information. In an embodiment of the present disclosure, the plurality of candidate graph features includes a randomly generated graph feature and a previously evaluated graph feature. In an embodiment of the present disclosure, each candidate graph feature is a Graphical Query Language (GQL) feature. In one embodiment of the present disclosure, determining the variance position and the variance value using the reinforcement learning strategy to obtain the feature feedback further comprises recommending the variance position and the variance value using the reinforcement learning strategy, applying the variance position and the variance value to the plurality of candidate map features to obtain a plurality of variance map features, and determining and evaluating a feature value of each variance map feature to obtain the feature feedback. In an embodiment of the present disclosure, determining and evaluating the feature value of each variation map feature to obtain the feature feedback further comprises determining a metric for the feature value of each variation map feature, screening out variation map features for which the metric does not satisfy a condition, and determining the importance of the retained variation map features as the feature feedback. In an embodiment of the present disclosure, the metrics include at least one of variance, missing value ratio, chi-square value, IV value, PSI value. In an embodiment of the present disclosure, determining the importance of the retained variation map features further comprises inputting the retained variation map features into a tree model to determine the importance. In an embodiment of the present disclosure, reducing the initi