Method, system, electronic device and storage medium for predicting user credit risk
By converting user behavior sequence data into time series encoding and institutional sequence encoding, combining Wavenet and DNN models, the problem of feature engineering relying on business knowledge in user credit behavior risk prediction is solved, and the generalization ability and prediction effect of the model are improved, so as to achieve Better credit risk control support.
Patent Information
- Application Number
- CN202110964860.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-08-19
- Publication Date
- 2024-12-13
- Estimated Expiration
- 2041-08-19
AI Technical Summary
In the prior art, the generalization ability of user credit behavior risk prediction models is low, resulting in poor prediction results, mainly because feature engineering relies on business domain knowledge and lacks effective solutions.
Using a method based on user behavior sequence dictionary, user behavior sequence data is converted into time series encoding vectors and institutional sequence encoding vectors. Low-dimensional dense eigenvectors are extracted through Embedding and Wavenet models, combined with DNN for prediction, and reducing dependence on business domain knowledge. , improve model generalization capabilities.
Through deep learning, the risk control model is constructed, and the regular and periodic information in user historical behavior is explored, the model's distinction ability and prediction effect are improved, the dependence of data preprocessing and feature engineering is reduced, and the effect of credit risk control is improved.
Smart Images
Figure CN113837858B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of computer technology, and in particular to a method, system, electronic device and storage medium for predicting user credit risk. Background Art
[0002] The prediction of user credit delinquency can essentially be abstracted into a binary classification modeling problem, which is the same technical framework as click-through rate prediction in the Internet advertising field, marketing response prediction, and text classification in the NLP field, but in different fields. Specifically, in the credit scenario, firstly, user profiling is performed, and data preprocessing and feature engineering are performed based on the user's static attributes, dynamic behaviors, etc., and statistics and business domain knowledge are comprehensively used to refine the features used for modeling. After a certain number of positive and negative samples are accumulated in the business, integrated learning algorithms such as logistic regression, decision trees, and gradient boosting trees, and even deep learning algorithms are used to model. Finally, the model parameters and model files are saved, and offline or online deployment is performed according to different usage scenarios to predict the risks of new users with relevant input features.
[0003] In related technologies, it is usually necessary to classify user behaviors, and then perform statistics on different types of behaviors in different time periods based on business knowledge, and describe users in many aspects from the aspects of times, frequencies, change rates, change trends, etc. to obtain user feature data. Therefore, the model effect is heavily dependent on the quality of features generated by feature engineering, and feature engineering depends on data quality and business domain knowledge, especially business domain knowledge, which is an obvious shortcoming in related technical solutions.
[0004] Currently, in related technologies, when conducting risk prediction and assessment of user credit behavior, there is still a problem of low model generalization ability and poor prediction effect due to the reliance of feature engineering on business domain knowledge. No effective solution has been proposed yet. Summary of the invention
[0005] The embodiments of the present application provide a method, system, electronic device and storage medium for predicting user credit risk, so as to at least solve the problem in the related art that when performing risk prediction and assessment on user credit behavior, the model has low generalization ability and poor prediction effect due to the reliance of feature engineering on business domain knowledge.
[0006] In a first aspect, an embodiment of the present application provides a method for predicting user credit risk, the method comprising:
[0007] According to the user behavior sequence dictionary, the user behavior sequence data is converted into a time series coding vector and an organization sequence coding vector;
[0008] The time series encoding vector and the organization sequence encoding vector are converted by Embedding to obtain a low-dimensional dense time series encoding vector and an organization sequence encoding vector, and the low-dimensional dense time series encoding vector and the organization sequence encoding vector are spliced to obtain a spliced vector;
[0009] The concatenated vector is processed by the wavenet module to extract the characteristic vector of the user behavior to obtain the target vector, and the target vector is input into the DNN module for prediction, and the target variable value to be predicted is output.
[0010] In some embodiments, the processing of the concatenated vector by a wavenet module to extract a feature vector of the user behavior to obtain a target vector includes:
[0011] The concatenated vector is expanded and convolved through the wavenet module to obtain the periodic information of the user behavior, and the long sequence information of the user behavior is obtained through multi-layer expansion, and the target vector is obtained through global maximum pooling processing.
