Fraud pattern recognition model generation method, device, electronic device and computer-readable storage medium

By automatically generating a fraud pattern recognition model and using feature mining and filtering rules to process risk control scenario data, the problem of untimely and costly update of the fraud pattern recognition model is solved, the recognition accuracy and confrontation effect are improved, and the management process is simplified.

CN118503653BActive Publication Date: 2024-12-06BEIJING CHENGSHI WANGLIN INFORMATION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410940663.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-15
Publication Date
2024-12-06
Estimated Expiration
2044-07-15

AI Technical Summary

Technical Problem

The existing fraud pattern recognition model is not updated in time, resulting in attenuation of recognition accuracy, and the update process requires manual participation, which is costly and difficult to manage.

Method used

By automatically generating a fraud pattern recognition model, the sample set is used to continuously collect black and white sample data in risk control scenarios, the sample data is characterized by using feature mining paradigms and preset feature screening rules, and features with high expression capabilities are selected for training Network model, select the best performance model as the latest fraud pattern recognition model.

Benefits of technology

The fraud pattern recognition model is updated in a timely manner, the recognition accuracy and confrontation effect are improved, the time and labor cost of manual participation are reduced, and the model management is simplified.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118503653B_ABST
    Figure CN118503653B_ABST
Patent Text Reader

Abstract

The present application discloses a method, device, electronic device and computer-readable storage medium for generating a fraud pattern recognition model. The method for generating a fraud pattern recognition model includes obtaining a sample set, which continuously collects black and white sample data in a risk control scenario; performing feature aggregation processing on the sample data in the sample set using a feature mining paradigm to obtain a feature set containing multi-dimensional aggregated features; performing feature screening on the feature set using preset feature screening rules; using the screened features to train a variety of different types of network models, and selecting the model with the best performance from the trained models as the latest fraud pattern recognition model. The technical solution of the present application can update the model in a timely manner by automatically generating the latest fraud pattern recognition model, thereby improving the confrontation effect between the risk control system and fraud means, so that the strategy accuracy of online governance issues is always maintained in the effective governance range.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a fraud pattern recognition model generation method, device, electronic device and computer-readable storage medium. Background Art

[0002] The current fraud pattern recognition model is updated irregularly. The model can often achieve relatively good recognition results in the early stage of launch. As the confrontation escalates and the fraud methods change and update, the effect of the fraud pattern recognition model will gradually decline. Figure 1 The figure shows the accuracy change trend of the SZP ("pig killing") recognition model. In the early stage of the model, it has good recognition accuracy. As time goes by, the recognition accuracy of the model becomes lower and lower. In order for the fraud pattern recognition model to learn the latest fraud patterns, it needs to be updated frequently.

[0003] In the existing technology, model updates are usually performed manually. When designing the sample features required for training the model, it is necessary to use expert experience to design anti-fraud rules, extract samples and construct features based on the anti-fraud rules, and develop and deploy models based on the features. Summary of the invention

[0004] Embodiments of the present application provide a fraud pattern recognition model generation method, device, electronic device, and computer-readable storage medium to automatically generate the latest fraud pattern recognition model.

[0005] The present application embodiment adopts the following technical solutions:

[0006] In a first aspect, an embodiment of the present application provides a method for generating a fraud pattern recognition model, the method comprising:

[0007] Obtaining a sample set, wherein the sample set continuously collects black and white sample data in a risk control scenario;

[0008] Performing feature aggregation processing on the sample data in the sample set by using a feature mining paradigm to obtain a feature set containing multi-dimensional aggregate features;

[0009] Performing feature screening on the feature library according to preset feature screening rules;

[0010] The screened features are used to train various types of network models, and the model with the best performance is selected from the trained models as the latest fraud pattern recognition model.

[0011] Optionally, the using a feature mining paradigm to perform feature aggregation processing on the sample data in the sample set to obtain a feature set containing multi-dimensional aggregate features includes:

[0012] Acquire the pre-constructed feature mining paradigm, wherein the feature mining paradigm includes a main dimension aggregation feature mining paradigm, a main and auxiliary dimension aggregation feature mining paradigm, and a main group dimension aggregation feature mining paradigm;

[0013] According to the feature mining paradigm, feature mining and aggregation processing are performed on the sample data in the sample set to obtain the main dimension aggregation features, main and auxiliary sub-dimension aggregation features and main group dimension aggregation features corresponding to each sample data.

[0014] Optionally, the feature mining paradigm is constructed through the following steps:

[0015] The feature mining paradigm is constructed according to an identification field, a time field, an indicator field and a calculation function, wherein the identification field and the time field are used to match sample data, the indicator field is used to indicate the focus content in the matched sample data, and the calculation function field is used to calculate the statistical value of the indicator field.

