Risk assessment method and device based on privacy calculation and electronic equipment

Abstract

The invention provides a risk assessment method and device based on privacy calculation and electronic equipment, and the method comprises the steps: obtaining basic data corresponding to each test sample of an insurance company from a plurality of different data sources, and carrying out the privacy intersection of the basic data; screening target input data corresponding to the target variable from the privacy intersection result based on a preset screening condition, and predicting the target input data through a preset model library to obtain an initial risk score corresponding to each model; training each model based on the target input data, the initial risk score and the annotation information of the test sample set, and adjusting the parameters of each model in the training process, so as to calculate the risk probability corresponding to each test sample based on the adjusted parameters through the trained models after the training is finished, and then determining a risk assessment result corresponding to each test sample based on the risk probability. According to the invention, the accuracy of insurance risk assessment can be improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a risk assessment method, device and electronic equipment based on privacy computing. Background Art

[0002] The use of external data sources is becoming very common in traditional insurance companies, and it is becoming the mainstream in the industry for traditional insurance companies (i.e. insurance companies) to seek external data to enhance their risk assessment models. In the process of cooperation with external data sources, traditional insurance companies will provide a portion of labeled samples that encrypt the three elements (name, mobile phone number and ID number) to the data source, and let the modeling engineers of the data source provide modeling services, so as to finally obtain a risk scoring model that meets the expectations, and finally provide services to the insurance company in the form of APIs by deploying the data source as a scoring calculation node; in addition, the data source side itself will also develop some risk scoring models based on its own data advantages, so as to provide solutions without the need for the insurance company to provide sample performance, but at present, due to the different situations of each data source, the products and services provided by each data source will also be uneven.

[0003] The implementation methods of traditional insurance companies in providing risk assessment services have the following three main defects: (1) Due to data security concerns, traditional insurance companies find it difficult to provide encrypted three-factor information. Instead, they need to use privacy computing to make the existing process more secure and compliant; (2) Customization of risk assessment services is not a trend in the insurance industry. Insurance companies’ user samples are usually not shared because this involves the output of user sensitive information. (3) At present, traditional insurance companies have not effectively integrated model training technology, but traditional insurance companies actually need to better manage or even share many training parameters in some situations. The insurance industry lacks relevant implementation solutions for effectively integrating model training technology. Summary of the invention

[0004] In view of this, an object of the present invention is to provide a risk assessment method, device and electronic device based on privacy computing to alleviate the above-mentioned problems existing in the related technology.

[0005] In a first aspect, an embodiment of the present invention provides a risk assessment method based on privacy computing, including: obtaining a test sample set of an insurance company; wherein each test sample in the test sample set includes corresponding three-factor information and an insurance time, and the three-factor information includes a name, an ID number, and a mobile phone number; obtaining basic data corresponding to each test sample from multiple different data sources, and performing private intersection on the basic data; wherein the private intersection result includes input data corresponding to multiple variables; screening out target input data corresponding to a target variable from the private intersection result based on a preset screening condition, and predicting the target input data through a preset model library; wherein the preset model library includes multiple models, and the prediction result includes an initial risk score obtained by each model after predicting the target input data; training each model based on the target input data, the initial risk score, and the annotation information of the test sample set, and adjusting the parameters of each model during the training process, so as to calculate the risk probability corresponding to each test sample based on the adjusted parameters through the trained multiple models after the training ends, and then determining the risk assessment result corresponding to each test sample based on the risk probability.

[0006] In a second aspect, an embodiment of the present invention further provides a risk assessment device based on privacy computing, including: an acquisition module, configured to obtain a test sample set of an insurance company; wherein each test sample in the test sample set includes corresponding three-factor information and an insurance time, and the three-factor information includes a name, an ID number, and a mobile phone number; a private intersection module, configured to obtain basic data corresponding to each test sample from multiple different data sources, and perform private intersection on the basic data; wherein the private intersection result includes input data corresponding to multiple variables; a prediction module, configured to screen out target input data corresponding to a target variable from the private intersection result based on a preset screening condition, and predict the target input data through a preset model library; wherein the preset model library includes multiple models, and the prediction result includes an initial risk score obtained by each model after predicting the target input data; an evaluation module, configured to train each model based on the target input data, the initial risk score, and the annotation information of the test sample set, and adjust the parameters of each model during the training process, so as to calculate the risk probability corresponding to each test sample based on the adjusted parameters through the trained multiple models after the training ends, and then determining the risk assessment result corresponding to each test sample based on the risk probability.

