User risk information identification method and computer device

Abstract

Embodiments of the present application relate to the field of user risk information identification, and specifically to user risk information identification methods and computer devices. A specific implementation of the method includes: in response to determining that the target user account has been logged in on the user information processing end, extracting key content from the page content corresponding to the user information verification page on the user information processing end to generate key content of the user information page; according to the key content of the user information page generated by the user information processing end and the user portrait information set corresponding to the target user account, using a pre-trained user risk information identification model to generate user risk identification information for the key content of the user information page; and sending the user risk identification information to the user information processing end to perform a risk warning operation on the target user account. This implementation can perform multi-dimensional risk identification on users, identify and intercept high-risk users in real time during the insurance application process, and shorten the verification time.

Description

Technical Field

[0001] Embodiments of the present application relate to the field of user risk information identification, and more particularly to a method for identifying user risk information and a computer device. Background Art

[0002] In the process of insuring Internet insurance products, the insurance product provider often has a shallow understanding of the customer. It can only judge the customer quality through the insurance application materials submitted by the customer and conclude an insurance contract. Without a strict risk control system, it may lead to frequent occurrences of black and gray production, fraudulent insurance applications, malicious claims, etc., bringing a large number of adverse consequences to the subsequent services of the insurance policy and affecting the healthy development of the insurance business. Currently, for the verification of the customer's insurance application material information, the commonly used method is: manual verification. However, manual verification usually has the following technical problems: manual verification takes a long time and easily causes backlogs of material information; in addition, manual verification is subjective to a certain extent and the verification accuracy is relatively low. Summary of the Invention

[0003] This section of the application is used to briefly introduce concepts, which will be described in detail in the subsequent Detailed Description section. This section of the application is not intended to identify the key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0004] Some embodiments of the present application propose a method for identifying user risk information, a computer device, and a computer-readable storage medium to solve one or more of the technical problems mentioned in the above Background Art section.

[0005] In a first aspect, some embodiments of the present application provide a method for identifying user risk information. The method includes: in response to determining that a target user account has logged in to a user information processing terminal, extracting key content from the page content corresponding to the user information verification page in the user information processing terminal to generate key content of the user information page; according to the key content of the user information page generated by the user information processing terminal and the user portrait information set corresponding to the target user account, using a pre-trained user risk information identification model to generate user risk identification information for the key content of the user information page; and sending the user risk identification information to the user information processing terminal to perform a risk warning operation for the target user account.

[0006] In a second aspect, the present application further provides a computer device. The computer device includes a processor, a memory, and a computer program stored on the memory and executable by the processor. When the computer program is executed by the processor, the method described in any implementation manner of the first aspect is implemented.

[0007] In a third aspect, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method described in any implementation manner of the first aspect is implemented.

[0008] The above-mentioned various embodiments of the present application have the following beneficial effects: A user risk information recognition model can be used to perform multi-dimensional risk recognition on users, and risk users can be identified and intercepted in real time during the insurance application process, shortening the verification time, reducing the economic losses caused by malicious insurance applications, reducing the increase in post-service costs caused by fraudulent insurance application behaviors, and ensuring the healthy and sound development of the business. First, in response to determining that the target user account has been logged in to the user information processing terminal, key content extraction is performed on the page content corresponding to the user information verification page in the user information processing terminal to generate the key content of the user information page. Thus, the historical insurance application information (key content) of the user can be extracted. Then, according to the key content of the user information page generated by the user information processing terminal and the user portrait information set corresponding to the target user account, a pre-trained user risk information recognition model is used to generate user risk recognition information for the key content of the user information page. Thus, multi-dimensional risk recognition of users can be performed through the user risk information recognition model. Finally, the user risk recognition information is sent to the user information processing terminal to perform a risk warning operation on the target user account. Thus, multi-dimensional risk recognition of users can be performed through the user risk information recognition model, risk users can be identified and intercepted in real time during the insurance application process, shortening the verification time, reducing the economic losses caused by malicious insurance applications, reducing the increase in post-service costs caused by fraudulent insurance application behaviors, and ensuring the healthy and sound development of the business. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In combination with the accompanying drawings and with reference to the following specific embodiments, the above and other features, advantages and aspects of the various embodiments of the present application will become more apparent. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the elements and elements are not necessarily drawn to scale.

