Data processing system, method, electronic device and storage medium

Abstract

The present invention discloses a system, method, electronic device and storage medium for data processing, and relates to the field of computer technology. A specific implementation of the system includes: a receiving module for receiving the health data of the user sent by the collection terminal; a construction module for constructing a feature vector according to the health data received by the receiving module; a grade probability calculation module for calculating the underwriting results corresponding to each probability model respectively according to the feature vector through N probability models; an underwriting grade calculation module for calculating the user's target underwriting grade according to the underwriting results corresponding to each probability model; an output module for outputting the user's target underwriting grade to the underwriting terminal, so that the underwriting terminal underwrites the user according to the target underwriting grade. This implementation can solve the problem that the manual evaluation method not only consumes a lot of labor costs, but also takes a lot of time, reducing the efficiency of underwriting the policyholder.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a data processing system, method, electronic device and storage medium. Background Art

[0002] In the insurance field, users need to evaluate the physical condition of the insured when purchasing insurance, and then underwrite the insured's policy based on the evaluation results. The evaluation of the insured is usually: based on the insured's physical examination data, the physical examination report data is manually analyzed, and then the insured's underwriting level is evaluated.

[0003] In the process of implementing the present invention, the inventors found that there are at least the following problems in the prior art:

[0004] Physical examination data is usually large in volume and of complex and diverse types. Manual evaluation not only requires a lot of labor costs, but also takes a lot of time, reducing the efficiency of underwriting for policyholders. Summary of the invention

[0005] In view of this, the embodiments of the present invention provide a data processing system, method, system and storage medium, which can solve the problem that the manual evaluation method not only requires a large amount of labor costs, but also takes a lot of time, reducing the efficiency of underwriting for policyholders.

[0006] To achieve the above objective, according to one aspect of an embodiment of the present invention, a data processing system is provided.

[0007] A data processing system according to an embodiment of the present invention includes: a receiving module for receiving health data of a user collected by a collection device; a building module for building a feature vector according to the health data received by the receiving module; a level probability calculation module for calculating, according to the feature vector and through N probability models, the underwriting results corresponding to each of the probability models, the underwriting results including the probability that the user belongs to each underwriting level, wherein N is an integer greater than 0; an underwriting level calculation module for calculating the underwriting level of the user according to the underwriting results corresponding to each of the probability models; and an output module for outputting the underwriting level of the user to an underwriting terminal, so that the underwriting terminal underwrites the user according to the underwriting level.

[0008] In one embodiment, the level probability calculation module is specifically used to:

[0009] Through each of the probability models, according to the first fully connected mapping function, the first transformation vector corresponding to the feature vector is calculated; according to the nonlinear transformation function, the first transformation vector is transformed into a transformation vector; according to the second fully connected mapping function, the second transformation vector corresponding to the transformation vector is calculated; and the second transformation vector is determined as the underwriting result corresponding to the probability model.

[0010] In yet another embodiment, the level probability calculation module is specifically used to:

[0011] Through each of the probability models, according to the connection matrix, features in the transformation vector are eliminated to obtain a connection vector corresponding to the transformation vector; and according to the second fully connected mapping, a second transformation vector corresponding to the connection vector is calculated.

[0012] In yet another embodiment, the level probability calculation module is specifically used to:

[0013] By means of each of the probability models, a first product of the feature vector and a first feature transformation matrix is ​​calculated, and a sum of the first product and a first offset is determined as the first conversion vector;

[0014] And by using each of the probability models, a second product of the transformation vector and a second characteristic transformation matrix is ​​calculated, and a sum of the second product and a second offset is determined as the second conversion vector.

[0015] In yet another embodiment, the level probability calculation module is specifically used to:

[0016] Through each of the probability models, the elements in the feature vector are normalized; the normalized feature vector is expanded and the expanded feature vector is concatenated into a column vector; and according to the first fully connected mapping function and the column vector, a first conversion vector corresponding to the feature vector is calculated.

[0017] In yet another embodiment, the underwriting level calculation module is used to:

[0018] For each of the underwriting levels, the probability that the user belongs to the underwriting level is screened from the underwriting results corresponding to each of the probability models, and a probability set that the user belongs to the underwriting level is obtained;

[0019] A target underwriting level for the user is determined according to a probability set that the user belongs to each of the underwriting levels.

[0020] In yet another embodiment, the data processing system further comprises a training module;

[0021] The training module is used to:

[0022] The N probability models are trained according to the residuals corresponding to the N probability models respectively, wherein the residual corresponding to the Mth probability model among the N probability models is the difference between the expected output of the probability model and the sum of the outputs of the first to Mth probability models, and M is an integer greater than 0 and less than or equal to N.

[0023] To achieve the above objective, according to one aspect of an embodiment of the present invention, a data processing method is provided.