[0007] In a third aspect, an embodiment of the present invention further provides an electronic device, including a processor and a memory, where the memory stores computer executable instructions that can be executed by the processor, and the processor executes the computer executable instructions to implement the risk assessment method based on privacy computing described in the first aspect above.

[0008] A risk assessment method, device and electronic device based on privacy computing provided by an embodiment of the present invention obtain a test sample set of an insurance company; obtain basic data corresponding to each test sample from multiple different data sources, and perform privacy intersection on the basic data; filter out target input data corresponding to target variables from the privacy intersection results based on preset screening conditions, and predict the target input data through a preset model library. The prediction results include initial risk scores obtained by each model in the preset model library after predicting the target input data; train each model based on the target input data, the initial risk scores and the annotation information of the test sample set, and adjust the parameters of each model during the training process, so as to calculate the risk probability corresponding to each test sample based on the adjusted parameters through the trained multiple models after the training ends, and then determine the risk assessment result corresponding to each test sample based on the risk probability. By adopting the above technology, the basic data corresponding to the three-element information in multiple different data sources can be utilized, the parameters of multiple models can be dynamically optimized through privacy computing, and then risk assessment can be carried out with the multiple models after optimizing the parameters, thereby improving the accuracy of insurance risk assessment.

[0009] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention are realized and obtained by the structures specifically pointed out in the specification, claims and drawings.

[0010] To make the above objectives, features and advantages of the present invention more obvious and understandable, the following specifically enumerates preferred embodiments and, in conjunction with the accompanying drawings, makes the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0012] Figure 1 It is a schematic flowchart of a risk assessment method based on privacy computing in an embodiment of the present invention;

[0013] Figure 2 It is an example diagram of realizing risk assessment based on privacy computing in an embodiment of the present invention;

[0014] Figure 3 It is a schematic structural diagram of a risk assessment device based on privacy computing in an embodiment of the present invention;

[0015] Figure 4 This is a schematic structural diagram of an electronic device in an embodiment of the present invention. Detailed implementation manners

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0017] Currently, the implementation methods of traditional insurance companies providing risk assessment services mainly have the following defects: Due to considerations of data security, traditional insurance companies are difficult to provide the encrypted three-factor information. Instead, the existing processes need to be made more secure and compliant through privacy computing methods; Customization of risk assessment services is not a trend in the insurance industry, and the user samples of insurance companies are usually not shared because this involves the output of user sensitive information; Currently, traditional insurance companies have not effectively integrated the technologies for model training. However, in some cases, traditional insurance companies actually need to better manage or even share many training parameters, and the insurance industry lacks relevant implementation solutions for effectively integrating model training technologies.

[0018] Based on this, a risk assessment method, device, and electronic device based on privacy computing provided by the embodiments of the present invention can alleviate the above problems existing in the related technologies.

[0019] To facilitate the understanding of this embodiment, a risk assessment method based on privacy computing disclosed in the embodiments of the present invention will be introduced in detail first. Refer to Figure 1 As shown, the method may include the following steps:

[0020] Step S102: Obtain a test sample set of an insurance company.

[0021] Each test sample in the test sample set may respectively include corresponding three-factor information and insurance application time, and the three-factor information may include name, ID number, and mobile phone number.

[0022] Exemplarily, an insurance company provides a batch of test samples, which may include encrypted or unencrypted three-factor information (i.e., name, ID number, mobile phone number), and at the same time include the insurance application time as the starting point of the retrospective variable. If the insurance company encrypts the three-factor information, it needs to tell the data source side what encryption method is used.

[0023] Step S104: Obtain the basic data corresponding to each test sample from multiple different data sources, and perform private intersection on the basic data.