[0016] Optionally, the identification field includes a risk control object identification, a risk control object attribute identification and the spatiotemporal information of the risk control object; the time field includes a fixed time window period or a sliding time window period; the indicator field includes at least one of a numerical field, a category field and a relationship field, the calculation function corresponding to the numerical field includes a data distribution feature calculation function, the calculation function corresponding to the category field includes a counting function and / or a proportion function, and the calculation function corresponding to the relationship field includes a whether function and / or a counting function.

[0017] Optionally, the sample data includes identification information and time information, and the performing feature mining and aggregation processing on the sample data in the sample set according to the feature mining paradigm includes:

[0018] According to the identification field and the time field in the feature mining paradigm and the identification information and the time information of the sample data, acquiring target sample data matching the identification field and the time field from the sample set;

[0019] Obtaining statistical values ​​of the indicator field in the target sample data according to the calculation function;

[0020] The multi-dimensional aggregation features corresponding to the target sample data are obtained according to the statistical values ​​of the target sample data.

[0021] Optionally, the preset feature screening rule includes at least one feature screening index selected from among an information importance index, a coverage index, a stability index and an iterative model subset index, and the feature screening of the feature library by using the preset feature screening rule includes:

[0022] Obtain each feature screening index value corresponding to each multi-dimensional aggregation feature in the feature library;

[0023] According to all feature screening index values ​​corresponding to each multi-dimensional aggregate feature, the feature expression capability value corresponding to each multi-dimensional aggregate feature is obtained;

[0024] Feature screening is performed based on the feature expression capability value corresponding to each multi-dimensional aggregate feature.

[0025] Optionally, after acquiring the latest fraud pattern recognition model, the method further includes:

[0026] The latest fraud pattern recognition model is used to perform historical data backtracking to determine the online threshold and recognition result label of the latest fraud pattern recognition model.

[0027] In a second aspect, an embodiment of the present application further provides a fraud pattern recognition model generation device, the device comprising:

[0028] A sample processing unit, used to obtain a sample set, which continuously collects black and white sample data in a risk control scenario;

[0029] A feature processing unit, used to perform feature aggregation processing on the sample data in the sample set using a feature mining paradigm to obtain a feature set containing multi-dimensional aggregate features;

[0030] A feature screening unit, used to perform feature screening on the feature library according to preset feature screening rules;

[0031] The model generation unit is used to train a variety of different types of network models using the screened features, and select the model with the best performance from the trained models as the latest fraud pattern recognition model.

[0032] In a third aspect, an embodiment of the present application further provides an electronic device, including:

[0033] A processor; and a memory arranged to store computer executable instructions, which, when executed, cause the processor to perform the aforementioned fraud pattern recognition model generation method.

[0034] In a fourth aspect, an embodiment of the present application also provides a computer-readable storage medium, which stores one or more programs. When the one or more programs are executed by an electronic device including multiple applications, the electronic device executes the aforementioned fraud pattern recognition model generation method.

[0035] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects:

[0036] The embodiment of the present application utilizes a sample set to continuously collect black and white sample data in risk control scenarios, and performs standardized feature processing on the sample data through a feature mining paradigm and preset feature screening rules, so that the screened features can improve the ability to express the current fraud pattern, thereby improving the recognition accuracy of the latest fraud pattern recognition model.

[0037] The embodiment of the present application can update the model in a timely manner by automatically generating the latest fraud pattern recognition model, thereby improving the effectiveness of the risk control system in combating fraud methods, so that the strategy accuracy of online governance issues is always maintained in the effective governance range, and no human intervention is required, thereby saving the time and manpower costs of model updates and facilitating the management of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0039] Figure 1 This is a schematic diagram of the recognition accuracy trend of the SZP recognition model;

[0040] Figure 2 A schematic diagram of a process for generating a fraud pattern recognition model in an embodiment of the present application;

[0041] Figure 3 A schematic diagram of a mining logic of a multi-dimensional aggregation feature in an embodiment of the present application;

[0042] Figure 4 This is a schematic diagram of the online process of a fraud pattern recognition model in an embodiment of the present application;

[0043] Figure 5 This is a schematic diagram of a fraud pattern recognition model generation process in an embodiment of the present application;

[0044] Figure 6 This is a schematic diagram showing the comparison of the accuracy of the successive fraud pattern recognition models and the initial fraud pattern recognition model in the embodiments of the present application;

[0045] Figure 7 This is a structural block diagram of a fraud pattern recognition model generation device in an embodiment of the present application;

[0046] Figure 8 This is a schematic diagram of the structure of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0048] The technical solutions provided by various embodiments of the present application are described in detail below in conjunction with the accompanying drawings.

[0049] Research has found that there are at least the following problems with manual model updating:

[0050] First, the model is not updated in time, and the online accuracy of the model decays quickly. If the update speed lags behind the decay speed, it may cause property losses.

[0051] Second, the cost of model updates is high, and each update requires at least one week of manpower;

[0052] Third, it is necessary to coordinate personnel from different departments to configure the model. When the model update frequency is increased, it will bring additional workload to employees and is not conducive to model management.

