WO2019232892A1 - 预测投保风险概率的方法、装置、计算机设备和存储介质 - Google Patents

预测投保风险概率的方法、装置、计算机设备和存储介质 Download PDF

Info

Publication number
WO2019232892A1
WO2019232892A1 PCT/CN2018/095504 CN2018095504W WO2019232892A1 WO 2019232892 A1 WO2019232892 A1 WO 2019232892A1 CN 2018095504 W CN2018095504 W CN 2018095504W WO 2019232892 A1 WO2019232892 A1 WO 2019232892A1
Authority
WO
WIPO (PCT)
Prior art keywords
vector
personal information
insurance
risk
risk probability
Prior art date
Application number
PCT/CN2018/095504
Other languages
English (en)
French (fr)
Inventor
金戈
徐亮
肖京
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2019232892A1 publication Critical patent/WO2019232892A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0635—Risk analysis of enterprise or organisation activities
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/08—Insurance

Definitions

  • the present invention relates to the field of computer technology, and in particular, to a method, an apparatus, a computer device, and a storage medium for predicting an insurance risk probability.
  • the salesman of an insurance company needs to review the insurance application submitted by the customer, determine whether to accept the underwriting business through the audit result, and determine the insurance premium rate after accepting the underwriting business.
  • underwriters will give customers different rates according to different risk categories to ensure business quality.
  • the existing method mainly uses the experience of the salesperson to artificially screen out risk orders and determine the risk category of customers.
  • the disadvantages of this method are low efficiency, time wasted, and it is easy to miss risk orders due to human negligence, and for risk The potential relationship between the customers corresponding to the order cannot be predicted. Therefore, how to provide a method that can efficiently and accurately predict the risk probability of an insurance application is an urgent problem.
  • the main purpose of the present invention is to provide a method, device, computer equipment and storage medium that can efficiently and accurately predict the probability of insurance risks.
  • the invention proposes a method for predicting the probability of insurance risks, including:
  • a vector matrix obtained by combining the first vector and the second vector is input to a preset deep neural network-based risk probability prediction model for calculation, where the risk probability prediction model uses a specified amount of personal information of the user and Business type information, and the user's personal information and the risk probability corresponding to the business type information are obtained by training as sample data to calculate the risk probability of the user's insurance;
  • a calculation result output by the risk probability prediction model is obtained, and the result is a risk probability that the user performs insurance.
  • the device for predicting the probability of insurance risk proposed by the present invention includes:
  • An obtaining unit configured to obtain personal information and service type information of a user in an insurance policy
  • a conversion unit configured to convert the personal information into a first vector and convert the business type information into a second vector
  • a computing unit configured to combine a vector matrix obtained by combining the first vector and the second vector into a preset deep neural network-based risk probability prediction model for calculation, wherein the risk probability prediction model passes a specified amount of The personal information and business type information of the user, and the risk probability corresponding to the personal information and business type information of the user are obtained as training data to calculate the risk probability of the user's insurance;
  • a first output unit is configured to obtain a calculation result output by the risk probability prediction model, where the result is a risk probability that a user applies for insuring the business.
  • the computer device includes a memory and a processor, and the memory stores calculation-readable instructions, and is characterized in that, when the processor executes the calculation-readable instructions, implements the steps of the foregoing method.
  • the computer-readable non-volatile storage medium stores calculation-readable instructions thereon, and is characterized in that, when the calculation-readable instructions are executed by a processor, the steps of the foregoing method are implemented.
  • the beneficial effects of the present invention are: compared with the existing selection of risk sheets through human experience, the screening efficiency is low, time is wasted, and the risk sheets are easily missed due to human negligence.
  • the risk probability prediction model is used to implement It can efficiently and accurately predict the risk probability of insurance business application, and it can also greatly reduce manpower and save time.
  • FIG. 1 is a schematic diagram of steps of a method for predicting an insurance risk probability in an embodiment of the present invention
  • FIG. 2 is a schematic diagram of steps of a method for predicting an insurance risk probability in another embodiment of the present invention.
  • FIG. 3 is a schematic structural diagram of an apparatus for predicting an insurance risk probability according to an embodiment of the present invention.
  • FIG. 4 is a schematic structural diagram of a conversion unit of a device for predicting an insurance risk probability according to an embodiment of the present invention
  • FIG. 5 is a schematic structural diagram of a conversion unit of a device for predicting an insurance risk probability in another embodiment of the present invention.
  • FIG. 6 is a schematic structural diagram of a calculation unit of a device for predicting an insurance risk probability according to an embodiment of the present invention
  • FIG. 7 is a schematic structural diagram of an apparatus for predicting an insurance risk probability in another embodiment of the present invention.
  • FIG. 8 is a schematic block diagram of a computer device according to an embodiment of the present invention.
  • a method for predicting an insurance risk probability in an embodiment of the present invention includes:
  • Step S1 obtaining personal information and service type information of a user in the insurance policy
  • Step S2 converting the personal information to obtain a first vector and converting the service type information to obtain a second vector;
  • step S3 a vector matrix obtained by combining the first vector and the second vector is input to a preset deep neural network-based risk probability prediction model for calculation, where the risk probability prediction model passes a specified amount of user's Personal information and business type information, as well as the user's personal information and business type information, the risk probability corresponding to training data is used to calculate the user's insurance risk probability;
  • Step S4 Obtain a calculation result output by the risk probability prediction model, where the result is a risk probability that the user insures.
  • step S1 when the user goes to an insurance company to insure an insurance business, the salesman of the insurance company needs the user to fill in the user's own personal information and the type of business insured in the insurance policy, so that according to the user's personal information and insurance Business type information to assess the risk probability of a user when they are insured; where the user's personal information includes some of the user's private information or information about the user's participation in the insurance business, such as education level, fixed assets, job title, whether you have purchased insurance, income, age , The number of insurance purchased and the amount of insurance policy corresponding to each insurance purchased; and the business type information is different types of insurance products of insurance companies.
  • the method for predicting the insurance risk probability in this embodiment obtains the user's personal information and business type information filled in by the user in the insurance policy, which is convenient for predicting and obtaining the risk probability of the user's insurance based on the above information.
  • step S2 the personal information and business type information of the user contains text information, and the preset risk probability prediction model based on the deep neural network requires a corresponding vector to be calculated, so the personal information is converted into The first vector and the above-mentioned service type information are converted into a second vector.
  • the user's personal information is converted by a preset first rule to obtain a corresponding first vector.
  • the preset first rule is that the user's personal information contains text information, which is converted by setting different scoring parameters. A corresponding vector is obtained. For personal information that is a number, it is directly used as the corresponding vector or after scaling, it is used as the corresponding vector.
  • the above service type information is converted to a corresponding second vector through a preset second rule, wherein the preset second rule is to convert the service type information to a number according to the coding rule, and then densely process the number to obtain the corresponding vector. .
  • step S3 the first vector and the second vector are combined to obtain a vector matrix, and the vector matrix is input to a preset deep neural network-based risk probability prediction model for calculation, where the risk probability prediction model is input by A specified amount of the user's personal information and business type information, and the risk probability corresponding to the user's personal information and business type information are obtained as training data.
  • the user's personal information and business type information are input to In the risk probability prediction model, the above-mentioned risk probability prediction model will calculate the risk probability of the user insured.
  • step S4 the calculation result output by the above-mentioned risk probability prediction model is obtained, and the result is the risk probability of the user insured, so that the salesman of the insurance company can evaluate the risk probability of the user when insured according to the above-mentioned risk probability.
  • the above-mentioned risk probability prediction model can be used to efficiently and accurately predict insurance.
  • the risk probability of business application for insurance can also greatly reduce manpower and save time.
  • the personal information of the user includes discrete personal information and continuous personal information
  • the step of converting the personal information into a first vector includes:
  • Step S210 detecting discrete personal information and continuous personal information in the personal information
  • Step S211 convert the discrete personal information into a discrete vector, cross the discrete personal information to obtain a cross vector, and scale the continuous personal information to obtain a continuous vector;
  • Step S212 Combine the discrete vector, the cross vector, and the continuous vector to obtain a first vector.
  • the user's personal information includes discrete personal information and continuous personal information.
  • the above-mentioned discrete personal information mainly refers to personal information with discrete characteristics such as education level, fixed assets, professional title, whether or not insurance has been purchased;
  • the above-mentioned continuous personal information mainly refers to age Personal information with continuous characteristics such as the number of insurances purchased. For the personal information, it is necessary to detect discrete personal information and continuous personal information in the personal information.
  • the discrete personal information needs to be converted into a discrete vector according to a method of setting different scoring parameters; among them, the method of setting different scoring parameters is to set a scoring level for the input discrete personal information, The way to set the scoring level will be classified according to the population corresponding to this information.
  • the scoring parameter is set to 1 for those who have received college education or above, and the corresponding parameter is set to 0 for those who have not received college education or above;
  • the corresponding parameters are set to 1, for non-real estate, the corresponding parameters are set to 0; for those who have purchased insurance, the corresponding parameters are set to 1, and for those who have not purchased insurance, the corresponding parameters are set to 1.
  • the parameter is set to 0.