[0010] Figure 1 is a flowchart of some embodiments of the user risk information recognition method according to the present application;

[0011] Figure 2 is a schematic structural diagram of a computer device suitable for implementing some embodiments of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0012] Embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although some embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. On the contrary, these embodiments are provided to more thoroughly and completely understand the present application. It should be understood that the drawings and embodiments of the present application are only for exemplary purposes and are not used to limit the protection scope of the present application.

[0013] In addition, it should be noted that for the convenience of description, only parts related to the relevant invention are shown in the drawings. Without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0014] It should be noted that the concepts such as "first" and "second" mentioned in the present application are only used to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependent relationships.

[0015] It should be noted that the modifications of "one" and "multiple" mentioned in the present application are illustrative rather than restrictive. Those skilled in the art should understand that unless otherwise clearly specified in the context, it should be understood as "one or more".

[0016] The names of the messages or information exchanged between multiple devices in the embodiments of the present application are only for illustrative purposes and are not used to limit the scope of these messages or information.

[0017] The present application will be described in detail below with reference to the drawings and in combination with embodiments.

[0018] Figure 1 Flow 100 of some embodiments of the user risk information identification method according to the present application is shown. The user risk information identification method includes the following steps:

[0019] Step 101, in response to determining that the target user account has been logged in at the user information processing end, extract key content from the page content corresponding to the user information verification page at the user information processing end to generate key content of the user information page.

[0020] In some embodiments, the execution subject of the user risk information identification method (e.g., a computing device) may, in response to determining that the target user account has been logged in to the user information processing end, extract key content from the page content corresponding to the user information verification page in the above-mentioned user information processing end to generate the key content of the user information page. The target user account may be the user account to be insured. For example, the target user account may be the account logged in to the insurance platform in the user information processing end. The insurance platform may be a platform that supports various processing operations such as user information risk determination, insurance content viewing, and insurance purchase. The user information processing end may be a processing system that supports automated process processing. For example, the user information processing end may be Selenium. Selenium is an open-source toolset for automated testing of web applications, supports multiple browsers, and allows users to write test scripts in multiple programming languages (such as Java, Python, etc.) to simulate the operations of real users in the browser. There is a corresponding operating device for the user information processing end. The operating device provides the necessary hardware and software environment for executing automated tasks. By correctly binding, logging in, and managing on the operating device, the scheduled tasks can be executed efficiently and reliably. In practice, the operating device may be a physical device and a virtual device. The physical device may be a computer and a server, and may also be a mobile phone and a tablet computer. The virtual device may be a cloud PC, and may also be a cloud phone. The corresponding operating environment needs to be configured in advance. The operating device must meet the operating environment requirements of the RPA task, including the operating system, software dependencies, network connection, and security configuration, etc. In the cloud environment, it is ensured that the resources of the virtual device (such as CPU, memory, storage) can meet the requirements of the insurance process processing task. The operating device needs to be continuously monitored to ensure the stable operation of the insurance process processing task. Any hardware or software failure needs to be processed in a timely manner. The management operations of the relevant devices include: regular updates, maintenance, and optimization to ensure an efficient and secure operating environment. It should be noted that for the platform login of the target account, the login can be performed by managing the login information, starting the browser or application, navigating to the login page, and entering the credentials. The user information verification page may be a page in the insurance platform to be verified for user information. For example, the user information verification page may be a page that displays the various insurance information of the current user. The page content may be the specific content in the user information verification page. For example, the key content extraction may be the content corresponding to the key insurance field elements in the page.

[0021] As an example, the key content of the user information page can be generated by extracting the key content from the page content corresponding to the user information verification page according to the pre-set insurance keyword field extraction rules. The insurance keyword field extraction rules may be the extraction rules set for extracting the insurance keyword fields, and may include the field type, field name, etc.