[0053] In response to the above problems, the embodiments of the present application provide a method for generating a fraud pattern recognition model. The following embodiments of the present application use a computer as an example of the execution subject of the fraud pattern recognition model generation method. It is understandable that the execution subject of the embodiments of the present application may also be other devices or systems, such as a server, a computing system, or a risk control system. Alternatively, the execution subject of the fraud pattern recognition model generation method of the embodiments of the present application may also be the cloud. The execution subject in the embodiments of the present application may be software or hardware, and different execution steps may be centralized or distributed.

[0054] like Figure 2 As shown, a flow chart of a fraud pattern recognition model generation method in an embodiment of the present application is provided, and the method at least includes the following steps S210 to S240:

[0055] Step S210, obtaining a sample set, wherein the sample set continuously collects black and white sample data in a risk control scenario.

[0056] The black and white sample data in this embodiment are related to the risk control scenario. The black sample corresponds to the sample data with risks, and the white sample corresponds to the sample data without risks. Taking the information service platform as an example, the black sample can be black user data involving "order fraud" and "pig killing fraud".

[0057] The sample set of this embodiment continuously collects black and white sample data in risk control scenarios to obtain the latest black sample data, providing a data basis for training the latest fraud pattern recognition model. It should be noted that this embodiment does not limit the source of sample data. In actual applications, it can come from the production environment or from a third party.

[0058] The sample data in this embodiment includes identification information and time information. The identification information includes unique identification information of the risk control object corresponding to the sample data and attribute information of the risk control object. The time information includes the production time or acquisition time corresponding to the sample data.

[0059] Step S220: performing feature aggregation processing on the sample data in the sample set using a feature mining paradigm to obtain a feature set containing multi-dimensional aggregated features.

[0060] Research has found that there are some attempts at automated feature mining in the industry, but the rationality of feature logic cannot be guaranteed, the mined features lack interpretability, and are not suitable for real production environments.

[0061] To address this problem, this embodiment pre-constructs a feature mining paradigm for mining multi-dimensional features and performing feature aggregation. By analyzing the characteristics of black sample data in risk control scenarios, for example, the published information involving "brushing fraud" and "pig killing fraud" often includes special content such as "0 yuan rebate". By reasonably designing the feature mining paradigm, the rationality and explainability of the feature logic can be guaranteed, and a relatively complete, fixed and standardized feature mining process can also be formed through the feature mining paradigm.

[0062] This embodiment uses the feature mining paradigm to perform feature mining and feature aggregation processing on the sample data in the sample set, and can obtain multi-dimensional aggregate features corresponding to each sample data. The multi-dimensional aggregate features can be understood as aggregate features of different dimensions about the risk control object, and the risk control object is related to the specific risk control scenario.

[0063] This embodiment exemplifies that each sample data includes three-dimensional aggregate features, namely, primary dimension aggregate features, primary and secondary dimension aggregate features, and primary group dimension aggregate features; wherein each dimension aggregate feature may include one or more statistical values. In other embodiments of the present application, each sample data may also include more dimensional aggregate features, and those skilled in the art may design corresponding dimensional information according to specific risk control scenarios.

[0064] If we take the information service platform mentioned above as an example, the risk control object is the information publisher. For the multi-dimensional aggregation features of the information publisher, the main dimension aggregation features can be one or more aggregation features produced with the user information of the information publisher (user information includes, for example, the ID, telephone number, ID number, etc. of the information publisher) as the main identifier; the main and auxiliary dimension aggregation features can be one or more aggregation features produced with the user information and attribute information of the information publisher (attribute information includes, for example, age, gender, address, education level, etc.) as the main identifier; the main group dimension aggregation features can be one or more aggregation features produced with the user information and spatiotemporal information of the information publisher (spatiotemporal information includes, for example, the IP address of the information publisher, the time of information release, etc.) as the main identifier.

[0065] It should be noted that the acquisition, use and analysis of all user information involved in this application have been explicitly authorized by the user.

[0066] Step S230: performing feature screening on the feature library using preset feature screening rules.

[0067] The feature set obtained through the above steps includes favorable features and unfavorable features. The favorable features can be understood as features that are conducive to expressing the current fraud pattern, and correspondingly, the unfavorable features can be understood as features that do not contribute to the expression of the current fraud pattern.

[0068] This embodiment uses preset feature screening rules to perform feature screening, which can increase the proportion of favorable features and thus improve the performance of the new model. Feature screening can also reduce the amount of model calculation and improve model update efficiency.

[0069] Step S240, using the screened features to train multiple different types of network models, and selecting the model with the best performance from the trained models as the latest fraud pattern recognition model.