[0024] Among them, the private intersection result may include input data corresponding to multiple variables.

[0025] For example, the data source can be the data scores of operators, social security bureaus, Internet platforms, etc. Corresponding private computing nodes can be deployed on each data source side, so that private computing can be performed through multiple private computing nodes to comprehensively evaluate the insurance application risk and improve the accuracy of evaluating the insurance application risk.

[0026] Step S106: Screen out the target input data corresponding to the target variable from the private intersection result based on a preset screening condition, and predict the target input data through a preset model library.

[0027] Among them, the preset model library may include multiple models, and the prediction result may include the initial risk scores obtained by each model after predicting the target input data respectively.

[0028] The models in the preset model library can select existing machine learning models according to actual needs, and the specific types and structures of the models in the preset model library are not limited herein.

[0029] Step S108: Train each model based on the target input data, the initial risk scores, and the annotation information of the test sample set, and adjust the parameters of each model during the training process, so that after the training is completed, the risk probability corresponding to each test sample can be calculated by the trained multiple models based on the adjusted parameters, and then the risk assessment result corresponding to each test sample is determined based on the risk probability.

[0030] A risk assessment method based on private computing provided by an embodiment of the present invention obtains a test sample set of an insurance company; obtains the basic data corresponding to each test sample from multiple different data sources, and performs private intersection on the basic data; screens out the target input data corresponding to the target variable from the private intersection result based on a preset screening condition, and predicts the target input data through a preset model library, and the prediction result includes the initial risk scores obtained by each model in the preset model library after predicting the target input data respectively; trains each model based on the target input data, the initial risk scores, and the annotation information of the test sample set, and adjusts the parameters of each model during the training process, so that after the training is completed, the risk probability corresponding to each test sample can be calculated by the trained multiple models based on the adjusted parameters, and then the risk assessment result corresponding to each test sample is determined based on the risk probability. By adopting the above technology, the basic data corresponding to the three-element information in multiple different data sources can be utilized, the parameters of multiple models can be dynamically optimized through private computing, and then risk assessment can be performed by means of multiple models with optimized parameters, thereby improving the accuracy of insurance application risk assessment.

[0031] As a possible implementation, obtaining the basic data corresponding to each test sample from multiple different data sources in step S104 above may include: for each test sample, querying each data source by using the three-element information included in the test sample, and using the query result corresponding to each data source as the corresponding basic data.

[0032] As a possible implementation, the above basic data may include the privacy information of each test sample corresponding to each data source, and the privacy information corresponding to each data source corresponds to at least one of the name, ID number, and mobile phone number included in the corresponding test sample; based on this, performing privacy intersection on the basic data in step S104 above may include: using a preset privacy intersection algorithm to perform privacy intersection on the privacy information corresponding to each data source to obtain the input data corresponding to multiple variables.

[0033] Among them, the specific variables can be customized according to the input requirements of the model, and no limitation is imposed thereon.

[0034] Continuing the previous example, after deploying the privacy computing nodes on the side of multiple data sources, each privacy computing node is deployed with a corresponding privacy intersection algorithm. The privacy information of each test sample corresponding to the corresponding data source can be obtained through each privacy computing node and the deployed privacy intersection algorithm can be used to perform privacy intersection, so as to obtain the privacy intersection result (that is, the input data corresponding to multiple variables).

[0035] As a possible implementation, the above preset screening conditions may include the coverage and / or the value ratio of the corresponding three-element information to the input data corresponding thereto; the coverage may refer to the proportion of the privacy information in the basic data queried by using the three-element information, and the value ratio may refer to the proportion of the data matching the fields corresponding to the three-element information in the basic data queried by using the three-element information. Based on this, screening out the target input data corresponding to the target variable from the privacy intersection result in step S106 above may include: screening out the target input data that meets the preset screening conditions from the privacy intersection result.

[0036] Among them, meeting the preset screening conditions may include: the coverage is greater than the preset coverage threshold, and / or, the value ratio is greater than the preset value ratio threshold.