  • the preset cross scoring rule is specifically that the input discrete feature information is also constructed as a cross vector to make the discrete feature information between Can be correlated.
  • the width of the input data can be increased.
  • the preset cross-scoring rule is to set the scoring parameter to 1 for customers who meet both a college degree or higher education and an insurance purchase, and set any corresponding parameter to 0 if any of the above conditions are not met.
  • a cross vector can be constructed between the pairs in the above manner.
  • three or more types of discrete personal information can also be intersected to obtain a cross vector.
  • continuous personal information generally it can be directly used as a risk prediction model for calculation and input data.
  • continuous personal information such as income, continuous feature information such as policy amount
  • the value is generally large and can be based on preset scaling rules.
  • the continuous personal information is scaled to obtain a continuous vector. Specifically, the continuous personal information can be reduced by a reduction function to reduce the value. Avoid too much data, which will increase the calculation amount of the risk prediction model.
  • the first vector is obtained by combining the foregoing discrete vector, cross vector, and continuous vector, which is convenient as input data for a risk probability prediction model.
  • the method includes:
  • Step S2120 performing dense processing on the discrete vector and the cross vector, respectively.
  • discrete vectors and cross vectors are generally sparse. When they are directly input into the risk probability prediction model, the calculation amount of the above risk probability prediction model will be increased, and the training time will be longer. Therefore, the above-mentioned discrete vectors, Cross vectors are densely processed separately.
  • the specific means of dense processing is to perform dense processing on the above discrete vectors and cross vectors through a processing layer similar to the hidden layer function of the risk probability prediction model.
  • the step of converting the service type information to obtain a second vector includes:
  • Step S220 converting the service type information into insurance policy number information
  • Step S221 Perform a dense processing on the insurance policy number information to obtain a second vector.
  • the above-mentioned service type information is converted into insurance policy number information through a preset coding rule.
  • the preset coding rule may be OneHot coding, and different insurance service types are set to insurance policy number information by means of OneHot coding, such as For one type of insurance business, the insurance number information is set to 0000000001, for another type of insurance business, the insurance number information is set to 0000000010, and so on, all different types of insurance services can be encoded.
  • Second vector The specific method adopted is to convert the above insurance policy number data to obtain a dense process to obtain a second vector by using the same processing layer as the hidden layer function of the risk probability prediction model.
  • the vector matrix obtained by combining the first vector and the second vector is input to a preset deep neural network-based risk probability prediction model for calculation step S3, include:
  • step S31 a vector matrix obtained by combining the first vector and the second vector is input to a preset deep neural network-based risk probability prediction model and calculated to obtain a result vector and a result constant;
  • a vector matrix obtained by combining the first vector and the second vector is input to a preset deep neural network-based risk probability prediction model for calculation; the risk prediction model of the insurance business specifically includes a first input layer and three hidden layers And an output layer.
  • the result vector a and the result constant b are output.
  • the customer's discrete vector and cross vector are also input into the above-mentioned risk probability calculation formula, so that the calculated risk probability of a customer applying for the insurance business is more accurate.
  • a method for predicting an insurance risk probability in another embodiment includes:
  • Step S5 matching the risk probability with a preset risk level table, where the risk level table includes a corresponding relationship between different risk probability ranges and risk levels;
  • step S6 the risk level is output according to the matching result.
  • the risk probability output by the risk probability prediction model will be matched with a preset risk level table.
  • the above risk level table includes the corresponding relationship between different risk probability ranges and risk levels, for example, when the risk probability is in the range of 0.9 to 1. When the risk probability is between 0.6 and 0.9, it is higher risk, when the risk probability is between 0.3 and 0.6, it is general risk, and when the risk probability is between 0 and 0.3 When it is low, the corresponding risk level is output according to the matching result.
  • the method for predicting the probability of insurance risk in this embodiment, after step S6 of outputting a risk level according to a matching result includes:
  • Step S7 Find a rate corresponding to the risk level in a preset rate mapping table, where the preset rate mapping table includes a corresponding relationship between different risk levels and rates.
  • the preset rate mapping table is used to find the corresponding rate of the above risk level.
  • the preset rate mapping table includes different risk levels.
  • the corresponding relationship with the premium rate is convenient to find the corresponding premium rate in the rate mapping table according to the above-mentioned risk level, so that the user can directly calculate the insurance premium rate for insurance.
  • the apparatus for predicting an insurance risk probability in this embodiment includes:
  • An obtaining unit 10 configured to obtain personal information and service type information of a user in an insurance policy
  • a converting unit 20 configured to convert the personal information into a first vector and convert the business type information into a second vector
  • a computing unit 30 is configured to combine the first vector and the second vector to obtain a vector matrix and input the vector matrix to a preset deep neural network-based risk probability prediction model for calculation, where the risk probability prediction model passes a specified amount
  • the personal information and business type information of the user, and the risk probability corresponding to the personal information and business type information of the user are obtained as training data for calculating the risk probability of the user's insurance;
  • the first output unit 40 is configured to obtain a calculation result output by the risk probability prediction model, where the result is a risk probability of a user applying for insurance of the business.
  • the salesperson of the insurance company needs the user to fill in the user's own personal information and the type of business insured in the insurance policy. Assess the user's risk probability when applying for insurance; where the user's personal information includes some of the user's personal information or information about the user's participation in insurance business, such as education level, fixed assets, job title, whether you have purchased insurance, income, age, insurance purchased The number of copies and the amount of insurance policy corresponding to each insurance purchased; and the business type information is different types of insurance products of insurance companies.
  • the obtaining unit 10 obtains the personal information and service type information of the user filled in by the user in the insurance policy, so that it is easy to predict the risk probability of the user's insurance application based on the above information.
  • the conversion unit 20 converts the personal information into a first vector. And converting the foregoing service type information into a second vector.
  • the user's personal information is converted by a preset first rule to obtain a corresponding first vector.
  • the preset first rule is that the user's personal information contains text information, which is converted by setting different scoring parameters. A corresponding vector is obtained. For personal information that is a number, it is directly used as the corresponding vector or after scaling, it is used as the corresponding vector.
  • the above service type information is converted to a corresponding second vector through a preset second rule, wherein the preset second rule is to convert the service type information to a number according to the coding rule, and then densely process the number to obtain the corresponding vector. .
  • the computing unit 30 combines the first vector and the second vector to obtain a vector matrix, and inputs the vector matrix into a preset deep neural network-based risk probability prediction model for calculation, where the risk probability prediction model is specified by input.
  • the user ’s personal information and business type information and the risk probability corresponding to the user ’s personal information and business type information are obtained as training data. After the training is completed, the user ’s personal information and business type information are entered into the risk.
  • the probability prediction model the above-mentioned risk probability prediction model will calculate the risk probability of the user insured.
  • the first output unit 40 obtains a calculation result output by the above-mentioned risk probability prediction model, and the result is the risk probability of the user's insurance, so that the salesman of the insurance company can evaluate the risk probability of the user when applying for insurance based on the above-mentioned risk probability.
  • the above-mentioned risk probability prediction model can be used to efficiently and accurately predict insurance.
  • the risk probability of business application for insurance can also greatly reduce manpower and save time.
  • the conversion unit 20 includes:
  • a detection module 210 configured to detect discrete personal information and continuous personal information in the personal information
  • the execution module 211 is configured to convert the discrete personal information to obtain a discrete vector, cross the discrete personal information to obtain a cross vector, and perform scaling processing on the continuous personal information to obtain a continuous vector;
  • a combining module 212 is configured to combine the discrete vector, the intersection vector, and the continuous vector to obtain a first vector.
  • the user's personal information includes discrete personal information and continuous personal information.
  • the above-mentioned discrete personal information mainly refers to personal information with discrete characteristics such as education level, fixed assets, professional title, whether or not insurance has been purchased;
  • the above-mentioned continuous personal information mainly refers to age Personal information with continuous characteristics such as the number of insurances purchased.
  • the detection module 210 needs to detect discrete personal information and continuous personal information in the personal information.
  • the execution module 211 needs to convert the above-mentioned discrete personal information into a discrete vector according to a method of setting different scoring parameters; wherein, the method of setting different scoring parameters is to set the input discrete personal information.
  • the scoring parameter is set to 1 for those who have received college education or above, and the corresponding parameter is set to 0 for those who have not received college education or above;
  • the corresponding parameters are set to 1, for non-real estate, the corresponding parameters are set to 0; for those who have purchased insurance, the corresponding parameters are set to 1, and for those who have not purchased insurance, the corresponding parameters are set to 1.
  • the parameter is set to 0.
  • the execution module 211 will cross the above discrete personal information to obtain a cross vector according to a preset cross scoring rule.
  • the preset cross scoring rule is specifically that the input discrete feature information is also constructed as a cross vector to make discrete features. The information can be correlated. By adding the cross vector to the input information of the risk prediction model of the insurance business, the width of the input data can be increased. At the same time, when the cross vector is input into the risk prediction model of the insurance business for training, it can also be used. Improve the generalization ability of the model.