[0012] In some embodiments, before converting the user behavior sequence data into a time series encoding vector and an organization sequence encoding vector according to the user behavior sequence dictionary, the method includes:
[0013] The user behavior sequence data in the user behavior sequence dictionary is encoded.
[0014] In some embodiments, before encoding the user behavior sequence data in the user behavior sequence dictionary, the method includes:
[0015] Obtaining the user behavior sequence dictionary and numbering it, wherein the user behavior sequence dictionary includes a partner code, all event types invoked by the partner code, and the time when the event occurs;
[0016] The event types are arranged and combined.
[0017] In some embodiments, before obtaining the user behavior sequence dictionary, the method includes:
[0018] The source data of the user behavior is obtained, wherein the source data is derived from the event call record in the partner's business process, and the event record includes the call time, the calling partner, and the business process.
[0019] In some of the embodiments, after converting the time series encoding vector and the organization sequence encoding vector by embedding to obtain low-dimensional dense time series encoding vector and organization sequence encoding vector, the method includes:
[0020] Different label data are set on the low-dimensional dense time series encoding vector and the organization sequence encoding vector to form a complete data sample for use in different binary classification business scenarios.
[0021] In a second aspect, an embodiment of the present application provides a system for predicting user credit risk, the system comprising:
[0022] The vector conversion module is used to convert the user behavior sequence data into a time series coding vector and an organization sequence coding vector according to the user behavior sequence dictionary.
[0023] The time series encoding vector and the organization sequence encoding vector are converted by Embedding to obtain a low-dimensional dense time series encoding vector and an organization sequence encoding vector, and the low-dimensional dense time series encoding vector and the organization sequence encoding vector are spliced to obtain a spliced vector;
[0024] The prediction module is used to process the concatenated vector through the wavenet module, extract the characteristic vector of the user behavior, obtain the target vector, and input the target vector into the DNN module for prediction, and output the target variable value to be predicted.
[0025] In some of the embodiments, the prediction module is also used to perform dilation convolution on the concatenated vector through a wavenet module to obtain periodic information of user behavior, obtain long sequence information of the user behavior through multi-layer dilation, and obtain the target vector through global maximum pooling processing.
[0026] In a third aspect, an embodiment of the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for predicting user credit risk as described in the first aspect above is implemented.
[0027] In a fourth aspect, an embodiment of the present application provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for predicting user credit risk as described in the first aspect above.
[0028] Compared with the related art, the method for predicting user credit risk provided by the embodiment of the present application converts user behavior sequence data into a time series coding vector and an institution sequence coding vector according to a user behavior sequence dictionary; then, the time series coding vector and the institution sequence coding vector are converted through Embedding to obtain low-dimensional dense time series coding vector and institution sequence coding vector, and the low-dimensional dense time series coding vector and the institution sequence coding vector are spliced to obtain a spliced vector; finally, the spliced vector is processed through the wavenet module to extract the feature vector of the user behavior to obtain a target vector, and the target vector is input into the DNN module for prediction, and the output is the target variable value to be predicted.
[0029] This application is based on the user's loan application behavior data, comprehensively utilizes Word Embedding and Wavenet models, uses deep learning to build a risk control model, and mines the regularity and periodicity of user historical behavior.
[0030] In the specific field of credit risk control, this application proposes a combination method based on the wavenet model, which innovatively extracts periodic and regular features and information from the user's historical loan application behavior, and feeds it into the framework of the NLP model for risk control modeling. Specifically, after obtaining the embedding of the user's historical loan application behavior data, the wavenet model is used to capture the periodic relationship between the user's global or long-term behavior, and the target feature vector is extracted, which can further enhance the model's ability to distinguish, provide support for artificial intelligence big data risk control of credit business, and improve customer credit risk control effects. In addition, this application eliminates the two steps of time-consuming data preprocessing and feature engineering that relies on business domain knowledge, and integrates data processing and model building into a deep learning network, which not only saves time, but also effectively improves the feature quality of the data and improves the model effect. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] 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:
[0032] Figure 1 is a flowchart of a method for predicting user credit risk according to an embodiment of the present application;
[0033] Figure 2 is a schematic diagram of the network structure of the wavenet module according to an embodiment of the present application;
[0034] Figure 3 is a schematic diagram of the network structure of a DNN module according to an embodiment of the present application;
[0035] Figure 4 is a structural block diagram of a system for predicting user credit risk according to an embodiment of the present application;
[0036] Figure 5 It is a schematic diagram of the internal structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION
[0037] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. Based on the embodiments provided 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. In addition, it can also be understood that although the efforts made in this development process may be complex and lengthy, for ordinary technicians in the field related to the contents disclosed in the present application, some changes such as design, manufacturing or production based on the technical contents disclosed in the present application are only conventional technical means, and should not be understood as insufficient contents disclosed in the present application.