[0022] Step 102: Based on the key content of the user information page generated by the above user information processing terminal and the user portrait information corresponding to the above target user account, use a pre-trained user risk information recognition model to generate user risk recognition information for the key content of the above user information page.

[0023] In some embodiments, the above-mentioned execution entity may generate user risk identification information for the key content of the user information page based on the key content of the user information page generated by the above-mentioned user information processing end and the user portrait information corresponding to the target user account, using a pre-trained user risk information identification model. The user risk identification information may represent the insurance application risk of the user corresponding to the target user account. For example, the user risk identification information may represent the insurance application risk of the user corresponding to the target user account. For example, the user risk identification information may indicate that: User A has a relatively high insurance application risk and needs to increase the insurance amount / or reject the insurance application for User A. Among them, the user risk information identification model may be a neural network model pre-trained with the key content of the user information page and the user portrait information as inputs and the user risk identification information as outputs. For example, the user risk information identification model may be a multi-layer feedforward neural network (MLFN), a recurrent neural network (RNN), a self-organizing neural network, or a Hopfield neural network. The user portrait information may include: gender, age, region, annual income, various historical insurance application behaviors, and various historical claim settlement behaviors. It should be noted that the user portrait information is the content that needs to be filled in when the target user account is registered and needs to be updated regularly. Multi-layer feedforward neural network (MLFN): It adopts a multi-layer local connection structure, has no feedback, the neuron function usually takes the Sigmoid function or RBF, generally runs in discrete time, and uses a supervised learning algorithm. This model is suitable for processing complex non-linear relationships and has good adaptability for the identification of user risk information. Recurrent neural network (RNN): It is mainly used for the identification, modeling, and control of non-linear dynamic systems. The recurrent neural network can process sequence data, which is very useful for processing user behavior data with time dependence, thereby helping to identify potential risk behaviors. Self-organizing neural network: Its main function is to achieve clustering of input feature vectors and, on this basis, to complete function approximation, classification, and pattern recognition and other mappings. This model is suitable for processing complex and non-linear user behavior data and helps to discover potential risk patterns in user behaviors. Hopfield neural network: This is a fully connected feedback network, and its operation can be carried out in continuous time or discrete time. The Hopfield neural network is suitable for processing complex optimization problems and has certain application value for the identification and prediction of user risk information. It should be noted that the user portrait information is real and verifiable portrait information. The historical claim settlement behavior information may represent the user's claim settlement behavior information. Each historical insurance application behavior and each historical claim settlement behavior information can be queried on the insurance application platform.

[0024] Among them, the user risk information identification model can be trained through the following steps:

[0025] Step 1: Determine the initial user risk information recognition model. Among them, the above-mentioned initial user risk information recognition model includes: the key content risk recognition network of the initial user information page and the risk recognition network of the initial user portrait information. The risk recognition network of the initial user portrait information includes: the initial model input layer, the initial user risk information output layer, and the loss function layer. The above-mentioned key content risk recognition network of the initial user information page includes: the initial abnormal user page key feature information generation network and the initial page key content risk recognition network.

[0026] The key content risk recognition network of the initial user information page can be a key content risk recognition network of the user information page that has not yet completed training. The key content risk recognition network of the user information page can be a neural network model that generates key content risk recognition information for the user information page. The key content risk recognition information for the user information page can be the insured risk level information obtained by identifying the key content in the user information page. The insured risk level information can represent the insured risk level of the user. For example, the insured risk level of the user can be {First level, Second level, Third level, Fourth level}. The risk corresponding to the first level is greater than the risk corresponding to the second level. The risk corresponding to the second level is greater than the risk corresponding to the third level. The risk corresponding to the third level is greater than the risk corresponding to the fourth level.