[0024] A data processing method of an embodiment of the present invention includes: receiving health data of a user collected by a collection device; constructing a feature vector based on the health data; based on the feature vector, using N probability models, respectively calculating the underwriting results corresponding to each of the probability models, the underwriting results including the probability that the user belongs to each underwriting level, wherein N is an integer greater than 0; calculating the underwriting level of the user based on the underwriting results corresponding to each of the probability models; outputting the underwriting level of the user to an underwriting terminal, so that the underwriting terminal underwrites the user based on the underwriting level.

[0025] In one embodiment, the calculating, based on the feature vector and using N probability models, respectively the underwriting result corresponding to each probability model includes:

[0026] The following process is performed for each of the N probability models:

[0027] Calculating a first conversion vector corresponding to the feature vector according to a first fully connected mapping function;

[0028] transforming the first conversion vector into a transformation vector according to a nonlinear transformation function;

[0029] Calculating a second conversion vector corresponding to the transformation vector according to a second fully connected mapping function;

[0030] The second conversion vector is determined as the underwriting result corresponding to the probability model.

[0031] In yet another embodiment, before calculating the second conversion vector corresponding to the transformation vector according to the second fully connected mapping, the method includes:

[0032] According to the connection matrix, the features in the transformation vector are eliminated to obtain the connection vector corresponding to the transformation vector;

[0033] According to the second fully connected mapping, a second conversion vector corresponding to the connection vector is calculated.

[0034] In yet another embodiment, calculating the first conversion vector corresponding to the feature vector according to the first fully connected mapping function includes:

[0035] Calculating a first product of the eigenvector and a first eigentransformation matrix, and determining a sum of the first product and a first offset as the first conversion vector;

[0036] The calculating, according to the second fully connected mapping function, a second conversion vector corresponding to the transformation vector comprises:

[0037] A second product of the transformation vector and a second characteristic transformation matrix is ​​calculated, and a sum of the second product and a second offset is determined as the second conversion vector.

[0038] In yet another embodiment, calculating the first conversion vector corresponding to the feature vector according to the first fully connected mapping function includes:

[0039] Normalizing the elements in the feature vector;

[0040] Expand the normalized feature vector and concatenate the expanded feature vector into a column vector;

[0041] A first conversion vector corresponding to the feature vector is calculated according to the first fully connected mapping function and the column vector.

[0042] In yet another embodiment, calculating the user's insurance level according to the insurance results corresponding to each of the probability models includes:

[0043] For each of the underwriting levels, the probability that the user belongs to the underwriting level is screened from the underwriting results corresponding to each of the probability models, and a probability set that the user belongs to the underwriting level is obtained;

[0044] The assurance level of the user is determined according to a probability set that the user belongs to each of the assurance levels.

[0045] In yet another embodiment, before respectively calculating the core guarantee knot corresponding to each probability model through N probability models according to the feature vector, the method further includes:

[0046] The N probability models are trained according to the residuals corresponding to the N probability models respectively, wherein the residual corresponding to the Mth probability model among the N probability models is the difference between the expected output of the probability model and the sum of the outputs of the first to Mth probability models, and M is an integer greater than 0 and less than or equal to N.

[0047] To achieve the above objective, according to another aspect of an embodiment of the present invention, an electronic device is provided.

[0048] An electronic device according to an embodiment of the present invention includes: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the data processing method provided by the embodiment of the present invention.

[0049] To achieve the above objective, according to another aspect of an embodiment of the present invention, a computer readable medium is provided.

[0050] A computer-readable medium according to an embodiment of the present invention stores a computer program, and when the program is executed by a processor, the data processing method provided by the embodiment of the present invention is implemented.

[0051] An embodiment of the above invention has the following advantages or beneficial effects: In the embodiment of the present invention, a feature vector is constructed for the user's health data, and then the underwriting results corresponding to each probability model can be calculated through N probability models, and the user's target underwriting level can be determined based on the underwriting results. In this way, in the embodiment of the present invention, the underwriting results can be calculated directly based on the feature vector of the health data based on N probability models, and then the user's target underwriting level can be determined, without the need for manual analysis of the health data, reducing the cost and time consumed, and improving the efficiency of underwriting for the insured.

[0052] The further effects of the above-mentioned non-conventional optional manner will be described below in conjunction with the specific implementation manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The accompanying drawings are used to better understand the present invention and do not constitute an improper limitation of the present invention.