[0038] Reference to "embodiments" in this application means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in this application may be combined with other embodiments without conflict.
[0039] Unless otherwise defined, the technical terms or scientific terms involved in this application should be understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a", "the" and the like involved in this application do not indicate a quantitative limitation, and may represent the singular or plural. The terms "include", "comprise", "have" and any of their variations involved in this application are intended to cover non-exclusive inclusions; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units that are not listed, or may also include other steps or units inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "multiple" involved in this application refers to greater than or equal to two. "And / or" describes the association relationship of associated objects, indicating that there can be three relationships, for example, "A and / or B" can represent: A exists alone, A and B exist at the same time, and B exists alone. The terms "first", "second", "third" and the like involved in the present application are merely used to distinguish similar objects and do not represent a specific ordering of the objects.
[0040] The present application embodiment provides a method for predicting user credit risk. Figure 1 is a flowchart of a method for predicting user credit risk according to an embodiment of the present application. Figure 1 As shown, the process includes the following steps:
[0041] Step S101, converting user behavior sequence data into a time series coding vector and an organization series coding vector according to the user behavior sequence dictionary;
[0042] In some of the embodiments, before converting the user behavior sequence data into a time series encoding vector and an organization sequence encoding vector according to the user behavior sequence dictionary, it is necessary to encode the user behavior sequence data in the user behavior sequence dictionary. Preferably, the present embodiment adopts a one-hot encoding method to encode the organization sequence and time series in the user behavior sequence data. For example, the behavior sequence data of a certain user is: mt, jie, wzyh, jdsc, zbank, mt, jie, wzyh, then, by querying the user behavior sequence dictionary, the number of each behavior of the user is obtained. Assuming that the numbers of mt, jie, wzyh, jdsc, and zbank in the user behavior sequence dictionary are 1, 2, 3, 4, and 5 respectively, the behavior sequence data of the above-mentioned certain user can be converted into an organization sequence encoding vector [1,2,3,4,5,1,2,3]. Similarly, assuming that the corresponding time series of the user behavior event is: Tuesday, Wednesday, Friday, Thursday, Saturday, Tuesday, Wednesday, by querying the user behavior sequence dictionary, obtain the number corresponding to each time. Assuming that the periodic event time of the event is in units of one week, Monday to Sunday are numbered 0, 1, 2, 3, 4, 5, 6, then the time series encoding vector corresponding to the user is [1, 2, 4, 3, 5, 1, 2].
[0043] In some of the embodiments, before encoding the user behavior sequence data in the user behavior sequence dictionary, it is necessary to construct a user behavior sequence dictionary, arrange and combine all possible partner codes and all possible event types that each partner code may call, obtain the user's possible behavior sequence dictionary and number them, wherein the time when the event occurred also needs to be obtained. The time series can clearly reflect the periodic feature information in the user behavior, and provide more effective feature information for subsequent feature extraction.
[0044] In some embodiments, before obtaining the user behavior sequence dictionary, the source data of the user behavior is obtained, wherein the source data is derived from an event call record in a partner's business process, and the event record includes the call time, the calling partner, the business process, etc.;
[0045] Step S102, converting the time series code vector and the organization sequence code vector through Embedding to obtain low-dimensional dense time series code vector and organization sequence code vector, and splicing the low-dimensional dense time series code vector and organization sequence code vector to obtain a spliced vector;
[0046] It should be noted that, due to the insufficient or small number of user behaviors, the time series encoding vector and the organization series encoding vector obtained by encoding in step S101 are both high-dimensional and sparse.
[0047] In this embodiment, in order to reduce the complexity of the model and facilitate the processing of subsequent models, the encoded time series coding vector and the organization sequence coding vector are respectively converted by the Embedding conversion layer to obtain two low-dimensional dense vectors, and these two low-dimensional dense time series coding vectors and organization sequence coding vectors are spliced to obtain a spliced vector.