  • the preset cross-scoring rule is to set the scoring parameter to 1 for customers who meet both university education and higher education and have purchased insurance, and set any corresponding parameter to 0 if any of the above conditions are not met.
  • a cross vector can be constructed between the pairs in the above manner.
  • three or more types of discrete personal information can also be intersected to obtain a cross vector.
  • continuous personal information For continuous personal information, generally it can be directly used as a risk prediction model for calculation and input data.
  • the value For certain types of continuous personal information, such as income and continuous feature information, the value is generally large.
  • the shrinking rule performs a shrinking process on the continuous personal information to obtain a continuous vector.
  • the shrinking function can be used to reduce the value to reduce the value. Avoid too much data, which will increase the calculation amount of the risk prediction model.
  • the combining module 212 combines the discrete vector, the cross vector, and the continuous vector to obtain a first vector, which is convenient as input data of a risk probability prediction model.
  • the conversion unit 20 further includes:
  • the first processing module 2120 is configured to perform dense processing on the discrete vector and the cross vector, respectively.
  • discrete vectors and cross vectors are generally sparse. When they are directly input into the risk probability prediction model, the calculation amount of the above risk probability prediction model will be increased, and the training time will be longer. Therefore, the above-mentioned discrete vectors, Cross vectors are densely processed separately.
  • the first processing module 2120 performs dense processing on the above discrete vectors and cross vectors through a processing layer similar to the hidden layer function of the risk probability prediction model.
  • the conversion unit 20 further includes:
  • a conversion module 220 configured to convert the service type information into insurance policy number information
  • the second processing module 221 is configured to perform dense processing on the insurance policy number information to obtain a second vector.
  • the conversion module 220 converts the above business type information into insurance policy number information. Specifically, the conversion module 220 converts the foregoing service type information into insurance policy number information through a preset coding rule.
  • the preset coding rule may be OneHot coding, and different insurance service types are set to insurance policy numbers by means of OneHot coding. Information, for example, insurance policy number information for one type of insurance business is set to 0000000001, insurance type number information for another type of insurance business is set to 0000000010, and so on, all different types of insurance services can be encoded.
  • the second processing module 221 will convert the above The insurance policy number information is densely processed to obtain a second vector.
  • the specific method adopted is to convert the above insurance policy number data to obtain a dense process to obtain a second vector by using the same processing layer as the hidden layer function of the risk probability prediction model.
  • the calculation unit 30 includes:
  • a first calculation module 31 configured to combine the first vector and the second vector to obtain a vector matrix and input it to a preset deep neural network-based risk probability prediction model to calculate and obtain a result vector and a result constant;
  • the first calculation module 31 combines the first vector and the second vector to obtain a vector matrix, and inputs the vector matrix to a preset deep neural network-based risk probability prediction model for calculation.
  • the risk prediction model of the insurance business specifically includes a first input layer. , Three hidden layers and one output layer. After the risk prediction model of the insurance business is calculated, a result vector a and a result constant b are output.
  • the apparatus for predicting an insurance risk probability in another embodiment further includes:
  • the matching unit 50 is configured to match the risk probability with a preset risk level table, where the risk level table includes a corresponding relationship between different risk probability ranges and risk levels;
  • the second output unit 60 is configured to output a risk level according to a matching result.
  • the matching unit 50 will match the preset risk level table.
  • the above risk level table includes the corresponding relationship between different risk probability ranges and risk levels. For example, when the risk probability is 0.9 to 1, When the risk probability is between 0.6 and 0.9, it is high risk, when the risk probability is between 0.3 and 0.6, it is general risk, and when the risk probability is between 0 and When it is between 0.3, it is low risk; the second output unit 60 outputs a corresponding risk level according to the matching result.
  • the searching unit 70 is configured to search for a rate corresponding to the risk level in a preset rate mapping table, where the preset rate mapping table includes a corresponding relationship between different risk levels and rates.
  • the searching unit 70 searches the preset rate mapping table for the rate corresponding to the above risk level, and the preset rate mapping table includes The corresponding relationship between different risk levels and rates makes it easy to find the corresponding rate in the rate mapping table according to the above-mentioned risk level, so that the user's insurance rate for insurance can be directly calculated.
  • an embodiment of the present invention further provides a computer device.
  • the computer device may be a server, and its internal structure may be as shown in FIG.
  • the computer device includes a processor, a memory, a network interface, and a database connected through a system bus.
  • the computer design processor is used to provide computing and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium and an internal memory.
  • the non-volatile storage medium stores an operating system, computer-readable instructions, and a database.
  • the memory provides an environment for operating systems and computing-readable instructions in a non-volatile storage medium.
  • the database of the computer equipment is used to preset data such as a method for predicting the insurance risk probability.
  • the network interface of the computer device is used to communicate with an external terminal through a network connection.
  • the computationally readable instructions are executed by a processor to implement a method of predicting an insurance risk probability.
  • the processor executes the steps of the method for predicting an insurance risk probability: obtaining personal information and service type information of a user in an insurance policy; converting the personal information into a first vector and converting the business type information into a second vector; A vector matrix obtained by combining the first vector and the second vector is input to a preset deep neural network-based risk probability prediction model for calculation, where the risk probability prediction model inputs a specified amount of personal information and business type of a user Information, and the user ’s personal risk and risk type corresponding to the business type information are trained as sample data and used to calculate the risk probability of the user's insurance; obtain the calculation result output by the above risk probability prediction model, and the result is the user's insurance Risk probability.
  • the personal information of the user includes discrete personal information and continuous personal information
  • the step of converting the personal information into a first vector includes: detecting the discrete personal information and continuous personal information in the personal information; and The discrete personal information is converted to obtain a discrete vector, and the discrete personal information is cross-referenced to obtain a cross vector, and the continuous personal information is scaled to obtain a continuous vector; the discrete vector, the cross vector, and the continuous vector are combined to obtain a first vector.
  • the method before the step of combining the discrete vector, the cross vector, and the continuous vector to obtain the first vector, the method includes: performing dense processing on the discrete vector and the cross vector, respectively.
  • the step of converting the service type information to a second vector includes: converting the service type information into insurance policy number information; and performing dense processing on the insurance policy number information to obtain a second vector.
  • the result is the risk probability of the user insured, including: matching the risk probability with a preset risk level table, and the risk level
  • the table includes the corresponding relationship between different risk probability ranges and risk levels; the risk levels are output according to the matching results.
  • the method includes: searching a preset rate mapping table for a corresponding rate of the risk level, and the preset rate mapping table includes different risk levels and Correspondence of rates.
  • FIG. 8 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer equipment to which the solution of the present application is applied.
  • An embodiment of the present invention also provides a computer non-volatile readable storage medium, which stores a computer-readable instruction, and a method for predicting an insurance risk probability when the computer-readable instruction is executed by a processor is specifically: Obtain personal information and business type information of users in the insurance policy; convert the personal information to obtain a first vector and convert the business type information to obtain a second vector; combine the first vector and the second vector to obtain a vector matrix input
  • the calculation is performed to a preset risk probability prediction model based on a deep neural network, where the risk probability prediction model is obtained by inputting a specified amount of a user's personal information and business type information, and the user's personal information and business type information.
  • the risk probability is obtained by training as sample data, and is used to calculate the risk probability of the user's insurance application.
  • the calculation result output by the above risk probability prediction model is obtained, and the result is the risk probability of the user's insurance application.
  • the personal information of the user includes discrete personal information and continuous personal information
  • the step of converting the personal information into a first vector includes detecting the discrete personal information in the personal information. And continuous personal information; converting the discrete personal information into a discrete vector, and intersecting the discrete personal information to obtain a cross vector, and scaling the continuous personal information to obtain a continuous vector; and converting the discrete vector, the cross vector, and the continuous vector
  • the vectors are combined to obtain a first vector.
  • the method before the step of combining the discrete vector, the cross vector, and the continuous vector to obtain the first vector, the method includes: performing dense processing on the discrete vector and the cross vector, respectively.
  • the step of converting the service type information to a second vector includes: converting the service type information into insurance policy number information; and performing dense processing on the insurance policy number information to obtain a second vector.
  • the result is the risk probability of the user insured, including: matching the risk probability with a preset risk level table, and the risk level
  • the table includes the corresponding relationship between different risk probability ranges and risk levels; the risk levels are output according to the matching results.
  • the method includes: searching a preset rate mapping table for a corresponding rate of the risk level, and the preset rate mapping table includes different risk levels and Correspondence of rates.
  • Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • Volatile memory can include random access memory (RAM) or external cache memory.
  • RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
  • the risk probability prediction model can be used to achieve high efficiency. Accurately predicting the risk probability of insurance business application, it can also greatly reduce manpower and save time.