[0037] Continuing with the previous example, after obtaining the input data corresponding to multiple variables through private intersection by multiple privacy computing nodes, the coverage and / or the value ratio of the three-element information included in each test sample corresponding to its corresponding input data can be calculated. If the coverage is greater than the preset coverage threshold and / or the value ratio is greater than the preset value ratio threshold, the input data corresponding to the corresponding privacy computing node will be input into the corresponding model for prediction to obtain the corresponding initial risk score. If the coverage is not greater than the preset coverage threshold and the value ratio is not greater than the preset value ratio threshold, the input data corresponding to the corresponding privacy computing node will not be input into the corresponding model for prediction.

[0038] As a possible implementation manner, the above annotation information can characterize whether each test sample has the risk of insurance fraud. Based on this, training each model based on the target input data, the initial risk score, and the annotation information of the test sample set in step S108 above, and adjusting the parameters of each model during the training process can include: inputting the target input data into each model for iterative training, so as to predict the target input data by each model during each training and output the corresponding risk score. After obtaining the risk score each time, adjust the parameters of each model based on the risk score, the initial risk score, and the annotation information until the training ends and stop adjusting the parameters of each model.

[0039] Continuing with the previous example, during the process of inputting the target input data into each model for iterative training, for each training, each model will predict the target input data and output the risk score corresponding to this training. The risk score output by the same model during this training can be compared with the initial risk score corresponding to the same model, and the parameters of the same model can be adjusted in combination with the score comparison result of the same model and the annotation information; until the set maximum number of iterations is reached and the training ends, at this time the parameter adjustment of each model is completed (that is, the parameters of each model reach the optimal).

[0040] As a possible implementation manner, calculating the risk probability corresponding to each test sample based on the adjusted parameters by the trained multiple models in step S108 above can include: taking the union of the adjusted parameters to obtain an optimized parameter set; calculating the risk probability corresponding to each test sample by the trained multiple models using the optimized parameter set.

[0041] Continuing with the previous example, after the training of each model in the preset model library ends, the parameters of each model have reached the optimal. The union of these optimal parameters can be obtained to get an optimal parameter set. This optimal parameter set can be used as the optimal model parameters of the preset model library as a whole. Then, each test sample is input into these trained models as a whole to calculate the risk probability corresponding to each test sample. Furthermore, the corresponding risk assessment result can be determined for each test sample using the risk probability corresponding to it.

[0042] As a possible implementation, there are the following two operation methods for determining the risk assessment result corresponding to each test sample based on the risk probability in step S108 above:

[0043] Operation method 1: Map the risk probability corresponding to each test sample into a score value within a preset score interval, and use the score value corresponding to each test sample as the corresponding risk assessment result.

[0044] Exemplarily, after obtaining the risk probability corresponding to a certain test sample, the risk probability can be linearly mapped into a score value within the interval [0, 100] as the risk assessment result of the test sample.

[0045] Operation method 2: For each test sample, if the risk probability corresponding to the test sample is greater than a preset first threshold, then assign a first value representing high risk to the test sample; if the risk probability corresponding to the test sample is less than a preset second threshold, then assign a second value representing low risk to the test sample; use the scoring result corresponding to each test sample as the corresponding risk assessment result.

[0046] Exemplarily, after obtaining the risk probability corresponding to a certain test sample, it can be determined whether the risk probability is greater than 0.8 or less than 0.2. If the risk probability is greater than 0.8, then assign 100 points to the test sample (indicating that the corresponding user belongs to a high-risk user) as the risk assessment result of the test sample; if the risk probability is less than 0.2, then assign 0 points to the test sample (indicating that the corresponding user belongs to a low-risk user) as the risk assessment result of the test sample. The insurance company directly rejects the insurance for high-risk users, while low-risk users can go through the normal insurance application process.

[0047] For the sake of easy understanding, the implementation method of the above risk assessment method based on privacy computing is described exemplarily as follows with a specific application as an example.

[0048] See Figure 2 As shown, the risk assessment based on privacy computing can be implemented by 5 main modules and 2 libraries. The 5 main modules are respectively a data preparation module, a privacy node deployment module, a privacy intersection module, a parameter-based optimization module, and an output module. The 2 libraries are respectively a model library and a parameter server library.