[0048] In some of the embodiments, after converting the time series encoding vector and the organization sequence encoding vector through Embedding to obtain low-dimensional dense time series encoding vector and organization sequence encoding vector, different label data can be set on the two low-dimensional dense time series encoding vectors and organization sequence encoding vectors to form a complete data sample for use in different binary classification business scenarios. For example, in the field of marketing customer acquisition recommendation, by adding corresponding label data of whether the user is marketing to the behavioral data, it can be used to model the user's multi-scenario marketing intention and apply it to marketing recommendations for user growth business; in addition, it can also be applied to insurance intention prediction, and transaction anti-fraud based on historical behavior and other business fields;
[0049] Step S103, processing the concatenated vector through the wavenet module, extracting the characteristic vector of the user behavior, obtaining the target vector, and inputting the target vector into the DNN module for prediction, and outputting the target variable value to be predicted.
[0050] Preferably, in this embodiment, the concatenated vector is dilated and convolved through the wavenet module to obtain the periodic information of the user behavior, and then the long sequence information of the user behavior is obtained through multi-layer dilation. Finally, the target vector is obtained through global maximum pooling processing.
[0051] Figure 2 is a schematic diagram of the network structure of the wavenet module according to an embodiment of the present application, such as Figure 2As shown in FIG. 1 , first, the mechanism sequence encoding vector (None, 500, 32) and the time series encoding vector (None, 500, 32) are added through the vector addition module add_15:Add to obtain the first combined vector (None, 500, 32); then, the first combined vector is convolved through the expansion convolution layers conv1d_32:ConvID, conv1d_33:ConvID and conv1d_34:ConvID to generate the output vector, and the vector dimension is converted from (None, 500, 32) to (None, 500, 32). ne, 500, 64) to extract the periodic information of user behavior, then input the output vector of the convolutional layer conv1d_33:ConvID and conv1d_34:ConvID into the vector multiplication module multiply_10:Multiply for vector multiplication calculation, and output the second combined vector (None, 500, 64); similarly, perform vector processing of multilayer dilated convolution, vector multiplication module and vector addition module on the second combined vector to obtain the long sequence information of user behavior; finally, after global maximum pooling processing, obtain the target vector;
[0052] Furthermore, the obtained target vector is input into the DNN module for prediction, and the output is the value of the target variable to be predicted. Specifically, Figure 3 is a schematic diagram of the network structure of the DNN module according to an embodiment of the present application, such as Figure 3 As shown, first, the feature vectors [(None, 64), (None, 64), (None, 128)] processed by the maximum pooling are input into the connection module concatenate_5: Concatenate for horizontal concatenation, and the feature vector (None, 256) is output; then, the feature vector is input into the subsequent deep neural network (Deep Neural Networks, DNN for short) module for target value prediction, and the target variable value to be predicted is output, which is the probability value of the user's credit overdue in the embodiment of the application. Specifically, Figure 3It can be seen that the DNN module calculates and processes the feature vector (None, 256) through Dense_26: Dense, h1_dropout: Dropout, Dense_27: Dense, h2_dropout: Dropout, Dense_28: Dense, h3_dropout: Dropout and Dense_29: Dense, and outputs the value of the target variable to be predicted. The main function of the Dropout layer is to randomly set the parameters of some units in each training to 0 in proportion to reduce the overfitting of the model; the Dense layer is a fully connected forward network layer. In this module, the dimension of the fully connected layer keeps decreasing from 256->192->64->32->1. It should be noted that the loss function of the last layer Dense_29: Dense is a sigmoid function in order to adapt to the variable to be predicted in the binary classification scenario. The output is a vector of length 1, that is, a scalar, and the value of the target variable to be predicted is obtained.
[0053] In this embodiment, the introduced Wavenet module can use the dilated convolution mechanism to perform weighted sum calculation and quantification on the information with fixed intervals in the user's historical behavior sequence, thus achieving a periodic weighted effect. In addition, for a given same historical loan application event sequence, when the behavior codes initiated by the existing loan application events appear periodically, the Wavenet module will mine this regular information based on the interval and jump event sequence information, further refine the model's mining and insight into user behavior, and enhance the model's ability to distinguish.