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Theoretical Computer Science (AREA)
  • Development Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • General Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • Tourism & Hospitality (AREA)
  • Quality & Reliability (AREA)
  • Operations Research (AREA)
  • Game Theory and Decision Science (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Educational Administration (AREA)
  • Technology Law (AREA)
  • Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)

Abstract

一种预测投保风险概率的方法、装置、计算机设备和存储介质，获取保险单中的用户的个人信息和业务类型信息（S1）；将个人信息转换得到第一向量以及将业务类型信息转换得到第二向量（S2）；将第一向量和第二向量进行组合得到向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算（S3）；获取用户进行投保的风险概率（S4）。

Description

预测投保风险概率的方法、装置、计算机设备和存储介质
本申请要求于2018年6月5日提交中国专利局、申请号为201810569999X，申请名称为“预测投保风险概率的方法、装置、计算机设备和存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本发明涉及到计算机技术领域，特别是涉及到一种预测投保风险概率的方法、装置、计算机设备和存储介质。
背景技术
在保险领域中，保险公司的业务员需要对客户提出的投保申请进行审核，通过审核结果来决定是否接受承保业务，并在接受承保业务后，确定保险费率。而在核保过程中，核保人员会根据不同风险类别给予客户不同的费率，保证业务质量。现有的主要通过业务员的经验来人为筛选出风险单，判断客户的风险类别，采取这种方式的缺点在于效率较低，浪费时间，而且还很容易由于人为疏忽错漏风险单，且对于风险单所对应的客户之间的潜在关系无法预测。因此如何提供一种能高效、准确地预测保险业务投保申请的风险概率的方法成为亟待解决的问题。
技术问题
本发明的主要目的为提供一种能高效、准确地预测投保风险概率的方法、装置、计算机设备和存储介质。
技术解决方案
本发明提出预测投保风险概率的方法，包括：
获取保险单中的用户的个人信息和业务类型信息；
将所述个人信息转换得到第一向量以及将所述业务类型信息转换得到第二向量；
将所述第一向量和第二向量进行组合得到向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算，其中，所述风险概率预测模型通过指定量的用户的个人信息和业务类型信息，以及用户的个人信息和业务类型信息所对应的风险概率作为样本数据进行训练所得，用于计算用户进行投保的风险概率；
获取所述风险概率预测模型输出的计算结果，该结果为用户进行投保的风险概率。
本发明提出的预测投保风险概率的装置，包括：
获取单元，用于获取保险单中的用户的个人信息和业务类型信息；
转换单元，用于将所述个人信息转换得到第一向量以及将所述业务类型信息转换得到第二向量；
计算单元，用于将所述第一向量和第二向量进行组合得到向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算，其中，所述风险概率预测模型通过指定量 的用户的个人信息和业务类型信息，以及用户的个人信息和业务类型信息所对应的风险概率作为样本数据进行训练所得，用于计算用户进行投保的风险概率；
第一输出单元，用于获取所述风险概率预测模型输出的计算结果，该结果为用户申请投保该业务的风险概率。
本发明提出的计算机设备，包括存储器和处理器，所述存储器存储有计算可读指令，其特征在于，所述处理器执行所述计算可读指令时实现上述方法的步骤。
本发明提出的计算机非易失性可读存储介质，其上存储有计算可读指令，其特征在于，所述计算可读指令被处理器执行时实现上述的方法的步骤。
有益效果
本发明的有益效果为：相较于现有的通过人为经验来选择风险单导致筛选效率低、浪费时间且还很容易由于人为疏忽错漏风险单的情况，本发明中通过上述风险概率预测模型实现能高效、准确地预测保险业务投保申请的风险概率，还能极大的减少人力，节约时间。
附图说明
图1为本发明一实施例中的预测投保风险概率的方法的步骤示意图；
图2为本发明另一实施例中的预测投保风险概率的方法的步骤示意图；
图3为本发明一实施例中的预测投保风险概率的装置的结构示意图；
图4为本发明一实施例中的预测投保风险概率的装置的转换单元的结构示意图；
图5为本发明另一实施例中的预测投保风险概率的装置的转换单元的结构示意图；
图6为本发明一实施例中的预测投保风险概率的装置的计算单元的结构示意图；
图7为本发明另一实施例中的预测投保风险概率的装置的结构示意图；
图8为本发明一实施例的计算机设备的结构示意框图。
本发明的最佳实施方式
参照图1，本发明实施例中的预测投保风险概率的方法，包括：
步骤S1，获取保险单中的用户的个人信息和业务类型信息；
步骤S2，将所述个人信息转换得到第一向量以及将所述业务类型信息转换得到第二向量；
步骤S3，将所述第一向量和第二向量进行组合得到向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算，其中，所述风险概率预测模型通过指定量的用户的个人信息和业务类型信息，以及用户的个人信息和业务类型信息所对应的风险概率作为样本数据进行训练所得，用于计算用户进行投保的风险概率；
步骤S4，获取所述风险概率预测模型输出的计算结果，该结果为用户进行投保的风险概率。
在步骤S1中，当用户到保险公司去投保某个保险业务时，保险公司的业务员需要用户在保险单中填写用户自己的个人信息以及投保的业务类型信息，从而根据用户的个人信息 以及投保的业务类型信息来评估用户投保时的风险概率；其中用户的个人信息包括一些用户的私人信息或用户参与保险业务的相关信息，例如教育水平、固定资产、职称、是否购买过保险、收入、年龄、已购买的保险份数以及购买的每份保险对应的保单金额等；而业务类型信息为保险公司不同类型的保险产品。本实施例中的预测投保风险概率的方法通过获取用户在保险单中填写的用户的个人信息和业务类型信息，便于根据上述信息预测得到用户进行投保的风险概率。
在步骤S2中，对于上述用户的个人信息以及业务类型信息中包含有文字信息，而预设的基于深度神经网络的风险概率预测模型需要输入对应的向量才能进行计算，因此将上述个人信息转换得到第一向量以及将上述业务类型信息转换得到第二向量。其中用户的个人信息通过预设的第一规则进行转换得到对应的第一向量，其中上述预设的第一规则为对于用户的个人信息中包含有文字信息，通过设置不同的评分参数将其转换得到对应的向量，对于为数字的个人信息，直接作为对应的向量或者进行放缩后作为对应的向量。上述业务类型信息通过预设的第二规则进行转换得到对应的第二向量，其中上述预设的第二规则为根据编码规则将业务类型信息转换得到数字，再将数字进行稠密处理得到对应的向量。
在步骤S3中，将上述第一向量和第二向量进行组合得到向量矩阵，将上述向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算，其中上述风险概率预测模型通过输入指定量的用户的个人信息和业务类型信息、以及用户的个人信息和业务类型信息所对应的风险概率作为样本数据进行训练所得，在训练完成后，当将用户的个人信息和业务类型信息输入到风险概率预测模型中，上述风险概率预测模型将计算用户进行投保的风险概率。
在步骤S4中，获取上述风险概率预测模型输出的计算结果，该结果为用户进行投保的风险概率，从而使得保险公司的业务员能根据上述风险概率来评估用户在进行投保时的风险概率，相较于现有的通过人为经验来选择风险单导致筛选效率低、浪费时间且还很容易由于人为疏忽错漏风险单的情况，本实施例中通过上述风险概率预测模型实现能高效、准确地预测保险业务投保申请的风险概率，还能极大的减少人力，节约时间。
本实施例中的预测投保风险概率的方法，所述用户的个人信息包括离散个人信息以及连续个人信息，所述将所述个人信息转换得到第一向量的步骤，包括：
步骤S210，检测出所述个人信息中的离散个人信息以及连续个人信息；
步骤S211，将所述离散个人信息转换得到离散向量，并且将所述离散个人信息进行交叉得到交叉向量，以及将所述连续个人信息进行放缩处理得到连续向量；
步骤S212，将所述离散向量、交叉向量以及连续向量进行组合得到第一向量。
用户的个人信息包括离散个人信息和连续个人信息，其中上述离散个人信息主要指的是教育水平、固定资产、职称、是否购买过保险等具有离散特征的个人信息；上述连续个人信息主要指的年龄、已购买过的保险份数等具有连续特征的个人信息。对于上述个人信息中，需要检测出上述个人信息中的离散个人信息以及连续个人信息。对于检测得到的上 述离散个人信息，需要根据设置不同的评分参数的方法将上述离散个人信息转换得到离散向量；其中，设置不同的评分参数的方法为对于输入的离散个人信息均会设置评分等级，其中设置评分等级的方式将依据这些信息对应的人群进行分类。具体的说，对于输入的用户的教育水平，对于接受过大学本科及以上高等教育的将其评分参数设置为1，若没有接受过大学本科及以上高等教育的将其对应的参数设置为0；同理，对于固定资产中有房产的将对应的参数设置为1，无房产的将对应的参数设置为0；对于购买过保险的将对应的参数设置为1，未购买过保险的将对应的参数设置为0。以此类推，将所有用户的离散特征信息转换得到对应的离散向量。
此外，还将根据预设的交叉评分规则将上述离散个人信息进行交叉得到交叉向量，其中预设的交叉评分规则具体为对于输入的离散特征信息还通过构造成交叉向量来使得离散特征信息之间能进行关联，通过增加交叉向量到保险业务的风险预测模型的输入信息中，既能增加输入数据的宽度，同时在将交叉向量输入到保险业务的风险预测模型进行训练时，还能提高模型的泛化能力。其中预设的交叉评分规则为，对于同时满足大学本科及以上高等教育和购买过保险的客户，将其评分参数设置为1，上述任意一个条件不满足则将对应的参数设置为0。同理，对于多种不同类型的离散个人信息，可以两两之间通过上述方式构造交叉向量。优选地，还可以将三种或多种类型的离散个人信息进行交叉得到交叉向量，其具体方式参照上述方法，在此不再赘述。
对于连续个人信息，一般可以直接作为风险预测模型进行计算输入数据，对于某些类型的连续个人信息，例如收入、保单金额等连续特征信息，其数值一般较大，可以根据预设的放缩规则将上述连续个人信息进行放缩处理得到连续向量，具体的说，可以适当通过缩小函数进行缩小，以减小其数值。避免其数据量过大，从而增大风险预测模型的计算量。
将上述离散向量、交叉向量以及连续向量进行组合得到第一向量，便于作为风险概率预测模型的输入数据。