[0027] The initial user risk information output layer can be a user risk information output layer that has not yet completed training. In practice, the user risk information output layer can include: the user risk feature information generation layer, the insured risk level feature information generation layer, and the insured risk value range feature information generation layer. The user risk feature information generation layer can be a network layer that generates user risk feature information. The insured risk level feature information generation layer can be a network layer that generates insured risk level feature information. The insured risk value range feature information generation layer can be a network layer that generates insured risk value range feature information. For example, the user risk feature information generation layer can be a multi-head attention mechanism model. The insured risk level feature information generation layer can be a convolutional layer. The insured risk value range feature information generation layer can be a convolutional layer. The loss function layer can include at least one loss function.

[0028] Among them, the initial model input layer can be a model input layer that has not yet completed training. The model input layer can be the input layer of data. For example, the model input layer can be a feature information extraction layer. For example, the feature extraction layer can be a multi-layer cascaded convolutional neural network.

[0029] Second, obtain the training sample set of user risk information. Among them, the training samples of user risk information include: key content of the sample user information page, sample user portrait information, sample user value circulation information sequence, sample key content abnormal label, and sample portrait value level label. The sample user value circulation information can represent the insured value transfer behavior information of the user at a certain time. For example, the sample user value circulation information can represent the insured value information purchased by the user on the 10th. The sample key content abnormal label can represent the risk field content in the key content of the sample user information page. The sample portrait value level label can represent the value risk level corresponding to the sample user portrait information.

[0030] Third, use the training sample set of user risk information to train the above initial user risk information recognition model to obtain a trained user risk information recognition model.

[0031] Among them, the above third step can include the following sub-steps:

[0032] The first sub-step is to select a target user risk information training sample from the above training sample set of user risk information. A user risk information training sample can be randomly selected from the above training sample set of user risk information as the target user risk information training sample.

[0033] The second sub-step is to perform the following training steps based on the target user risk information training sample:

[0034] First, use the initial model input layer included in the initial user portrait information risk recognition network to generate the user portrait feature information corresponding to the sample user portrait information included in the above target user risk information training sample and the value circulation feature information corresponding to the sample user value circulation information sequence included.

[0035] Among them, the above execution entity can use the above initial model input layer to perform the following processing steps:

[0036] The first processing step is to generate the first initial user portrait feature information for the above sample user portrait information and generate the first initial user value circulation feature information corresponding to the above sample user value circulation information sequence. Among them, the first initial user portrait feature information can be the feature information obtained by fusing each portrait feature in the sample user portrait information. The first initial user portrait feature information can represent the feature semantics corresponding to each portrait feature in the sample user portrait information. The first initial user value circulation feature information can be the feature information obtained by fusing each user value circulation feature in the sample user value circulation information sequence. The first initial user value circulation feature information can represent the feature semantics corresponding to each user value circulation feature in the sample user value circulation information sequence. Both the first initial user portrait feature information and the first initial user value circulation feature information can be feature information in the form of matrix vectors. For example, the above execution entity can generate the first initial user portrait feature information and the first initial user value circulation feature information through onehot encoding.

[0037] The second processing step is to perform low-dimensional embedding processing on the above first initial user portrait feature information and the above first initial user value circulation feature information respectively to generate embedded user portrait feature information and embedded user value circulation feature information. Among them, the embedded user portrait feature information and the embedded user value circulation feature information can be feature information in the form of matrix vectors. The matrix vector dimension corresponding to the embedded user portrait feature information is smaller than the matrix vector dimension corresponding to the first initial user portrait feature information. The matrix vector dimension corresponding to the embedded user value circulation feature information is smaller than the matrix vector dimension corresponding to the first initial user value circulation feature information.

[0038] As an example, the above execution entity can use a low-dimensional embedding layer (embedding layer) to perform low-dimensional embedding processing on the above first initial user portrait feature information and the above first initial user value circulation feature information respectively to generate embedded user portrait feature information and embedded user value circulation feature information.

[0039] The third processing step is to input the above-embedded user portrait feature information into the initial squeeze-and-excitation layer included in the initial model input layer to obtain the main user portrait feature information. Among them, the main user portrait feature information can be the feature information of the main portrait features in the sample user portrait information. As an example, first, the above execution entity can perform average pooling on the embedded user portrait feature information to generate average pooling information. Second, multiply the above average pooling information by the first parameter to obtain the first multiplication information. After that, add the first multiplication information to the second parameter to obtain the first addition information. Then, input the first addition information into the RELU activation function to generate the first output result. Next, multiply the first output result by the third parameter to obtain the second multiplication result. Second, add the second multiplication result to the fourth parameter to obtain the second addition information. Finally, input the second addition information into the Sigmoid activation function to output the main user portrait feature information.