[0070] This embodiment can construct a pipeline model Pipline, which supports the embedding of various network models, such as single models such as tree models and nearest neighbor models, and can also include fusion models (fusion models refer to large models obtained by fusing multiple different models). The selected features are used to train each network model in the pipeline model, and the model with the best performance is selected as the latest fraud pattern recognition model through performance indicators.

[0071] The performance index includes, for example, the AUC (Area Under Curve) index. AUC refers to the area under the ROC (Receiver Operating Characteristic Curve) and the coordinate axis. The value of this area will not be greater than 1. The closer the AUC is to 1.0, the higher the authenticity of the recognition method. When it is equal to 0.5, the authenticity is the lowest. Of course, those skilled in the art may also choose other performance indicators, and the embodiments of the present application do not impose specific restrictions on this.

[0072] like Figure 2 It can be seen from the fraud pattern recognition model generation method shown that the embodiment of the present application uses a sample set to continuously collect black and white sample data in the risk control scenario, and performs standardized feature processing on the sample data through a feature mining paradigm and preset feature screening rules, so that the screened features can improve the ability to express the current fraud pattern, thereby improving the recognition accuracy of the latest fraud pattern recognition model. By automatically generating the latest fraud pattern recognition model, the embodiment of the present application can update the model in a timely manner, thereby improving the effectiveness of the risk control system in confronting fraudulent means, so that the strategy accuracy of online governance issues is always maintained in the effective governance range, and no human intervention is required, thereby saving the time and labor costs of model updates and facilitating the management of the model.

[0073] In some embodiments of the present application, in the above step S220, feature aggregation processing is performed on the sample data in the sample set using a feature mining paradigm to obtain a feature set containing multi-dimensional aggregate features, specifically including:

[0074] Acquire the pre-constructed feature mining paradigm, wherein the feature mining paradigm includes a main dimension aggregation feature mining paradigm, a main and auxiliary dimension aggregation feature mining paradigm, and a main group dimension aggregation feature mining paradigm;

[0075] According to the feature mining paradigm, feature mining and aggregation processing are performed on the sample data in the sample set to obtain the main dimension aggregation features, main and auxiliary sub-dimension aggregation features and main group dimension aggregation features corresponding to each sample data.

[0076] The embodiment of the present application can pre-construct a feature mining paradigm based on the features of risk scenarios and black samples. In some possible implementation schemes of the present embodiment, the feature mining paradigm can be constructed through the following steps:

[0077] The feature mining paradigm is constructed according to the identification field, the time field, the indicator field and the calculation function, that is, the feature mining paradigm = identification field + time field + indicator field + calculation function.

[0078] The identification field and the time field are used to match sample data, the indicator field is used to indicate the focus content in the matched sample data (the focus content can be understood as content related to fraudulent means), and the calculation function is used to calculate the statistical value of the indicator field.

[0079] The identification field includes the risk control object identification (the risk control object identification refers to the identification information that can uniquely identify the risk control object, such as ID, telephone number, ID number, etc.), the risk control object attribute identification (the risk control object attribute identification refers to the attribute information of the risk control object, such as age, gender, address, education level, etc.) and the risk control object's spatiotemporal information (the spatiotemporal information can be understood as the time information and IP address of the risk control object, etc.).

[0080] The time field includes a fixed time window period or a sliding time window period. The fixed time window period is, for example, a fixed time period from XXX time to XXX time, and the sliding time window period is, for example, a non-fixed time period of nearly XX days or nearly XX hours.

[0081] The indicator field includes at least one of a numerical field, a category field, and a relationship field (the relationship field includes a comparison field and / or an inclusion field); the calculation function corresponding to the numerical field includes a data distribution feature calculation function; the calculation function corresponding to the category field includes a counting function and / or a proportion function; and the calculation function corresponding to the relationship field includes a whether function and / or a counting function.

[0082] Each feature mining paradigm in this embodiment is used to produce aggregate features of one dimension, for example, the main dimension aggregate feature mining paradigm is used to produce the main dimension aggregate features, the main and auxiliary dimension aggregate feature mining paradigm is used to produce the main and auxiliary dimension aggregate features, and the main group dimension aggregate feature mining paradigm is used to produce the main group dimension aggregate features. Specifically, the structures of the three feature mining paradigms are as follows:

[0083] The main dimension aggregation feature mining paradigm = risk control object identification field + time field + indicator field + calculation function;

[0084] The main and auxiliary dimension aggregate feature mining paradigm = risk control object identification field + risk control object attribute identification field + time field + indicator field + calculation function;

[0085] The main group dimension aggregation feature mining paradigm = risk control object identification field + risk control object spatiotemporal field + time field + indicator field + calculation function.

[0086] In some other possible implementations of this embodiment, feature mining and aggregation processing are performed on the sample data in the sample set according to the feature mining paradigm, including:

[0087] According to the identification field and the time field in the feature mining paradigm and the identification information and the time information of the sample data, acquiring target sample data matching the identification field and the time field from the sample set;

[0088] Obtaining statistical values ​​of the indicator field in the target sample data according to the calculation function;

[0089] The multi-dimensional aggregation features corresponding to the target sample data are obtained according to the statistical values ​​of the target sample data.