[0049] The first part is the data preparation module. The insurance company provides a batch of test samples, which can be encrypted information of the three elements of name, ID number, and mobile phone number, and at the same time includes the insurance application time as the starting point of the retrospective variable. The insurance company needs to tell the data source side what kind of encryption method is used.

[0050] The second part is the privacy node deployment module. In this module, privacy nodes need to be deployed on the insurance company side and multiple data source sides respectively (such as Figure 2 deploying nodes on the data source 1 to data source N sides and on the insurance company side in Figure 2 ). First, the privacy node deployed on the insurance company side is the source node. It is necessary to query multiple different data sources through the three-element information provided by the insurance company side at the source node to obtain the basic data corresponding to the three-element information (for example, the basic data of the operator obtained from data source 1, the basic data of the social security bureau obtained from data source 2, and the basic data of a certain Internet platform obtained from data source N). Deploying multiple privacy nodes can avoid the singularity of data dimensions caused by a single data source and is conducive to comprehensively judging the risks of insured customers.

[0051] The third part is the privacy intersection module. This module mainly performs privacy intersection on the basic data obtained corresponding to the test samples provided by the insurance company. After privacy intersection, the intersection sample M' is obtained. The intersection sample M' actually includes the input data corresponding to multiple variables (i.e., independent variables and dependent variables). Considering that the number of primary keys of the basic data corresponding to the three-element information in different data sources is inconsistent (for example, the primary key of the basic data of the operator in data source 1 is the mobile phone number + name, and only the ID number is the primary key for the basic data of the social security bureau in data source 2), the intersection sample M' obtained through privacy intersection can include more privacy information compared with the result obtained by performing privacy intersection on a single data source. In addition, the privacy intersection module also includes a filter, and the filtering rules can be directly set in the filter so that the privacy intersection module can filter out the target input data corresponding to the target variables with a value rate of not less than x% from the intersection sample M' through the filter. The target input data will be input into the model library to participate in subsequent calculations, and the input data with a value rate less than x% will not be input into the model library and will not participate in subsequent calculations.

[0052] The fourth part is a model library including multiple data models (such as Figure 2 models 1 to model M in Figure 2 ). This library retains the target input data corresponding to the target variables obtained through privacy intersection and filtering by the privacy intersection module. After selecting the data model, the target input data corresponding to the target variables can be input into the model library to calculate scores through each data model respectively as the initial risk scores.

[0053] The fifth part is a parameter server library including multiple parameter servers, which is used to obtain the parameters of different data models in the model library through different parameter servers, that is, each parameter server can only obtain the parameters of a single data model. The parameters of the data model can include tree depth (depth), number of leaf nodes (leaves), learning rate (learning_rate), L1 regularization term (rambda_l1), and the weights of the privacy nodes corresponding to each data source (such as Figure 2 the weights a1 to a N ).

[0054] The sixth part is an optimization module based on parameters. The parameters on which this module is based are correspondingly output by each data model in the model library and include the model parameter settings after the development and finalization of each data model. The working process of the optimization module based on parameters mainly includes: for the selected data models (such as Figure 2 models 1 to M in), the final parameters of each data model have been obtained through the parameter server library, and at the same time, the insurance company has also obtained M model scores. The insurance company needs to combine its own dependent variable information y (indicating whether there is insurance fraud, y = 1 represents a fraud risk, y = 0 represents no fraud risk) and use the obtained target input data as independent variable information to iteratively train multiple data models in the model library, and adjust the parameters of these data models during the training process. After the training is completed, an optimal parameter combination is obtained by polling the adjusted parameters of each data model and taking the union. Then, the test samples provided by the insurance company side are input into the model library, and the risk probability of each test sample is calculated by the trained data models in the model library using this optimal parameter combination. It can be seen that the working process of the optimization module based on parameters is equivalent to calculating a risk probability in a machine learning manner for a single test sample using multiple initial risk scores corresponding to multiple different data sources.