[0040] The fourth processing step is to generate user portrait feature information and value circulation feature information based on the above main user portrait feature information and the above-embedded user value circulation feature information.

[0041] Among them, the above fourth processing step may include:

[0042] 1. Generate initial embedded user value circulation feature information based on the above-embedded user value circulation feature information. As an example, the above execution entity can perform average pooling on the embedded user value circulation feature information to generate pooled user value circulation feature information as the initial embedded user value circulation feature information.

[0043] 2. Generate initial main user portrait feature information based on the above main user portrait feature information and the above-embedded user portrait feature information. As an example, first, the above execution entity can perform an exclusive NOR operation on the main user portrait feature information and the embedded user portrait feature information to generate an exclusive NOR operation result. Then, perform average pooling on the exclusive NOR operation result to generate the initial main user portrait feature information.

[0044] 3. Generate user portrait feature information and value circulation feature information based on the above initial embedded user value circulation feature information and the above initial main user portrait feature information. For example, the above execution entity can input the initial embedded user value circulation feature information and the above initial main user portrait feature information into the Bert model to generate user portrait feature information and value circulation feature information.

[0045] Second, use the above initial user risk information output layer to perform the following first processing step:

[0046] 1. Generate at least one user value characteristic information for at least one user value dimension based on the above user portrait characteristic information and the above value circulation characteristic information. Among them, the user value dimension can be the value dimension corresponding to the user. The at least one user value dimension can describe the value of the user's insurance purchase from multiple value dimensions. In practice, the at least one user value dimension can include: the dimension of the insurance purchase category that the user is interested in, and the dimension of the insurance purchase value category that the user is interested in. For example, the dimension of the insurance purchase category that the user is interested in can represent the type of insurance that the user is interested in purchasing. The dimension of the insurance purchase value category that the user is interested in can represent the category dimension of the insurance value range that the user is interested in. For example, the above execution entity can input the above user portrait characteristic information and the above value circulation characteristic information into the user risk characteristic information generation layer to generate at least one user value characteristic information for at least one user value dimension.

[0047] 2. Generate the value dimension characteristic information and the portrait value level characteristic information of the sample portrait value level label included in the training sample of the target user risk information based on the above at least one user value characteristic information. As an example, the above execution entity can use the insurance risk level characteristic information generation layer and the insurance risk value range characteristic information generation layer to generate the value dimension characteristic information and the portrait value level characteristic information of the sample portrait value level label included in the training sample of the target user risk information based on the above at least one user value characteristic information.

[0048] Third, through the loss function layer, determine the loss information between the above value dimension characteristic information and the above portrait value level characteristic information. The loss function layer can include a preset loss function. For example, the loss function can be a cross-entropy loss function or a hinge loss function.

[0049] Fourth, in response to determining that the above loss information indicates that the model training is completed, determine the initial user portrait information risk recognition network as the trained user portrait information risk recognition network.

[0050] Thus, the trained user portrait information risk recognition network can be used to identify the insurance risk corresponding to the user portrait information. Therefore, during the insurance purchase process, risk users can be screened and intercepted in real time, shortening the verification time, reducing the economic losses caused by malicious insurance purchases, reducing the increase in post-service costs caused by fraudulent insurance purchase behaviors, and ensuring the healthy and sound development of the business.