本实施例中的所述的预测投保风险概率的方法，所述将所述离散向量、交叉向量以及连续向量进行组合得到第一向量的步骤S212之前，包括：
步骤S2120，对所述离散向量、交叉向量分别进行稠密处理。
需要指出的是，离散向量、交叉向量一般比较稀疏，当将其直接输入到风险概率预测模型中，会提高上述风险概率预测模型的计算量，使得训练时间较长，因此需要对上述离散向量、交叉向量分别进行稠密处理。其中稠密处理的具体手段为通过与风险概率预测模型的隐藏层功能一样处理层对上述离散向量、交叉向量进行稠密处理。
本实施例中的预测投保风险概率的方法，所述将所述业务类型信息转换得到第二向量的步骤，包括：
步骤S220，将所述业务类型信息转换为保险单号信息；
步骤S221，对所述保险单号信息进行稠密处理得到第二向量。
对于业务类型信息在输入到风险概率预测模型中进行计算之前，需要对不同的保险业 务类型进行区分。具体的，通过预设的编码规则将上述业务类型信息转换为保险单号信息，预设的编码规则可以为OneHot编码，通过OneHot编码的方式将不同的保险业务类型设置为保险单号信息，例如对于某一类型保险业务的保险单号信息设置为0000000001，另外一种类型的险业务的保险单号信息设置为0000000010，以此类推，可以将所有不同类型的保险业务进行编码。需要指出的是，对于由于该数据比较稀疏，当其直接输入到保险业务的风险预测模型中，会提高保险业务的风险预测模型的计算量，使得训练时间较长，需要经过处理转换得到稠密的第二向量。其中采取的具体手段为通过与风险概率预测模型的隐藏层功能一样处理层对上述通保险单号数据进行转换得到稠密处理得到第二向量。
本实施例中的预测投保风险概率的方法，所述将所述第一向量和第二向量进行组合得到向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算的步骤S3，包括：
步骤S31，将所述第一向量和第二向量进行组合得到向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算得到结果向量和结果常数；
步骤S32，通过公式
Figure PCTCN2018095504-appb-000001
计算得到风险概率，其中
Figure PCTCN2018095504-appb-000002
均为权重常数，a为所述结果向量，b为所述结果常数，φ(x)为所述交叉向量，x为所述离散向量，P(Y＝1|x)为所述风险概率。
将第一向量和第二向量进行组合得到向量矩阵输入到预设的基于深度神经网络的风险概率预测模型中进行计算；其中上述保险业务的风险预测模型具体包括第一输入层，三个隐藏层和一个输出层。其中上述保险业务的风险预测模型进行计算后，将输出得到结果向量a和结果常数b。为了准确的预测出结果，通过公式
Figure PCTCN2018095504-appb-000003
计算得到风险概率，其中
Figure PCTCN2018095504-appb-000004
均为权重常数，a为所述结果向量，b为所述结果常数，φ(x)为所述交叉向量，x为所述离散向量，P(Y＝1|x)为所述风险概率。在本公式中还将该客户的离散向量和交叉向量输入到上述风险概率的计算公式中，使得计算得到的某客户申请该保险业务时的风险概率更加准确。
参照图2，另一实施例中的预测投保风险概率的方法，所述获取所述风险概率预测模型输出的计算结果，该结果为用户进行投保的风险概率的步骤S4之后，包括：
步骤S5，将所述风险概率与预设的风险等级表进行匹配，所述风险等级表包括不同风险概率范围与风险等级的对应关系；
步骤S6，根据匹配结果输出风险等级。
对于风险概率预测模型输出的风险概率，会将与预设的风险等级表进行匹配，上述风险等级表包括不同风险概率范围与风险等级的对应关系，例如当风险概率处于0.9到1的范围之间时，则为高风险，而当风险概率处于0.6到0.9之间，则为较高风险，当风险概率处于0.3到0.6之间时，则为一般风险，而当风险概率处于0到0.3之间时，则为低风险；根据匹配结果输出对应的风险等级。
本实施例中的预测投保风险概率的方法，所述根据匹配结果输出风险等级的步骤S6之后，包括：
步骤S7，在预设的费率映射表中查找所述风险等级对应的费率，所述预设的费率映射表包括不同风险等级与费率的对应关系。
对于得到的风险等级，为了计算出该风险等级的保险产品的费率，会在预设的费率映射表中查找上述风险等级对应的费率，上述预设的费率映射表包括不同风险等级与费率的对应关系，便于根据上述风险等级在费率映射表查找得到对应的费率，从而能直接计算出用户进行投保的保险费率。
参照图3，本实施例中的预测投保风险概率的装置，包括：
获取单元10，用于获取保险单中的用户的个人信息和业务类型信息；
转换单元20，用于将所述个人信息转换得到第一向量以及将所述业务类型信息转换得到第二向量；
计算单元30，用于将所述第一向量和第二向量进行组合得到向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算，其中，所述风险概率预测模型通过指定量的用户的个人信息和业务类型信息，以及用户的个人信息和业务类型信息所对应的风险概率作为样本数据进行训练所得，用于计算用户进行投保的风险概率；
第一输出单元40，用于获取所述风险概率预测模型输出的计算结果，该结果为用户申请投保该业务的风险概率。
当用户到保险公司去投保某个保险业务时，保险公司的业务员需要用户在保险单中填写用户自己的个人信息以及投保的业务类型信息，从而根据用户的个人信息以及投保的业务类型信息来评估用户投保时的风险概率；其中用户的个人信息包括一些用户的私人信息或用户参与保险业务的相关信息，例如教育水平、固定资产、职称、是否购买过保险、收入、年龄、已购买的保险份数以及购买的每份保险对应的保单金额等；而业务类型信息为保险公司不同类型的保险产品。获取单元10获取用户在保险单中填写的用户的个人信息和业务类型信息，便于根据上述信息预测得到用户进行投保的风险概率。
对于上述用户的个人信息以及业务类型信息中包含有文字信息，而预设的基于深度神经网络的风险概率预测模型需要输入对应的向量才能进行计算，转换单元20将上述个人信息转换得到第一向量以及将上述业务类型信息转换得到第二向量。其中用户的个人信息通过预设的第一规则进行转换得到对应的第一向量，其中上述预设的第一规则为对于用户的个人信息中包含有文字信息，通过设置不同的评分参数将其转换得到对应的向量，对于为数字的个人信息，直接作为对应的向量或者进行放缩后作为对应的向量。上述业务类型信息通过预设的第二规则进行转换得到对应的第二向量，其中上述预设的第二规则为根据编码规则将业务类型信息转换得到数字，再将数字进行稠密处理得到对应的向量。
计算单元30将上述第一向量和第二向量进行组合得到向量矩阵，并将上述向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算，其中上述风险概率预测模 型通过输入指定量的用户的个人信息和业务类型信息、以及用户的个人信息和业务类型信息所对应的风险概率作为样本数据进行训练所得，在训练完成后，当将用户的个人信息和业务类型信息输入到风险概率预测模型中，上述风险概率预测模型将计算用户进行投保的风险概率。
第一输出单元40获取上述风险概率预测模型输出的计算结果，该结果为用户进行投保的风险概率，从而使得保险公司的业务员能根据上述风险概率来评估用户在进行投保时的风险概率，相较于现有的通过人为经验来选择风险单导致筛选效率低、浪费时间且还很容易由于人为疏忽错漏风险单的情况，本实施例中通过上述风险概率预测模型实现能高效、准确地预测保险业务投保申请的风险概率，还能极大的减少人力，节约时间。
参照图4，本实施例中的预测投保风险概率的装置，所述转换单元20，包括：
检测模块210，用于检测出所述个人信息中的离散个人信息以及连续个人信息；
执行模块211，用于将所述离散个人信息转换得到离散向量，并且将所述离散个人信息进行交叉得到交叉向量，以及将所述连续个人信息进行放缩处理得到连续向量；
组合模块212，用于将所述离散向量、交叉向量以及连续向量进行组合得到第一向量。
用户的个人信息包括离散个人信息和连续个人信息，其中上述离散个人信息主要指的是教育水平、固定资产、职称、是否购买过保险等具有离散特征的个人信息；上述连续个人信息主要指的年龄、已购买过的保险份数等具有连续特征的个人信息。对于上述个人信息中，检测模块210需要检测出上述个人信息中的离散个人信息以及连续个人信息。对于检测得到的上述离散个人信息，执行模块211需要根据设置不同的评分参数的方法将上述离散个人信息转换得到离散向量；其中，设置不同的评分参数的方法为对于输入的离散个人信息均会设置评分等级，其中设置评分等级的方式将依据这些信息对应的人群进行分类。具体的说，对于输入的用户的教育水平，对于接受过大学本科及以上高等教育的将其评分参数设置为1，若没有接受过大学本科及以上高等教育的将其对应的参数设置为0；同理，对于固定资产中有房产的将对应的参数设置为1，无房产的将对应的参数设置为0；对于购买过保险的将对应的参数设置为1，未购买过保险的将对应的参数设置为0。以此类推，将所有用户的离散特征信息转换得到对应的离散向量。
此外，执行模块211还将根据预设的交叉评分规则将上述离散个人信息进行交叉得到交叉向量，其中预设的交叉评分规则具体为对于输入的离散特征信息还通过构造成交叉向量来使得离散特征信息之间能进行关联，通过增加交叉向量到保险业务的风险预测模型的输入信息中，既能增加输入数据的宽度，同时在将交叉向量输入到保险业务的风险预测模型进行训练时，还能提高模型的泛化能力。其中预设的交叉评分规则为对于同时满足大学本科及以上高等教育和购买过保险的客户，将其评分参数设置为1，上述任意一个条件不满足则将对应的参数设置为0。同理，对于多种不同类型的离散个人信息，可以两两之间通过上述方式构造交叉向量。优选地，还可以将三种或多种类型的离散个人信息进行交叉得到交叉向量，其具体方式参照上述方法，在此不再赘述。
对于连续个人信息，一般可以直接作为风险预测模型进行计算输入数据，对于某些类型的连续个人信息，例如收入、保单金额等连续特征信息，其数值一般较大，执行模块211根据预设的放缩规则将上述连续个人信息进行放缩处理得到连续向量，具体的说，可以适当通过缩小函数进行缩小，以减小其数值。避免其数据量过大，从而增大风险预测模型的计算量。
组合模块212将上述离散向量、交叉向量以及连续向量进行组合得到第一向量，便于作为风险概率预测模型的输入数据。
本实施例中的预测投保风险概率的装置，所述转换单元20，还包括：
第一处理模块2120，用于对所述离散向量、交叉向量分别进行稠密处理。
需要指出的是，离散向量、交叉向量一般比较稀疏，当将其直接输入到风险概率预测模型中，会提高上述风险概率预测模型的计算量，使得训练时间较长，因此需要对上述离散向量、交叉向量分别进行稠密处理。其中第一处理模块2120通过与风险概率预测模型的隐藏层功能一样处理层对上述离散向量、交叉向量进行稠密处理。
参照图5，另一实施例中的预测投保风险概率的装置，所述转换单元20，还包括：
转换模块220，用于将所述业务类型信息转换为保险单号信息；
第二处理模块221，用于对所述保险单号信息进行稠密处理得到第二向量。