[0090] Figure 3 A schematic diagram of the production process for producing the above-mentioned main dimension aggregation features, main and auxiliary sub-dimension aggregation features and main group dimension aggregation features in the information service scenario is shown, as Figure 3 As shown:

[0091] When using the main dimension aggregate feature mining paradigm for feature mining, firstly, the sample data in the sample set is matched according to the risk control object identification field, time field and the identification information and time information of the sample data to obtain the target sample data that matches the risk control object identification field and time field; then, the statistical value of the indicator field in the target sample data is calculated according to the calculation function, and the statistical value corresponding to the indicator field is an aggregate feature of the target sample data under the main dimension. When the main dimension aggregate feature mining paradigm includes multiple indicator fields and / or multiple calculation functions, other aggregate features of the target sample data under the main dimension can be obtained. In other words, each sample data can have more than one aggregate feature under each dimension.

[0092] Similarly, the main and auxiliary dimension aggregation feature mining paradigm is used to obtain all the aggregation features of each sample data under the main and auxiliary dimensions, and the main group dimension aggregation feature mining paradigm is used to obtain all the aggregation features of each sample data under the main group dimension.

[0093] After obtaining all the aggregate features of each sample data in each dimension, all the aggregate features are sorted in a certain order, so that the multi-dimensional aggregate features of each sample data can be obtained.

[0094] In some embodiments of the present application, the preset feature screening rule includes at least one feature screening index among an information importance index, a coverage index, a stability index and an iterative model subset index, wherein:

[0095] Information Value (IV) is an indicator for evaluating the predictive ability of a feature for a target variable (especially the target variable in a binary classification problem). It is used to measure the ability of a feature to distinguish the target variable after binning. The coverage index can be understood as the degree of feature missingness.

[0096] The Population Stability Index (PSI) measures the stability of the data by comparing the fluctuations of the historical sample distribution with the current sample distribution;

[0097] The iterative model subset index adopts forward iteration or backward iteration to continuously add the gain brought by feature comparison or reduce the feature comparison loss to determine the importance of the feature.

[0098] In the above step S230, the feature library is screened by using the preset feature screening rules, which specifically includes:

[0099] Obtain each feature screening index value corresponding to each multi-dimensional aggregation feature in the feature library;

[0100] According to all feature screening index values ​​corresponding to each multi-dimensional aggregate feature, the feature expression capability value corresponding to each multi-dimensional aggregate feature is obtained;

[0101] Feature screening is performed based on the feature expression capability value corresponding to each multi-dimensional aggregate feature.

[0102] This embodiment can perform weighted calculation on the feature screening index value, and use the weighted calculation result as the feature expression capability value corresponding to each multi-dimensional aggregated feature. Of course, the feature expression capability value can also be obtained through other existing calculation schemes, and the embodiment of the present application does not make specific restrictions on this.

[0103] After obtaining the feature expression capability value corresponding to each multi-dimensional aggregate feature, the multi-dimensional aggregate features with higher feature expression capability values ​​can be screened out. Optionally, the principal component analysis method (PCA) or linear discriminant analysis method (LDA) can be used to perform feature dimensionality reduction on the screened features to reduce feature redundancy and improve feature expression capability.

[0104] In some embodiments of the present application, after obtaining the latest fraud pattern recognition model, Figure 2 The methods also include:

[0105] The latest fraud pattern recognition model is used to perform historical data backtracking to determine the online threshold and recognition result label of the latest fraud pattern recognition model.

[0106] Combine the following Figure 4 and Figure 5 The fraud pattern recognition model generation process of the embodiment of the present application is described in detail.

[0107] like Figure 4As shown in the figure, the risk control system is deployed with monitoring service and label service. The label service is used to manage the identification labels output by the online fraud pattern recognition model, and the monitoring service is used to monitor the label value of the identification label, trigger the model update process according to the label value, or the monitoring service triggers the model update process regularly. When the label value is lower than the preset label threshold, it means that the online fraud pattern recognition model cannot effectively identify fraud problems, and the model should be updated at this time.

[0108] like Figure 5 As shown, when the model needs to be updated, the sample set needs to be processed first. Taking the information service platform as an example, all black samples such as complaints and audits are continuously collected through the business platform, and abnormal samples in the sample set are removed. Abnormal samples are, for example, samples with too long a time period, etc. Abnormal samples are related to specific risk control scenarios. Then, the sample data is subjected to feature aggregation processing using the feature mining paradigm. The specific processing process refers to the relevant embodiments in the previous text. After obtaining the multi-dimensional aggregation features corresponding to each sample data, the multi-dimensional aggregation features can also be subjected to feature normalization processing, such as normalization processing, missing value processing, and abnormal processing. In practical applications, a data engine can be developed, and the data engine manages the sample set and the feature set. Then, the network model is built, and then the feature library is subjected to feature screening using the preset feature screening rules. The preset feature screening rules include, for example, information importance index, coverage index, stability index, and iterative model subset index. The screened features are used to train the tree model, the nearest neighbor model, or the fusion model respectively. Based on the performance index of the model, the model with the best performance index is used as the new fraud pattern recognition model, and the online threshold of the new fraud pattern recognition model is determined, and the score is converted into an identification label.