[0051] Fifth, determine the initial abnormal recognition configuration information and initial abnormal feature recognition indication information corresponding to the user information page. The initial abnormal recognition configuration information can be the abnormal recognition configuration information before each execution of the training step. The above-mentioned abnormal recognition configuration information can be information for configuring the abnormal data recognition method. The above-mentioned abnormal recognition configuration information can include a set of recognition feature types, a set of data abnormal thresholds, and a set of abnormal type recognition method identifiers. The recognition feature types in the above-mentioned set of recognition feature types can represent the types of data corresponding to the features required for abnormal data recognition. The data abnormal thresholds included in the above-mentioned set of data abnormal thresholds can be the thresholds for generating data abnormalities for the corresponding data types. The abnormal type recognition method identifiers in the set of abnormal type recognition method identifiers can be the unique identifiers of the abnormal type recognition methods. The above-mentioned abnormal type recognition method can be a classification method for detecting the abnormal types of value generation data. The above classification method can be: Lightgbm (Light Gradient Boosting Machine) machine learning model, LOF (Local Outlier Factor, local outlier factor). The above-mentioned abnormal types can be, but are not limited to, one of the following: abnormal, normal. The above-mentioned initial abnormal feature recognition guiding information can be the abnormal feature recognition guiding information before each execution of the training step. The above-mentioned abnormal feature recognition guiding information can be information for guiding the feature extraction model to extract the required features. The above-mentioned initial abnormal feature recognition guiding information can include retrieving a set of value keyword texts, a set of abnormal numerical types, a set of value verification types, and numerical fluctuation information. In practice, the above-mentioned execution entity can determine the preset abnormal recognition configuration information and preset abnormal feature recognition guiding information as the initial abnormal recognition configuration information and initial abnormal feature recognition guiding information respectively. Among them, the above-mentioned preset abnormal recognition configuration information can be the preset abnormal recognition configuration information. The above-mentioned preset abnormal feature recognition guiding information can be the preset abnormal feature recognition guiding information.

[0052] Sixth, input the key content of the sample user information page included in the target user risk information training sample into the initial abnormal user page key feature information generation network corresponding to the initial abnormal feature recognition indication information to obtain the abnormal feature information of the sample page key content. The key content of the sample user information page can include: user's historical insurance information, user's claim information (reporting insurance), user's income information, user's insurance collection information (information on which insurances are collected), user's insurance browsing information, etc. The initial abnormal user page key feature information generation network can be a feature extraction model for extracting the features guided by the initial abnormal feature recognition indication information in the key content of the sample user information page. The above-mentioned initial abnormal user page key feature information generation network can be a GPT model.

[0053] Seventh, according to the initial page key content risk identification network corresponding to the initial abnormal identification configuration information, perform content risk identification processing on the above sample page key content abnormal feature information to obtain sample page key content risk identification information. The initial abnormal user page key feature information generation network includes: a reason information generation sub-network and an abnormal type generation sub-network. Among them, the above reason information generation sub-network and can be a model for automatically generating abnormal reason information. The above abnormal reason can be the reason for the abnormality of the page key content. The above reason information generation sub-network can be a GPT model. The abnormal type generation sub-network can be: a model corresponding to at least one abnormal type identification method among the respective abnormal type identification methods corresponding to the initial abnormal identification configuration information. The respective abnormal type identification methods corresponding to the initial abnormal identification configuration information can be: the respective abnormal type identification methods corresponding to the set of abnormal type identification method identifiers included in the initial abnormal identification configuration information.

[0054] Among them, performing content risk identification processing on the above sample page key content abnormal feature information includes:

[0055] 1. Input the above sample page key content abnormal feature information into the above abnormal type generation sub-network to obtain the sample page key content abnormal type. Among them, the above sample page key content abnormal type can represent whether there is an abnormality in the page key content. The above sample page key content abnormal type can be: abnormal, no abnormality.

[0056] 2. In response to determining that the above sample page key content abnormal type meets the preset abnormal type condition, input the above sample page key content abnormal feature information into the reason information generation sub-network to obtain the sample page key content abnormal reason information. Among them, the above preset abnormal type condition can be: the sample page key content abnormal type represents that there is an abnormality in the page key content.

[0057] 3. Combine the above sample page key content abnormal type and the above sample page key content abnormal reason information into sample page key content risk identification information.