对于业务类型信息在输入到风险概率预测模型中进行计算之前，需要对不同的保险业务类型进行区分，转换模块220将上述业务类型信息转换为保险单号信息。具体的，转换模块220通过预设的编码规则将上述业务类型信息转换为保险单号信息，预设的编码规则可以为OneHot编码，通过OneHot编码的方式将不同的保险业务类型设置为保险单号信息，例如对于某一类型保险业务的保险单号信息设置为0000000001，另外一种类型的险业务的保险单号信息设置为0000000010，以此类推，可以将所有不同类型的保险业务进行编码。需要指出的是，对于由于该数据比较稀疏，当其直接输入到保险业务的风险预测模型中，会提高保险业务的风险预测模型的计算量，使得训练时间较长，第二处理模块221将上述保险单号信息进行稠密处理得到第二向量。其中采取的具体手段为通过与风险概率预测模型的隐藏层功能一样处理层对上述通保险单号数据进行转换得到稠密处理得到第二向量。
参照图6，本实施例中的预测投保风险概率的装置，所述计算单元30，包括：
第一计算模块31，用于将所述第一向量和第二向量进行组合得到向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算得到结果向量和结果常数；
第二计算模块32，用于通过公式
Figure PCTCN2018095504-appb-000005
计算得到风险概率，其中
Figure PCTCN2018095504-appb-000006
均为权重常数，a为所述结果向量，b为所述结果常数，φ(x)为所述交叉向量，x为所述离散向量，P(Y＝1|x)为所述风险概率。
第一计算模块31将第一向量和第二向量进行组合得到向量矩阵输入到预设的基于深度神经网络的风险概率预测模型中进行计算；其中上述保险业务的风险预测模型具体包括第一输入层，三个隐藏层和一个输出层。其中上述保险业务的风险预测模型进行计算后， 将输出得到结果向量a和结果常数b。为了准确的预测出结果，第二计算模块32通过公式
Figure PCTCN2018095504-appb-000007
计算得到风险概率，其中
Figure PCTCN2018095504-appb-000008
均为权重常数，a为所述结果向量，b为所述结果常数，φ(x)为所述交叉向量，x为所述离散向量，P(Y＝1|x)为所述风险概率。在本公式中还将该客户的离散向量和交叉向量输入到上述风险概率的计算公式中，使得计算得到的某客户申请该保险业务时的风险概率更加准确。
参照图7，另一实施例中的预测投保风险概率的装置，还包括：
匹配单元50，用于将所述风险概率与预设的风险等级表进行匹配，所述风险等级表包括不同风险概率范围与风险等级的对应关系；
第二输出单元60，用于根据匹配结果输出风险等级。
对于风险概率预测模型输出的风险概率，匹配单元50会将与预设的风险等级表进行匹配，上述风险等级表包括不同风险概率范围与风险等级的对应关系，例如当风险概率处于0.9到1的范围之间时，则为高风险，而当风险概率处于0.6到0.9之间，则为较高风险，当风险概率处于0.3到0.6之间时，则为一般风险，而当风险概率处于0到0.3之间时，则为低风险；第二输出单元60根据匹配结果输出对应的风险等级。
本实施例中的预测投保风险概率的装置，还包括：
查找单元70，用于在预设的费率映射表中查找所述风险等级对应的费率，所述预设的费率映射表包括不同风险等级与费率的对应关系。
对于得到的风险等级，为了计算出该风险等级的保险产品的费率，查找单元70会在预设的费率映射表中查找上述风险等级对应的费率，上述预设的费率映射表包括不同风险等级与费率的对应关系，便于根据上述风险等级在费率映射表查找得到对应的费率，从而能直接计算出用户进行投保的保险费率。
参照图8，本发明实施例中还提供一种计算机设备，该计算机设备可以是服务器，其内部结构可以如图8所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口和数据库。其中，该计算机设计的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统、计算可读指令和数据库。该内存器为非易失性存储介质中的操作系统和计算可读指令的运行提供环境。该计算机设备的数据库用于预设预测投保风险概率的方法等数据。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算可读指令被处理器执行时以实现预测投保风险概率的方法。
上述处理器执行上述预测投保风险概率的方法的步骤：获取保险单中的用户的个人信息和业务类型信息；将上述个人信息转换得到第一向量以及将上述业务类型信息转换得到第二向量；将上述第一向量和第二向量进行组合得到向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算，其中，上述风险概率预测模型通过输入指定量的用户的个人信息和业务类型信息，以及用户的个人信息和业务类型信息所对应的风险概率作为样本数据进行训练所得，用于计算用户进行投保的风险概率；获取上述风险概率预测模 型输出的计算结果，该结果为用户进行投保的风险概率。
上述计算机设备，上述用户的个人信息包括离散个人信息以及连续个人信息，上述将上述个人信息转换得到第一向量的步骤，包括：检测出上述个人信息中的离散个人信息以及连续个人信息；将上述离散个人信息转换得到离散向量，并且将上述离散个人信息进行交叉得到交叉向量，以及将上述连续个人信息进行放缩处理得到连续向量；将上述离散向量、交叉向量以及连续向量进行组合得到第一向量。
在一个实施例中，上述将上述离散向量、交叉向量以及连续向量进行组合得到第一向量的步骤之前，包括：对上述离散向量、交叉向量分别进行稠密处理。
在一个实施例中，上述将上述第一向量和第二向量进行组合得到向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算的步骤，包括：将上述第一向量和第二向量进行组合得到向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算得到结果向量和结果常数；通过公式
Figure PCTCN2018095504-appb-000009
计算得到风险概率，其中
Figure PCTCN2018095504-appb-000010
均为权重常数，a为所述结果向量，b为所述结果常数，φ(x)为所述交叉向量，x为所述离散向量，P(Y＝1|x)为所述风险概率。
在一个实施例中，上述将上述业务类型信息转换得到第二向量的步骤，包括：将上述业务类型信息转换为保险单号信息；对上述保险单号信息进行稠密处理得到第二向量。
在一个实施例中，上述获取上述风险概率预测模型输出的计算结果，该结果为用户进行投保的风险概率的步骤之后，包括：将上述风险概率与预设的风险等级表进行匹配，上述风险等级表包括不同风险概率范围与风险等级的对应关系；根据匹配结果输出风险等级。
在一个实施例中，上述根据匹配结果输出风险等级的步骤之后，包括：在预设的费率映射表中查找上述风险等级对应的费率，上述预设的费率映射表包括不同风险等级与费率的对应关系。
本领域技术人员可以理解，图8中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定。
本发明一实施例还提供一种计算机非易失性可读存储介质，其上存储有计算可读指令，计算可读指令被处理器执行时实现一种预测投保风险概率的方法，具体为：获取保险单中的用户的个人信息和业务类型信息；将上述个人信息转换得到第一向量以及将上述业务类型信息转换得到第二向量；将上述第一向量和第二向量进行组合得到向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算，其中，上述风险概率预测模型通过输入指定量的用户的个人信息和业务类型信息，以及用户的个人信息和业务类型信息所对应的风险概率作为样本数据进行训练所得，用于计算用户进行投保的风险概率；获取上述风险概率预测模型输出的计算结果，该结果为用户进行投保的风险概率。
上述计算机非易失性可读存储介质，上述用户的个人信息包括离散个人信息以及连续个人信息，上述将上述个人信息转换得到第一向量的步骤，包括：检测出上述个人信息中 的离散个人信息以及连续个人信息；将上述离散个人信息转换得到离散向量，并且将上述离散个人信息进行交叉得到交叉向量，以及将上述连续个人信息进行放缩处理得到连续向量；将上述离散向量、交叉向量以及连续向量进行组合得到第一向量。
在一个实施例中，上述将上述离散向量、交叉向量以及连续向量进行组合得到第一向量的步骤之前，包括：对上述离散向量、交叉向量分别进行稠密处理。
在一个实施例中，上述将上述第一向量和第二向量进行组合得到向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算的步骤，包括：将上述第一向量和第二向量进行组合得到向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算得到结果向量和结果常数；通过公式
Figure PCTCN2018095504-appb-000011
计算得到风险概率，其中
Figure PCTCN2018095504-appb-000012
均为权重常数，a为所述结果向量，b为所述结果常数，φ(x)为所述交叉向量，x为所述离散向量，P(Y＝1|x)为所述风险概率。
在一个实施例中，上述将上述业务类型信息转换得到第二向量的步骤，包括：将上述业务类型信息转换为保险单号信息；对上述保险单号信息进行稠密处理得到第二向量。
在一个实施例中，上述获取上述风险概率预测模型输出的计算结果，该结果为用户进行投保的风险概率的步骤之后，包括：将上述风险概率与预设的风险等级表进行匹配，上述风险等级表包括不同风险概率范围与风险等级的对应关系；根据匹配结果输出风险等级。
在一个实施例中，上述根据匹配结果输出风险等级的步骤之后，包括：在预设的费率映射表中查找上述风险等级对应的费率，上述预设的费率映射表包括不同风险等级与费率的对应关系。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算可读指令来指令相关的硬件来完成，所述的计算可读指令可存储与一非易失性计算机可读取存储介质中，该计算可读指令在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的和实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可以包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限，RAM一多种形式可得，诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双速据率SDRAM(SSRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
综上所述，相较于现有的通过人为经验来选择风险单导致筛选效率低、浪费时间且还很容易由于人为疏忽错漏风险单的情况，本发明中通过上述风险概率预测模型实现能高效、准确地预测保险业务投保申请的风险概率，还能极大的减少人力，节约时间。
以上所述仅为本发明的优选实施例，并非因此限制本发明的专利范围，凡是利用本发 明说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本发明的专利保护范围内。