[0109] like Figure 4 As shown in the figure, after obtaining the new fraud pattern recognition model, model evaluation is still required. In practical applications, the new model can be evaluated through a multi-layer evaluation system. For example, the accuracy of the new model can be verified offline and through different traffic. If the evaluation fails, the automatic rollback mechanism will be initiated. After the evaluation passes, the risk control system will provide a container for seamless model replacement, and use the new model to seamlessly replace the old online model.

[0110] In this way, a new fraud pattern recognition model can be automatically generated and updated in a timely manner, improving the adversarial effect and ensuring that the accuracy of the online governance strategy is always maintained within the effective governance range. Figure 6 As shown, the accuracy of successive versions of the fraud pattern recognition model has always been above 50%.

[0111] The present application embodiment also provides a fraud pattern recognition model generation device 700, such as Figure 7As shown, a schematic diagram of the structure of a fraud pattern recognition model generation device in an embodiment of the present application is provided, wherein the device 700 includes: a sample processing unit 710, a feature processing unit 720, a feature screening unit 730 and a model generation unit 740, wherein:

[0112] A sample processing unit 710 is used to obtain a sample set, which continuously collects black and white sample data in a risk control scenario;

[0113] A feature processing unit 720 is used to perform feature aggregation processing on the sample data in the sample set using a feature mining paradigm to obtain a feature set containing multi-dimensional aggregate features;

[0114] A feature screening unit 730 is used to screen the feature library according to preset feature screening rules;

[0115] The model generation unit 740 is used to train a plurality of different types of network models using the screened features, and select the model with the best performance from the trained models as the latest fraud pattern recognition model.

[0116] In some embodiments of the present application, the feature processing unit 720 is specifically used to obtain the pre-constructed feature mining paradigm, which includes a main dimension aggregation feature mining paradigm, a main and auxiliary dimension aggregation feature mining paradigm, and a main group dimension aggregation feature mining paradigm; according to the feature mining paradigm, feature mining and aggregation processing are performed on the sample data in the sample set to obtain the main dimension aggregation features, main and auxiliary secondary dimension aggregation features, and main group dimension aggregation features corresponding to each sample data.

[0117] In some embodiments of the present application, the fraud pattern recognition model generation device 700 further includes a paradigm construction unit;

[0118] A paradigm construction unit is used to construct the feature mining paradigm according to an identification field, a time field, an indicator field and a calculation function, wherein the identification field and the time field are used to match sample data, the indicator field is used to indicate the focus content in the matched sample data, and the calculation function is used to calculate the statistical value of the indicator field; the identification field includes a risk control object identification, a risk control object attribute identification and the spatiotemporal information of the risk control object; the time field includes a fixed time window period or a sliding time window period; the indicator field includes at least one of a numerical field, a category field and a relationship field, the calculation function corresponding to the numerical field includes a data distribution feature calculation function, the calculation function corresponding to the category field includes a counting function and / or a proportion function, and the calculation function corresponding to the relationship field includes a whether function and / or a counting function.

[0119] In some embodiments of the present application, the sample data includes identification information and time information, and the feature processing unit 720 is further used to obtain target sample data matching the identification field and the time field from the sample set according to the identification field and the time field in the feature mining paradigm and the identification information and time information of the sample data; obtain the statistical value of the indicator field in the target sample data according to the calculation function; and obtain the multi-dimensional aggregation features corresponding to the target sample data according to the statistical value of the target sample data.

[0120] In some embodiments of the present application, the preset feature screening rules include at least one feature screening indicator selected from the group consisting of an information importance index, a coverage index, a stability index, and an iterative model subset index. The feature screening unit 730 is specifically used to obtain each feature screening indicator value corresponding to each multi-dimensional aggregate feature in the feature library; obtain the feature expression capability value corresponding to each multi-dimensional aggregate feature according to all feature screening indicator values ​​corresponding to each multi-dimensional aggregate feature; and perform feature screening according to the feature expression capability value corresponding to each multi-dimensional aggregate feature.

[0121] In some embodiments of the present application, the model generation unit 740 is also used to, after acquiring the latest fraud pattern recognition model, use the latest fraud pattern recognition model to perform historical data backtracking to determine the online threshold and recognition result label of the latest fraud pattern recognition model.