[0058] Eighth, determine the identification loss value between the corresponding sample key content abnormal label and the above sample page key content risk identification information. For example, the identification loss value between the corresponding sample key content abnormal label and the above sample page key content risk identification information can be determined through a preset loss function. The loss function can be a hinge loss function or a cross-entropy loss function.

[0059] Ninth, in response to determining that the above identification loss value is less than or equal to the preset identification loss value, determine the initial user information page key content risk identification network as the trained user information page key content risk identification network.

[0060] Thus, when the key content risk recognition network for the user information page, which has completed training, performs anomaly recognition on the key content of the user information page, the accuracy of recognizing key content with abnormal value can be improved.

[0061] Step 103: Send the above-mentioned user risk recognition information to the above-mentioned user information processing end to perform a risk warning operation on the above-mentioned target user account.

[0062] In some embodiments, the above-mentioned execution subject may send the above-mentioned user risk recognition information to the above-mentioned user information processing end to perform a risk warning operation on the above-mentioned target user account. For example, a risk level label may be assigned to the target user account to avoid insuring the user.

[0063] For another example, the risk warning operation may include:

[0064] First: Input the current insurance application data into the risk control model, score the user, and perform verification according to the risk control strategy of the current product. If the risk label score is higher than the threshold, reject.

[0065] Second: Request the risk control score from a third-party information source. If it is higher than the threshold, reject.

[0066] Third: Feed the scoring data of the third-party information source back into the risk control model system to provide decision-making indicators for subsequent scoring.

[0067] Fourth: For the rejected users, if they meet the rules for adding to the list and have not been added to the list, add them to the list rules.

[0068] This application also provides a computer device 200. As Figure 2 shown, the computer device 200 includes: a bus 201, a processor 202, a memory 203, and a communication interface 204. The processor 202, the memory 203, and the communication interface 204 communicate with each other through the bus 201. The computer device 200 may be a server or a terminal device. It should be understood that this application does not limit the number of processors and memories in the computer device 200.

[0069] The bus 201 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of easy representation, Figure 2It is represented by only one line in the figure, but it does not mean that there is only one bus or one type of bus. The bus 201 may include a path for transmitting information between various components of the computer device 200 (for example, the memory 203, the processor 202, and the communication interface 204).

[0070] The processor 202 may include any one or more of processors such as a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0071] The memory 203 may include a volatile memory, such as a random access memory (RAM). The memory 203 may also include a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD), or a solid state drive (SSD).

[0072] The memory 203 stores executable program codes, and the processor 202 executes the executable program codes to respectively implement the functions of the foregoing acquisition module, sampling module, determination module, and mixing module, so as to implement the method for identifying user risk information. That is, the memory 203 stores instructions for executing the method for identifying user risk information.

[0073] The communication interface 204 uses a transceiver module such as, but not limited to, a network interface card or a transceiver to implement communication between the computer device 200 and other devices or a communication network.

[0074] The embodiment of the present application further provides a chip, which includes a processor and a data interface. The processor reads instructions stored in a memory through the data interface to execute the method for identifying user risk information.

[0075] The embodiments of the present application also provide a computer-readable storage medium. The above computer-readable storage medium can be any available medium that a computing device can store or a data storage device such as a data center containing one or more available media. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid-state drive), etc. The computer-readable storage medium includes instructions that direct the computing device to execute the method for identifying user risk information described above.