Claims (20)

  1. 一种预测投保风险概率的方法，其特征在于，包括：
    获取保险单中的用户的个人信息和业务类型信息；
    将所述个人信息转换得到第一向量以及将所述业务类型信息转换得到第二向量；
    将所述第一向量和第二向量进行组合得到向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算，其中，所述风险概率预测模型通过输入指定量的用户的个人信息和业务类型信息，以及用户的个人信息和业务类型信息所对应的风险概率作为样本数据进行训练所得，用于计算用户进行投保的风险概率；
    获取所述风险概率预测模型输出的计算结果，该结果为用户进行投保的风险概率。
  2. 根据权利要求1所述的预测投保风险概率的方法，其特征在于，所述用户的个人信息包括离散个人信息以及连续个人信息，所述将所述个人信息转换得到第一向量的步骤，包括：
    检测出所述个人信息中的离散个人信息以及连续个人信息；
    将所述离散个人信息转换得到离散向量，并且将所述离散个人信息进行交叉得到交叉向量，以及将所述连续个人信息进行放缩处理得到连续向量；
    将所述离散向量、交叉向量以及连续向量进行组合得到第一向量。
  3. 根据权利要求2所述的预测投保风险概率的方法，其特征在于，所述将所述离散向量、交叉向量以及连续向量进行组合得到第一向量的步骤之前，包括：
    对所述离散向量、交叉向量分别进行稠密处理。
  4. 根据权利要求2所述的预测投保风险概率的方法，其特征在于，所述将所述第一向量和第二向量进行组合得到向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算的步骤，包括：
    将所述第一向量和第二向量进行组合得到向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算得到结果向量和结果常数；
    通过公式
    Figure PCTCN2018095504-appb-100001
    计算得到风险概率，其中
    Figure PCTCN2018095504-appb-100002
    Figure PCTCN2018095504-appb-100003
    均为权重常数，a为所述结果向量，b为所述结果常数，φ(x)为所述交叉向量，x为所述离散向量，P(Y＝1|x)为所述风险概率。
  5. 根据权利要求1所述的预测投保风险概率的方法，其特征在于，所述将所述业务类型信息转换得到第二向量的步骤，包括：
    将所述业务类型信息转换为保险单号信息；
    对所述保险单号信息进行稠密处理得到第二向量。
  6. 根据权利要求1所述的预测投保风险概率的方法，其特征在于，所述获取所述风险概率预测模型输出的计算结果，该结果为用户进行投保的风险概率的步骤之后，包括：
    将所述风险概率与预设的风险等级表进行匹配，所述风险等级表包括不同风险概率范 围与风险等级的对应关系；
    根据匹配结果输出风险等级。
  7. 根据权利要求6所述的预测投保风险概率的方法，其特征在于，所述根据匹配结果输出风险等级的步骤之后，包括：
    在预设的费率映射表中查找所述风险等级对应的费率，所述预设的费率映射表包括不同风险等级与费率的对应关系。
  8. 一种预测投保风险概率的装置，其特征在于，包括：
    获取单元，用于获取保险单中的用户的个人信息和业务类型信息；
    转换单元，用于将所述个人信息转换得到第一向量以及将所述业务类型信息转换得到第二向量；
    计算单元，用于将所述第一向量和第二向量进行组合得到向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算，其中，所述风险概率预测模型通过指定量的用户的个人信息和业务类型信息，以及用户的个人信息和业务类型信息所对应的风险概率作为样本数据进行训练所得，用于计算用户进行投保的风险概率；
    第一输出单元，用于获取所述风险概率预测模型输出的计算结果，该结果为用户申请投保该业务的风险概率。
  9. 根据权利要求8所述的预测投保风险概率的装置，其特征在于，所述转换单元，包括：
    检测模块，用于检测出所述个人信息中的离散个人信息以及连续个人信息；
    执行模块，用于将所述离散个人信息转换得到离散向量，并且将所述离散个人信息进行交叉得到交叉向量，以及将所述连续个人信息进行放缩处理得到连续向量；
    组合模块，用于将所述离散向量、交叉向量以及连续向量进行组合得到第一向量。
  10. 根据权利要求9所述的预测投保风险概率的装置，其特征在于，所述转换单元，还包括：
    第一处理模块，用于对所述离散向量、交叉向量分别进行稠密处理。
  11. 根据权利要求9所述的预测投保风险概率的装置，其特征在于，所述计算单元，包括：
    第一计算模块，用于将所述第一向量和第二向量进行组合得到向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算得到结果向量和结果常数；
    第二计算模块，用于通过公式
    Figure PCTCN2018095504-appb-100004
    计算得到风险概率，其中
    Figure PCTCN2018095504-appb-100005
    均为权重常数，a为所述结果向量，b为所述结果常数，φ(x)为所述交叉向量，x为所述离散向量，P(Y＝1|x)为所述风险概率。
  12. 根据权利要求8所述的预测投保风险概率的装置，其特征在于，所述转换单元，还包括：
    转换模块，用于将所述业务类型信息转换为保险单号信息；
    第二处理模块，用于对所述保险单号信息进行稠密处理得到第二向量。
  13. 根据权利要求8所述的预测投保风险概率的装置，其特征在于，所述预测投保风险概率的装置，还包括：
    匹配单元，用于将所述风险概率与预设的风险等级表进行匹配，所述风险等级表包括不同风险概率范围与风险等级的对应关系；
    第二输出单元，用于根据匹配结果输出风险等级。
  14. 根据权利要求13所述的预测投保风险概率的装置，其特征在于，所述预测投保风险概率的装置，还包括：
    查找单元，用于在预设的费率映射表中查找所述风险等级对应的费率，所述预设的费率映射表包括不同风险等级与费率的对应关系。
  15. 一种计算机设备，包括存储器和处理器，所述存储器存储有计算可读指令，其特征在于，所述处理器执行所述计算可读指令时实现预测投保风险概率的方法，该方法包括：
    获取保险单中的用户的个人信息和业务类型信息；
    将所述个人信息转换得到第一向量以及将所述业务类型信息转换得到第二向量；
    将所述第一向量和第二向量进行组合得到向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算，其中，所述风险概率预测模型通过输入指定量的用户的个人信息和业务类型信息，以及用户的个人信息和业务类型信息所对应的风险概率作为样本数据进行训练所得，用于计算用户进行投保的风险概率；
    获取所述风险概率预测模型输出的计算结果，该结果为用户进行投保的风险概率。
  16. 根据权利要求1所述的计算机设备，其特征在于，所述用户的个人信息包括离散个人信息以及连续个人信息，所述将所述个人信息转换得到第一向量的步骤，包括：
    检测出所述个人信息中的离散个人信息以及连续个人信息；
    将所述离散个人信息转换得到离散向量，并且将所述离散个人信息进行交叉得到交叉向量，以及将所述连续个人信息进行放缩处理得到连续向量；
    将所述离散向量、交叉向量以及连续向量进行组合得到第一向量。
  17. 根据权利要求16所述的计算机设备，其特征在于，所述将所述离散向量、交叉向量以及连续向量进行组合得到第一向量的步骤之前，包括：
    对所述离散向量、交叉向量分别进行稠密处理。
  18. 一种计算机非易失性可读存储介质，其上存储有计算可读指令，其特征在于，所述计算可读指令被处理器执行时实现预测投保风险概率的方法，该方法包括：
    获取保险单中的用户的个人信息和业务类型信息；
    将所述个人信息转换得到第一向量以及将所述业务类型信息转换得到第二向量；
    将所述第一向量和第二向量进行组合得到向量矩阵输入至预设的基于深度神经网络的风险概率预测模型中进行计算，其中，所述风险概率预测模型通过输入指定量的用户的个人信息和业务类型信息，以及用户的个人信息和业务类型信息所对应的风险概率作为样 本数据进行训练所得，用于计算用户进行投保的风险概率；
    获取所述风险概率预测模型输出的计算结果，该结果为用户进行投保的风险概率。
  19. 根据权利要求18所述的计算机非易失性可读存储介质，其特征在于，所述用户的个人信息包括离散个人信息以及连续个人信息，所述将所述个人信息转换得到第一向量的步骤，包括：
    检测出所述个人信息中的离散个人信息以及连续个人信息；
    将所述离散个人信息转换得到离散向量，并且将所述离散个人信息进行交叉得到交叉向量，以及将所述连续个人信息进行放缩处理得到连续向量；
    将所述离散向量、交叉向量以及连续向量进行组合得到第一向量。
  20. 根据权利要求19所述的计算机非易失性可读存储介质，其特征在于，所述将所述离散向量、交叉向量以及连续向量进行组合得到第一向量的步骤之前，包括：
    对所述离散向量、交叉向量分别进行稠密处理。
PCT/CN2018/095504 2018-06-05 2018-07-12 预测投保风险概率的方法、装置、计算机设备和存储介质 WO2019232892A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810569999.XA CN109002900A (zh) 2018-06-05 2018-06-05 预测投保风险概率的方法、装置、计算机设备和存储介质
CN201810569999.X 2018-06-05