[0055] The seventh part is an output module, which converts the risk probability generated in the sixth part into a risk score from 0 to 100 through logical transformation. For example, the risk probability can be linearly mapped into a score value from 0 to 100; for another example, test samples with a risk probability greater than 0.8 can be directly rated 100 points (indicating high-risk users, and the insurance company can directly reject the insurance), and test samples with a risk probability less than 0.2 can be directly rated 0 points (indicating low-risk users, and the insurance company can go through the normal insurance application process).

[0056] The above seven parts work together to form an organic whole, from sample preparation, privacy node deployment and privacy intersection to model selection and parameter optimization, etc., forming a complete set of solutions.

[0057] By adopting the above-mentioned risk assessment method based on privacy computing, the initial risk score can be obtained by means of multiple data sources and multiple models, and the training and parameter optimization of multiple models are completed, so as to calculate a more reliable risk score by using multiple trained models, and help the insurance company make more reasonable decisions.

[0058] Based on the above-mentioned risk assessment method based on privacy computing, the embodiment of the present invention further provides a risk assessment device based on privacy computing. Refer to Figure 3 As shown, the device may include the following modules:

[0059] The data preparation module 302 is used to obtain the test sample set of the insurance company; wherein, each test sample in the test sample set includes corresponding three-factor information and the insurance time, and the three-factor information includes name, ID number and mobile phone number.

[0060] The private intersection module 304 is used to obtain the basic data corresponding to each test sample from multiple different data sources and perform private intersection on the basic data; wherein, the private intersection result includes the input data corresponding to multiple variables.

[0061] The prediction module 306 is used to screen out the target input data corresponding to the target variable from the private intersection result based on the preset screening conditions, and predict the target input data through the preset model library; wherein, the preset model library includes multiple models, and the prediction result includes the initial risk score obtained by each model after predicting the target input data.

[0062] The training and output module 308 is used to train each model based on the target input data, the initial risk score and the annotation information of the test sample set, and adjust the parameters of each model during the training process, so as to calculate the risk probability corresponding to each test sample through the multiple trained models based on the adjusted parameters after the training is completed, and then determine the risk assessment result corresponding to each test sample based on the risk probability.

[0063] The risk assessment device based on privacy computing provided by the embodiment of the present invention can utilize the basic data corresponding to the three-factor information in multiple different data sources, dynamically optimize the parameters of multiple models through privacy computing, and then perform risk assessment by means of the multiple models with optimized parameters, so as to improve the accuracy of the insured risk assessment.

[0064] The above-mentioned private intersection module 304 can also be used for: for each test sample, query each data source by using the three-factor information included in the test sample, and use the query result corresponding to each data source as the corresponding basic data.

[0065] The above basic data may include the privacy information of each test sample corresponding to each data source, and the privacy information corresponding to each data source corresponds to at least one of the name, ID number, and mobile phone number included in the corresponding test sample; based on this, the above privacy intersection module 304 may further be configured to: perform privacy intersection on the privacy information corresponding to each data source by using a preset privacy intersection algorithm to obtain input data corresponding to multiple variables.

[0066] The above preset screening conditions may include the coverage and / or the value ratio of the corresponding three-element information corresponding to the input data thereto; based on this, the above prediction module 306 may further be configured to: screen out target input data that meets the preset screening conditions from the privacy intersection result; wherein, meeting the preset screening conditions includes: the coverage is greater than a preset coverage threshold, and / or, the value ratio is greater than a preset value ratio threshold.

[0067] The above annotation information may characterize whether each test sample has the risk of insurance fraud; based on this, the above training and output module 308 may further be configured to: input the target input data into each model for iterative training, so as to predict the target input data by each model during each training and output the corresponding risk score, and adjust the parameters of each model based on the risk score, the initial risk score, and the annotation information after each risk score is obtained until the training ends and the adjustment of the parameters of each model stops.

[0068] The above training and output module 308 may further be configured to perform the following operations after the training ends: take the union of the adjusted parameters to obtain an optimized parameter set; calculate the risk probability corresponding to each test sample by using the optimized parameter set through the trained multiple models.

[0069] The above training and output module 308 may further be configured to: map the risk probability corresponding to each test sample into a score value within a preset score interval, and use the score value corresponding to each test sample as the corresponding risk assessment result.