[0054] Through the above steps S101 to S103, the embodiment of the present application, in the credit scenario, comprehensively utilizes WordEmbedding, wavenet and other methods, builds a risk control model through deep learning, mines the regularity and periodicity feature information in the user's historical behavior, provides support for the artificial intelligence big data risk control of the credit business, and improves the customer credit risk control effect. It solves the problem of low model generalization ability and poor prediction effect due to feature engineering relying on business domain knowledge when predicting and evaluating the risk of user credit behavior, enhances the generalization ability of the model, and improves the distinguishing effect of the model.
[0055] It should be noted that the steps shown in the above process or the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.
[0056] This embodiment also provides a system for predicting user credit risk, which is used to implement the above embodiments and preferred implementations, and will not be repeated here. As used below, the terms "module", "unit", "subunit", etc. can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.
[0057] Figure 4 is a structural block diagram of a system for predicting user credit risk according to an embodiment of the present application, such as Figure 4 As shown, the system includes a vector conversion module 41 and a prediction module 42:
[0058] The vector conversion module 41 is used to convert the user behavior sequence data into a time series coding vector and an organization sequence coding vector according to the user behavior sequence dictionary, convert the time series coding vector and the organization sequence coding vector through Embedding to obtain a low-dimensional dense time series coding vector and an organization sequence coding vector, and splice the low-dimensional dense time series coding vector and the organization sequence coding vector to obtain a spliced vector; the prediction module 42 is used to process the spliced vector through the wavenet module, extract the feature vector of the user behavior, obtain the target vector, and input the target vector into the DNN module for prediction, and output the target variable value to be predicted.
[0059] Through the above system, the embodiment of the present application, in the credit scenario, comprehensively utilizes Word Embedding, Wavenet and other methods, builds a risk control model through deep learning, mines the regularity and periodicity feature information in the user's historical behavior, provides support for the artificial intelligence big data risk control of the credit business, and improves the customer's credit risk control effect. It solves the problem of low model generalization ability and poor prediction effect due to feature engineering relying on business domain knowledge when predicting and evaluating the risk of user credit behavior, enhances the generalization ability of the model, and improves the model's differentiation effect.
[0060] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementation modes, and this embodiment will not be described in detail here.
[0061] In addition, it should be noted that the above modules can be functional modules or program modules, and can be implemented by software or hardware. For modules implemented by hardware, the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.
[0062] This embodiment further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.
[0063] Optionally, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.
[0064] In addition, in combination with the method for predicting user credit risk in the above embodiment, the embodiment of the present application can provide a storage medium for implementation. The storage medium stores a computer program; when the computer program is executed by a processor, any one of the methods for predicting user credit risk in the above embodiment is implemented.
[0065] In one embodiment, a computer device is provided, which may be a terminal. The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for predicting user credit risk is implemented. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball, or touchpad provided on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.
[0066] In one embodiment, Figure 5 is a schematic diagram of the internal structure of an electronic device according to an embodiment of the present application, such as Figure 5 As shown, an electronic device is provided, which may be a server, and its internal structure diagram may be as shown in Figure 5 As shown. The electronic device includes a processor, a network interface, an internal memory and a non-volatile memory connected through an internal bus, wherein the non-volatile memory stores an operating system, a computer program and a database. The processor is used to provide computing and control capabilities, the network interface is used to communicate with an external terminal through a network connection, the internal memory is used to provide an environment for the operation of the operating system and the computer program, the computer program is executed by the processor to implement a method for predicting user credit risk, and the database is used to store data.
[0067] Those skilled in the art will understand that Figure 5The structure shown in the figure is merely a block diagram of a partial structure related to the scheme of the present application, and does not constitute a limitation on the electronic device to which the scheme of the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.
[0068] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
[0069] Those skilled in the art should understand that the technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0070] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.