Publications (1)

Publication Number Publication Date
WO2019232892A1 true WO2019232892A1 (zh) 2019-12-12

Family

ID=64574314

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/095504 WO2019232892A1 (zh) 2018-06-05 2018-07-12 预测投保风险概率的方法、装置、计算机设备和存储介质

Country Status (2)

Country Link
CN (1) CN109002900A (zh)
WO (1) WO2019232892A1 (zh)

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109934389A (zh) * 2019-02-18 2019-06-25 平安科技（深圳）有限公司 基于预测模型的绩效预测方法、装置及存储介质
CN109784779B (zh) * 2019-03-04 2021-06-18 广州杰赛科技股份有限公司 财务风险预测方法、装置及存储介质
CN109978700A (zh) * 2019-03-29 2019-07-05 英大长安保险经纪有限公司 环境污染责任保险费率的调整方法及计算设备
CN110288488A (zh) * 2019-06-24 2019-09-27 泰康保险集团股份有限公司 医疗险欺诈预测方法、装置、设备和可读存储介质
CN110555749B (zh) * 2019-07-26 2021-10-29 创新先进技术有限公司 基于神经网络的信用行为预测方法以及装置
CN110706117A (zh) * 2019-08-22 2020-01-17 中国平安财产保险股份有限公司 业务处理方法、装置、计算机装置及存储介质
CN111553800B (zh) * 2020-04-30 2023-08-25 上海商汤智能科技有限公司 一种数据处理方法及装置、电子设备和存储介质
CN112559971A (zh) * 2021-02-25 2021-03-26 北京芯盾时代科技有限公司 一种概率预测方法、装置及计算机可读存储介质
CN113537560A (zh) * 2021-06-07 2021-10-22 同盾科技有限公司 用户投保意愿预测的方法、系统、电子装置和存储介质
CN113850686B (zh) * 2021-10-08 2023-11-28 同盾网络科技有限公司 投保概率确定方法、装置、存储介质及电子设备

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106127576A (zh) * 2016-07-01 2016-11-16 武汉泰迪智慧科技有限公司 一种基于用户行为的银行风险评估系统
CN106296195A (zh) * 2015-05-29 2017-01-04 阿里巴巴集团控股有限公司 一种风险识别方法及装置
CN107566358A (zh) * 2017-08-25 2018-01-09 腾讯科技（深圳）有限公司 一种风险预警提示方法、装置、介质及设备
CN107993140A (zh) * 2017-11-22 2018-05-04 深圳市耐飞科技有限公司 一种个人信贷风险评估方法及系统

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3852019A1 (en) * 2015-03-27 2021-07-21 Equifax, Inc. Optimizing neural networks for risk assessment
CN106530091A (zh) * 2015-09-15 2017-03-22 平安科技（深圳）有限公司 投保额度的计算方法及服务器
CN107798448A (zh) * 2016-12-15 2018-03-13 平安科技（深圳）有限公司 黑名单用户的判断方法及装置
CN107292528A (zh) * 2017-06-30 2017-10-24 阿里巴巴集团控股有限公司 车险风险预测方法、装置及服务器
CN107818513A (zh) * 2017-11-24 2018-03-20 泰康保险集团股份有限公司 风险评估方法及装置、存储介质、电子设备

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106296195A (zh) * 2015-05-29 2017-01-04 阿里巴巴集团控股有限公司 一种风险识别方法及装置
CN106127576A (zh) * 2016-07-01 2016-11-16 武汉泰迪智慧科技有限公司 一种基于用户行为的银行风险评估系统
CN107566358A (zh) * 2017-08-25 2018-01-09 腾讯科技（深圳）有限公司 一种风险预警提示方法、装置、介质及设备
CN107993140A (zh) * 2017-11-22 2018-05-04 深圳市耐飞科技有限公司 一种个人信贷风险评估方法及系统

Also Published As

Publication number Publication date
CN109002900A (zh) 2018-12-14

Similar Documents

Publication Publication Date Title
WO2019232892A1 (zh) 预测投保风险概率的方法、装置、计算机设备和存储介质
US20230275817A1 (en) Parallel computational framework and application server for determining path connectivity
WO2019196546A1 (zh) 确定业务请求事件的风险概率的方法及装置
JP6182279B2 (ja) データ分析システム、データ分析方法、データ分析プログラム、および、記録媒体
CN108133013B (zh) 信息处理方法、装置、计算机设备和存储介质
Dutta et al. Scenario analysis in the measurement of operational risk capital: a change of measure approach
US20150154520A1 (en) Automated Data Breach Notification
WO2020253357A1 (zh) 数据产品推荐方法、装置、计算机设备和存储介质
US9355370B2 (en) System and method for generating legal documents
Quijano Xacur et al. Generalised linear models for aggregate claims: to Tweedie or not?
JP6650502B2 (ja) 判定装置、判定方法、及び判定プログラム
WO2023000794A1 (zh) 保护数据隐私的业务预测模型训练的方法及装置
TW202004636A (zh) 保險服務優化方法、系統及電腦程式產品
US11494860B2 (en) Systems and methods for implementing search and recommendation tools for attorney selection
WO2020034801A1 (zh) 医疗特征筛选方法、装置、计算机设备和存储介质
JP6060298B1 (ja) 情報配信装置、情報配信方法および情報配信プログラム
WO2016084642A1 (ja) 与信審査用サーバと与信審査用システム及び与信審査用プログラム
WO2023086954A1 (en) Bayesian modeling for risk assessment based on integrating information from dynamic data sources
Kuang et al. Generalized log-normal chain-ladder
Youssef et al. Robust SURE estimates of profitability in the Egyptian insurance market
Yu et al. Asymptotic properties and information criteria for misspecified generalized linear mixed models
Jones et al. New radiocarbon calibration software
Zhao et al. Claim reserving for insurance contracts in line with the International Financial Reporting Standards 17: a new paid-incurred chain approach to risk adjustments
US20140324523A1 (en) Missing String Compensation In Capped Customer Linkage Model
CN115758271A (zh) 数据处理方法、装置、计算机设备和存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18921744

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18921744

Country of ref document: EP

Kind code of ref document: A1