[0122] It can be understood that the above-mentioned fraud pattern recognition model generation device can implement the various steps of the fraud pattern recognition model generation method provided in the aforementioned embodiment, and the relevant explanations about the fraud pattern recognition model generation method are applicable to the fraud pattern recognition model generation device and will not be repeated here.

[0123] Figure 8 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present application. Figure 8 At the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and a memory. The memory may include a memory, such as a high-speed random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage. Of course, the electronic device may also include hardware required for other services.

[0124] The processor, network interface and memory can be interconnected through an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 8 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0125] The memory is used to store the program. Specifically, the program may include a program code, and the program code includes a computer operation instruction. The memory may include a memory and a non-volatile memory, and provides instructions and data to the processor.

[0126] The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, forming a fraud pattern recognition model generation device at the logical level. The processor executes the program stored in the memory and is specifically used to perform the following operations:

[0127] Obtaining a sample set, wherein the sample set continuously collects black and white sample data in a risk control scenario;

[0128] Performing feature aggregation processing on the sample data in the sample set by using a feature mining paradigm to obtain a feature set containing multi-dimensional aggregate features;

[0129] Performing feature screening on the feature library according to preset feature screening rules;

[0130] The screened features are used to train various types of network models, and the model with the best performance is selected from the trained models as the latest fraud pattern recognition model.

[0131] The above application Figure 2The method performed by the fraud pattern recognition model generation device disclosed in the illustrated embodiment can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be directly embodied as a hardware decoding processor for execution, or a combination of hardware and software modules in the decoding processor for execution. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the fraud pattern recognition model generation method in combination with its hardware.

[0132] The electronic device may also perform Figure 2 The method executed by the fraud pattern recognition model generation device in the present invention is realized by the fraud pattern recognition model generation device in the present invention. Figure 2 The functions of the illustrated embodiment will not be described in detail in the embodiments of the present application.

[0133] The present application also provides a computer-readable storage medium, which stores one or more programs, wherein the one or more programs include instructions, which, when executed by an electronic device including multiple application programs, enable the electronic device to execute Figure 2 The method performed by the fraud pattern recognition model generation device in the illustrated embodiment is specifically used to perform:

[0134] Obtaining a sample set, wherein the sample set continuously collects black and white sample data in a risk control scenario;

[0135] Performing feature aggregation processing on the sample data in the sample set by using a feature mining paradigm to obtain a feature set containing multi-dimensional aggregate features;

[0136] Performing feature screening on the feature library according to preset feature screening rules;

[0137] The screened features are used to train various types of network models, and the model with the best performance is selected from the trained models as the latest fraud pattern recognition model.

[0138] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0139] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0140] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0141] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0142] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0143] Memory may include non-permanent storage in a computer-readable medium, in the form of random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0144] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0145] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0146] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0147] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A fraud pattern recognition model generation method, characterized in that: The method comprises: Acquire a sample set, wherein the sample set continuously collects black and white sample data in a risk control scenario, wherein the sample data includes identification information and time information, wherein the identification information includes unique identification information of a risk control object corresponding to the sample data and attribute information of the risk control object, and the time information includes production time or acquisition time corresponding to the sample data; Performing feature aggregation processing on the sample data in the sample set by using a feature mining paradigm to obtain a feature set containing multi-dimensional aggregate features; Performing feature screening on the feature set by using preset feature screening rules; Use the selected features to train different types of network models, and select the model with the best performance from the trained models as the latest fraud pattern recognition model; The feature mining paradigm is constructed through the following steps: The feature mining paradigm is constructed according to an identification field, a time field, an indicator field and a calculation function, wherein the identification field and the time field are used to match sample data, the identification field includes a risk control object identification, a risk control object attribute identification and the spatiotemporal information of the risk control object, the risk control object identification refers to identification information that can uniquely identify the risk control object, including ID, telephone number, and certificate number, the risk control object attribute identification refers to attribute information of the risk control object, including age, gender, address, and education, the spatiotemporal information includes time information and IP address of the risk control object, the time field includes a fixed time window period or a sliding time window period, the indicator field is used to indicate the content of interest in the matched sample data, and the calculation function is used to calculate the statistical value of the indicator field; The method of performing feature aggregation processing on the sample data in the sample set by using the feature mining paradigm to obtain a feature set containing multi-dimensional aggregate features includes: Obtain the pre-constructed feature mining paradigm, which includes a main dimension aggregate feature mining paradigm, a main and auxiliary dimension aggregate feature mining paradigm, and a main group dimension aggregate feature mining paradigm, wherein the main dimension aggregate feature mining paradigm = risk control object identification field + time field + indicator field + calculation function, the main and auxiliary dimension aggregate feature mining paradigm = risk control object identification field + risk control object attribute identification field + time field + indicator field + calculation function, and the main group dimension aggregate feature mining paradigm = risk control object identification field + risk control object spatiotemporal information field + time field + indicator field + calculation function; perform feature mining and aggregation processing on the sample data in the sample set according to the feature mining paradigm, and obtain the main dimension aggregate feature, main and auxiliary dimension aggregate feature, and main group dimension aggregate feature corresponding to each sample data; The performing feature mining and aggregation processing on the sample data in the sample set according to the feature mining paradigm includes: According to the identification field and time field in the feature mining paradigm and the identification information and time information of the sample data, target sample data matching the identification field and the time field is obtained from the sample set; according to the calculation function, the statistical value of the indicator field in the target sample data is obtained; according to the statistical value of the target sample data, the multi-dimensional aggregation features corresponding to the target sample data are obtained.