[0070] The above training and output module 308 may further be configured to: for each test sample, if the risk probability corresponding to the test sample is greater than a preset first threshold, score the test sample with a first value representing high risk, and if the risk probability corresponding to the test sample is less than a preset second threshold, score the test sample with a second value representing low risk; use the scoring result corresponding to each test sample as the corresponding risk assessment result.

[0071] The risk assessment device based on privacy computing provided by the embodiments of the present invention has the same implementation principle and technical effects as those of the foregoing embodiments of the risk assessment method based on privacy computing. For the sake of brief description, for the parts not mentioned in the embodiments of the risk assessment device based on privacy computing, reference may be made to the corresponding content in the foregoing embodiments of the risk assessment method based on privacy computing.

[0072] The embodiments of the present invention also provide an electronic device, as Figure 4 shown, which is a schematic structural diagram of the electronic device. Among them, the electronic device includes a processor 41 and a memory 40. The memory 40 stores computer-executable instructions that can be executed by the processor 41, and the processor 41 executes the computer-executable instructions to implement the foregoing risk assessment method based on privacy computing.

[0073] In Figure 4 the illustrated embodiment, the electronic device further includes a bus 42 and a communication interface 43. Among them, the processor 41, the communication interface 43, and the memory 40 are connected through the bus 42.

[0074] Among them, the memory 40 may include a high-speed random access memory (RAM, Random Access Memory), and may also include a non-volatile memory, such as at least one disk memory. Through at least one communication interface 43 (which can be wired or wireless), a communication connection is established between the system network element and at least one other network element, and the Internet, wide area network, local area network, metropolitan area network, etc. can be used. The bus 42 may be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus 42 can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, Figure 4 only a bidirectional arrow is used in

[0075] The processor 41 may be an integrated circuit chip with signal processing capabilities. In the implementation process, the steps of the above method can be completed by the integrated logic circuit of the hardware in the processor 41 or instructions in the form of software. The above-mentioned processor 41 may be a general-purpose processor, including a central processing unit (CPU for short), a network processor (NP for short), etc.; it may also be a digital signal processor (DSP for short), an application specific integrated circuit (ASIC for short), a field-programmable gate array (FPGA for short), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present invention can be directly embodied as being executed and completed by a hardware decoding processor, or executed and completed by a combination of hardware and software modules in the decoding processor. The software module may be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. This storage medium is located in the memory, and the processor 41 reads the information in the memory and combines its hardware to complete the steps of the risk assessment method based on privacy computing in the foregoing embodiments.

[0076] Unless otherwise specifically stated, the relative steps, numerical expressions, and numerical values of the components and steps set forth in these embodiments do not limit the scope of the present invention.

[0077] If the described function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a non-volatile computer-readable storage medium executable by a processor. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc that can store program codes.

[0078] In the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation to the present invention. In addition, the terms "first", "second", and "third" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance.

[0079] Finally, it should be noted that the above-described embodiments are only specific embodiments of the present invention, which are used to illustrate the technical solutions of the present invention, rather than limiting them. The protection scope of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that any person skilled in the art within the technical scope disclosed by the present invention can still modify the technical solutions recorded in the foregoing embodiments or easily conceive of changes, or equivalently replace some of the technical features; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A risk assessment method based on privacy computing, characterized in that: include: Obtain a test sample set from an insurance company; wherein each test sample in the test sample set includes corresponding three-element information and insurance time, and the three-element information includes name, ID number and mobile phone number; Obtain basic data corresponding to each test sample from multiple different data sources, and perform privacy intersection on the basic data; wherein the privacy intersection result includes input data corresponding to multiple variables; Filtering target input data corresponding to the target variable from the privacy intersection result based on preset screening conditions, and predicting the target input data through a preset model library; wherein the preset model library includes multiple models, and the prediction results include an initial risk score obtained after each model predicts the target input data; Each model is trained based on the target input data, the initial risk score, and the annotation information of the test sample set, and the parameters of each model are adjusted during the training process, so that after the training is completed, the risk probability corresponding to each test sample is calculated based on the adjusted parameters through the trained multiple models, and then the risk assessment result corresponding to each test sample is determined based on the risk probability.