Claims
1. A method for predicting user credit risk, It is characterized in that The method comprises: According to the user behavior sequence dictionary, the user behavior sequence data is converted into a time series coding vector and an organization sequence coding vector, wherein the source data of the user behavior sequence data is the business call record in the business process of the partner; The time series encoding vector and the organization sequence encoding vector are converted by Embedding to obtain a low-dimensional dense time series encoding vector and an organization sequence encoding vector, and the low-dimensional dense time series encoding vector and the organization sequence encoding vector are spliced to obtain a spliced vector; The concatenated vector is processed by a wavenet module to extract a feature vector of the user behavior, specifically comprising: performing dilation convolution on the concatenated vector by the wavenet module to obtain periodic information of the user behavior, obtaining long sequence information of the user behavior by multi-layer dilation, and performing global maximum pooling processing to obtain a target vector; Among them, the concatenated vector is expanded and convolved by the wavenet module to obtain the periodic information of the user behavior, and the long sequence information of the user behavior is obtained by multi-layer expansion, including: The organization sequence encoding vector and the time series encoding vector are added together through a vector addition module to obtain a first combination vector; the first combination vector is convolved through an expansion convolution layer to generate an output vector; the output vector is input into a vector multiplication module for vector multiplication calculation, and a second combination vector is output; the second combination vector is subjected to vector processing through multiple layers of expansion convolution, vector multiplication module and vector addition module to obtain the long sequence information of the user behavior; the Wavenet module uses the expansion convolution mechanism to perform weighted sum calculation and quantification on the information with fixed interval time in the user's historical behavior sequence; The target vector is input into the DNN module for prediction, and the output is the target variable value to be predicted.
2. The method according to claim 1, It is characterized in that Before converting the user behavior sequence data into a time series encoding vector and an organization sequence encoding vector according to the user behavior sequence dictionary, the method includes: The user behavior sequence data in the user behavior sequence dictionary is encoded.
3. The method according to claim 2, It is characterized in that Before encoding the user behavior sequence data in the user behavior sequence dictionary, the method includes: Obtaining the user behavior sequence dictionary and numbering it, wherein the user behavior sequence dictionary includes a partner code, all event types invoked by the partner code, and the time when the event occurs; The event types are arranged and combined.
4. The method according to claim 3, It is characterized in that Before obtaining the user behavior sequence dictionary, the method includes: The source data of the user behavior is obtained, wherein the source data comes from the event call record in the partner's business process, and the event record includes the call time, the calling partner, and the business process.
5. The method according to claim 1, It is characterized in that After converting the time series encoding vector and the organization sequence encoding vector through Embedding to obtain low-dimensional dense time series encoding vector and organization sequence encoding vector, the method includes: Different label data are set on the low-dimensional dense time series encoding vector and the organization sequence encoding vector to form a complete data sample for use in different binary classification business scenarios.
6. A system for predicting user credit risk, It is characterized in that The system comprises: A vector conversion module is used to convert the user behavior sequence data into a time series coding vector and an organization sequence coding vector according to the user behavior sequence dictionary, wherein the source data of the user behavior sequence data is the business call record in the business process of the partner, The time series encoding vector and the organization sequence encoding vector are converted by Embedding to obtain a low-dimensional dense time series encoding vector and an organization sequence encoding vector, and the low-dimensional dense time series encoding vector and the organization sequence encoding vector are spliced to obtain a spliced vector; The prediction module is used to process the spliced vector through the wavenet module to extract the characteristic vector of the user behavior, specifically including: performing expansion convolution on the spliced vector through the wavenet module to obtain the periodic information of the user behavior, obtaining the long sequence information of the user behavior through multi-layer expansion, and performing global maximum pooling processing to obtain the target vector, and inputting the target vector into the DNN module for prediction, and outputting the target variable value to be predicted, wherein the periodic information of the user behavior is obtained by performing expansion convolution on the spliced vector through the wavenet module, and the long sequence information of the user behavior is obtained by multi-layer expansion. The following steps are included: The organization sequence encoding vector and the time series encoding vector are added together through a vector addition module to obtain a first combination vector; the first combination vector is convolved through an expansion convolution layer to generate an output vector; the output vector is input into a vector multiplication module for vector multiplication calculation, and a second combination vector is output; the second combination vector is subjected to vector processing of multiple layers of expansion convolution, vector multiplication module and vector addition module to obtain the long sequence information of the user behavior; the Wavenet module uses the expansion convolution mechanism to perform weighted sum calculation and quantification on the information with fixed interval time in the user's historical behavior sequence.
7. An electronic device comprising a memory and a processor, It is characterized in that A computer program is stored in the memory, and the processor is configured to run the computer program to execute the method for predicting user credit risk according to any one of claims 1 to 5.
8. A storage medium, It is characterized in that The storage medium stores a computer program, wherein the computer program is configured to execute the method for predicting user credit risk according to any one of claims 1 to 5 when running.