2. The fraud pattern recognition model generation method according to claim 1, characterized in that: The time field includes a fixed time window period or a sliding time window period; The indicator field includes at least one of a numerical field, a category field, and a relationship field. The calculation function corresponding to the numerical field includes a data distribution feature calculation function. The calculation function corresponding to the category field includes a counting function and / or a proportion function. The calculation function corresponding to the relationship field includes a whether function and / or a counting function.

3. The fraud pattern recognition model generation method according to claim 1, characterized in that: The preset feature screening rule includes at least one feature screening index selected from among an information importance index, a coverage index, a stability index, and an iterative model subset index. The feature screening of the feature set by using the preset feature screening rule includes: Obtain each feature screening index value corresponding to each multi-dimensional aggregation feature in the feature set; According to all feature screening index values ​​corresponding to each multi-dimensional aggregate feature, the feature expression capability value corresponding to each multi-dimensional aggregate feature is obtained; Feature screening is performed based on the feature expression capability value corresponding to each multi-dimensional aggregate feature.

4. The fraud pattern recognition model generation method according to any one of claims 1 to 3, characterized in that: After acquiring the latest fraud pattern recognition model, the method further includes: The latest fraud pattern recognition model is used to perform historical data backtracking to determine the online threshold and recognition result label of the latest fraud pattern recognition model.

5. A fraud pattern recognition model generation device, characterized in that: The device comprises: A sample processing unit, used to obtain a sample set, wherein the sample set continuously collects black and white sample data in a risk control scenario, wherein the sample data includes identification information and time information, wherein the identification information includes unique identification information of a risk control object corresponding to the sample data and attribute information of the risk control object, and the time information includes production time or acquisition time corresponding to the sample data; A feature processing unit, used for performing feature aggregation processing on the sample data in the sample set by using a feature mining paradigm, and obtaining a feature set containing multi-dimensional aggregated features; specifically used for obtaining the pre-constructed feature mining paradigm, which includes a main dimension aggregated feature mining paradigm, a main and auxiliary dimension aggregated feature mining paradigm, and a main group dimension aggregated feature mining paradigm, wherein the main dimension aggregated feature mining paradigm = risk control object identification field + time field + indicator field + calculation function, the main and auxiliary dimension aggregated feature mining paradigm = risk control object identification field + risk control object attribute identification field + time field + indicator field + calculation function, and the main group dimension aggregated feature mining paradigm = risk control object identification field + risk control object spatiotemporal information field + time field + indicator field + calculation function; performing feature mining and aggregation processing on the sample data in the sample set according to the feature mining paradigm, and obtaining the main dimension aggregated features, main and auxiliary dimension aggregated features, and main group dimension aggregated features corresponding to each sample data; A feature screening unit, used to perform feature screening on the feature set according to preset feature screening rules; A model generation unit is used to train multiple different types of network models using the screened features, and select the model with the best performance from the trained models as the latest fraud pattern recognition model; A paradigm construction unit, used to construct the feature mining paradigm according to an identification field, a time field, an indicator field and a calculation function, wherein the identification field and the time field are used to match sample data, the identification field includes a risk control object identification, a risk control object attribute identification and spatiotemporal information of the risk control object, the risk control object identification refers to identification information that can uniquely identify the risk control object, including ID, telephone number, and certificate number, the risk control object attribute identification refers to attribute information of the risk control object, including age, gender, address, and education, the spatiotemporal information includes time information and IP address of the risk control object, the time field includes a fixed time window period or a sliding time window period, the indicator field is used to indicate the content of interest in the matched sample data, and the calculation function is used to calculate the statistical value of the indicator field; The feature processing unit is also used to obtain target sample data matching the identification field and the time field from the sample set according to the identification field and the time field in the feature mining paradigm and the identification information and time information of the sample data; obtain the statistical value of the indicator field in the target sample data according to the calculation function; and obtain the multi-dimensional aggregation features corresponding to the target sample data according to the statistical value of the target sample data.

6. An electronic device comprising: processor; as well as A memory arranged to store computer executable instructions, which, when executed, cause the processor to execute the fraud pattern recognition model generation method as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing one or more programs, which, when executed by an electronic device including multiple application programs, enables the electronic device to execute the fraud pattern recognition model generation method described in any one of claims 1 to 4.