2. The risk assessment method based on privacy computing according to claim 1 is characterized in that: Obtain basic data corresponding to each test sample from multiple different data sources, including: For each test sample, each data source is queried using the three-factor information included in the test sample, so as to use the query result corresponding to each data source as the corresponding basic data.

3. The risk assessment method based on privacy computing according to claim 1 is characterized in that: The basic data includes the privacy information of each test sample corresponding to each data source, and the privacy information corresponding to each data source corresponds to at least one of the name, ID number and mobile phone number included in the corresponding test sample; The basic data is subjected to privacy protection, including: A preset privacy intersection algorithm is used to perform privacy intersection on the privacy information corresponding to each data source to obtain input data corresponding to multiple variables.

4. The risk assessment method based on privacy computing according to claim 1 is characterized in that: The preset screening conditions include the coverage and / or value rate of the corresponding three-element information corresponding to the corresponding input data; Filtering target input data corresponding to the target variable from the privacy intersection result based on preset filtering conditions includes: Filter out target input data that meets the preset screening conditions from the privacy intersection results; wherein, meeting the preset screening conditions includes: coverage greater than a preset coverage threshold, and / or, a value rate greater than a preset value rate threshold.

5. The risk assessment method based on privacy computing according to claim 1 is characterized in that: The annotation information indicates whether each test sample has an insurance fraud risk; training each model based on the target input data, the initial risk score, and the annotation information of the test sample set, and adjusting the parameters of each model during the training process, including: The target input data is input into each model for iterative training, so that each model predicts the target input data and outputs a corresponding risk score during each training. After the risk score is obtained each time, the parameters of each model are adjusted based on the risk score, the initial risk score and the annotation information until the training is completed, and the parameters of each model are stopped from being adjusted.

6. The risk assessment method based on privacy computing according to claim 1 is characterized in that: The risk probability corresponding to each test sample is calculated based on the adjusted parameters of multiple trained models, including: Take the union of the adjusted parameters to obtain the optimized parameter set; The risk probability corresponding to each test sample is calculated using the optimized parameter set through the trained multiple models.

7. The risk assessment method based on privacy computing according to claim 1 is characterized in that: Determining a risk assessment result corresponding to each test sample based on the risk probability includes: The risk probability corresponding to each test sample is mapped into a score value within a preset score range, and the score value corresponding to each test sample is used as the corresponding risk assessment result.

8. The risk assessment method based on privacy computing according to claim 1 is characterized in that: Determining a risk assessment result corresponding to each test sample based on the risk probability includes: For each test sample, if the risk probability corresponding to the test sample is greater than a preset first threshold, the test sample is scored as a first value representing a high risk; if the risk probability corresponding to the test sample is less than a preset second threshold, the test sample is scored as a second value representing a low risk; The scoring result corresponding to each test sample is used as the corresponding risk assessment result.

9. A risk assessment device based on privacy computing, characterized in that: include: A data preparation module is used to obtain a test sample set from an insurance company; wherein each test sample in the test sample set includes corresponding three-element information and insurance time, and the three-element information includes name, ID number and mobile phone number; A privacy intersection module is used to obtain basic data corresponding to each test sample from multiple different data sources and perform privacy intersection on the basic data; wherein the privacy intersection result includes input data corresponding to multiple variables; A prediction module, used to filter out target input data corresponding to the target variable from the privacy intersection result based on preset screening conditions, and predict the target input data through a preset model library; wherein the preset model library includes multiple models, and the prediction results include an initial risk score obtained after each model predicts the target input data; The training and output module is used to train each model based on the target input data, the initial risk score and the annotation information of the test sample set, and adjust the parameters of each model during the training process, so that after the training is completed, the risk probability corresponding to each test sample can be calculated based on the adjusted parameters through the trained multiple models, and then the risk assessment result corresponding to each test sample can be determined based on the risk probability.

10. An electronic device, characterized in that: It includes a processor and a memory, wherein the memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement the risk assessment method based on privacy computing as described in any one of claims 1 to 8.

Images