[0076] The technical features of the above-described embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0077] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for identifying user risk information, comprising: In response to determining that the target user account has logged in to the user information processing end, extracting key content from the page content corresponding to the user information verification page in the user information processing end to generate key content of the user information page; Determining an initial user risk information recognition model, wherein the initial user risk information recognition model includes: an initial key content risk recognition network for the user information page and an initial user portrait information risk recognition network. The initial user portrait information risk recognition network includes: an initial model input layer, an initial user risk information output layer, and a loss function layer. The initial key content risk recognition network for the user information page includes: an initial abnormal user page key feature information generation network and an initial page key content risk recognition network; Obtaining a user risk information training sample set, wherein the user risk information training sample includes: sample key content of the user information page, sample user portrait information, sample user value circulation information sequence, sample key content abnormal label, and sample portrait value level label; Training the initial user risk information recognition model through the user risk information training sample set to obtain a trained user risk information recognition model; According to the key content of the user information page generated by the user information processing end and the user portrait information set corresponding to the target user account, using the pre-trained user risk information recognition model to generate user risk recognition information for the key content of the user information page; Sending the user risk recognition information to the user information processing end to perform a risk warning operation for the target user account; Wherein, the training the initial user risk information recognition model through the user risk information training sample set to obtain a trained user risk information recognition model includes: Selecting a target user risk information training sample from the user risk information training sample set; Based on the target user risk information training sample, performing the following training steps: Using the initial model input layer included in the initial user portrait information risk recognition network to generate user portrait feature information corresponding to the sample user portrait information included in the target user risk information training sample and value circulation feature information corresponding to the sample user value circulation information sequence included; Using the initial user risk information output layer to perform the following first processing step: Generating at least one user value feature information for at least one user value dimension according to the user portrait feature information and the value circulation feature information; Generating value dimension feature information and portrait value level feature information for the sample portrait value level label included in the target user risk information training sample according to the at least one user value feature information; Determining loss information between the value dimension feature information and the portrait value level feature information through the loss function layer; In response to determining that the loss information indicates that the model training is completed, determining the initial user portrait information risk recognition network as a trained user portrait information risk recognition network; Determine the initial abnormal recognition configuration information and the initial abnormal feature recognition indication information for the corresponding user information page; Input the key content of the sample user information page included in the target user risk information training sample into the initial abnormal user page key feature information generation network corresponding to the initial abnormal feature recognition indication information, and obtain the sample page key content abnormal feature information. The initial abnormal user page key feature information generation network includes: a cause information generation sub-network and an abnormal type generation sub-network; Perform content risk recognition processing on the sample page key content abnormal feature information according to the initial page key content risk recognition network corresponding to the initial abnormal recognition configuration information, and obtain the sample page key content risk recognition information; Determine the recognition loss value between the corresponding sample key content abnormal label and the sample page key content risk recognition information; In response to determining that the recognition loss value is less than or equal to the preset recognition loss value, determine the initial user information page key content risk recognition network as the trained user information page key content risk recognition network; Among them, the use of the initial model input layer included in the initial user portrait information risk recognition network to generate the user portrait feature information corresponding to the sample user portrait information included in the target user risk information training sample and the value circulation feature information corresponding to the sample user value circulation information sequence included includes: Use the initial model input layer to perform the following processing steps: Generate the first initial user portrait feature information for the sample user portrait information, and generate the first initial user value circulation feature information corresponding to the sample user value circulation information sequence; Perform low-dimensional embedding processing on the first initial user portrait feature information and the first initial user value circulation feature information respectively to generate embedded user portrait feature information and embedded user value circulation feature information; Input the embedded user portrait feature information into the initial squeeze-and-excitation layer included in the initial model input layer to obtain the main user portrait feature information; Generate user portrait feature information and value circulation feature information according to the main user portrait feature information and the embedded user value circulation feature information; Among them, the performing content risk recognition processing on the sample page key content abnormal feature information according to the initial page key content risk recognition network corresponding to the initial abnormal recognition configuration information to obtain the sample page key content risk recognition information includes: Input the sample page key content abnormal feature information into the abnormal type generation sub-network to obtain the sample page key content abnormal type; In response to determining that the sample page key content abnormal type meets the preset abnormal type condition, input the sample page key content abnormal feature information into the cause information generation sub-network to obtain the sample page key content abnormal cause information; Merge the sample page key content abnormal type and the sample page key content abnormal cause information into the sample page key content risk recognition information.

2. A computer device, wherein, The computer device includes a processor, a memory, and a computer program stored on the memory and executable by the processor, wherein when the computer program is executed by the processor, the steps of the method as claimed in claim 1 are implemented.

3. A computer-readable storage medium, wherein, A computer program is stored on the computer-readable storage medium, wherein when the computer program is executed by a processor, the steps of the method as claimed in claim 1 are implemented.

Images