[0054] Figure 1 is a schematic diagram of a network architecture of a data processing system according to an embodiment of the present invention;

[0055] Figure 2 is a schematic diagram of a main flow of a data processing method according to an embodiment of the present invention;

[0056] Figure 3 is a schematic diagram of another main flow of a data processing method according to an embodiment of the present invention;

[0057] Figure 4 is a schematic diagram of a structure of probability model training according to an embodiment of the present invention;

[0058] Figure 5 is a schematic diagram of a network structure of a probability model according to an embodiment of the present invention;

[0059] Figure 6 is a schematic diagram of an underwriting result according to an embodiment of the present invention;

[0060] Figure 7 is a schematic diagram of main modules of a data processing system according to an embodiment of the present invention;

[0061] Figure 8 is another exemplary system architecture diagram to which the embodiments of the present invention can be applied;

[0062] Fig. 9 It is a schematic diagram of the structure of a computer system suitable for implementing the embodiment of the present invention. DETAILED DESCRIPTION

[0063] The following is a description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0064] It should be pointed out that the embodiments of the present invention and the features in the embodiments may be combined with each other if there is no conflict.

[0065] An embodiment of the present invention provides a data processing system that can be used in scenarios where insurance underwriting is performed on insured users. Figure 1 FIG. 1 is a system architecture diagram of a data processing system in an embodiment of the present invention. Figure 1 As shown, the data processing system can be communicatively connected with the collection terminal and the underwriting terminal. The collection terminal can collect the user's health data and transmit the collected health data to the data processing system. After the data processing system derives the user's target underwriting level, it outputs it to the underwriting terminal. The underwriting terminal can underwrite the user according to the target underwriting level.

[0066] At the same time, the data processing system can also store the target underwriting level obtained for each user. When the underwriting terminal needs to query the target underwriting level of different users, it can send a query request to the data processing system. The data processing system sends the content of the underwriting terminal's query to the underwriting terminal to facilitate the underwriting terminal to perform underwriting.

[0067] In the embodiment of the present invention, the data processing system may further include a display module for displaying the obtained target underwriting level for the convenience of business personnel to view.

[0068] The embodiment of the present invention provides a data processing method, which can be executed by a data processing system, such as Figure 2 As shown, the method includes the following steps.

[0069] S201: Receive the user's health data sent by the collection terminal.

[0070] In this step, the user's health data can be received from the collection terminal.

[0071] S202: Construct a feature vector according to the user's health data.

[0072] Among them, the probability model processes vector data, so in this step, the user's health data is constructed into a feature vector, and then operations are performed based on the feature vector.

[0073] It should be noted that, in the embodiment of the present invention, the health data may specifically be physical examination data. The way to construct the feature vector for the physical examination data may be to use the values ​​of each item of the physical examination data as corresponding feature vector elements to construct the feature vector.

[0074] S203: Based on the feature vector, N probability models are used to calculate the underwriting results corresponding to each probability model.

[0075] The underwriting result includes the probability that the user belongs to each underwriting level, where N is an integer greater than 0.

[0076] The N probability models can have the same probability model structure and are used to classify feature vectors after training. The output of the probability model is the probability that the input data belongs to each underwriting level. For example, if the underwriting level is divided into four levels, the output of the probability model is a vector including four elements, and each of the four elements in the vector represents the probability that the user belongs to the corresponding underwriting level.

[0077] Specifically, the processing process of the input feature vector by each probability model in the N probability models can be executed as follows: according to the first fully connected mapping function, the first transformation vector corresponding to the feature vector is calculated; according to the nonlinear transformation function, the first transformation vector is transformed into a transformation vector; according to the second fully connected mapping function, the second transformation vector corresponding to the transformation vector is calculated; and the second transformation vector is determined as the underwriting result corresponding to the probability model.

[0078] After the feature vector is input into the probability model, it is first mapped by the first fully connected layer, specifically, the first transformation vector is calculated according to the first fully connected mapping function and the feature vector. In the embodiment of the present invention, the first fully connected mapping function can use a linear function. For example, the calculation process can be: calculating the first product of the feature vector and the first feature transformation matrix, and then determining the sum of the first product and the first offset as the first transformation vector. The first feature transformation matrix and the first offset are both parameters determined during the probability model training.

[0079] After the feature vector is mapped by the first fully connected mapping function, a first conversion vector is obtained, and then the first conversion vector can be transformed according to the nonlinear transformation function to obtain a transformation vector. By transforming through the nonlinear transformation function, the feature vector can be filtered. In the embodiment of the present invention, the transformation method in this step can adopt a ReLU function, that is, the nonlinear transformation function is implemented by the ReLU function to obtain a transformed transformation vector.

[0080] After obtaining the transformation vector, the transformation vector can be mapped by the second fully connected layer to obtain the underwriting result. The specific method is to calculate the second transformation vector based on the second fully connected mapping function and the transformation vector. In an embodiment of the present invention, the second fully connected mapping function can also use a linear function. For example, the calculation process can be: calculate the second product of the transformation vector and the second feature transformation matrix, and then determine the sum of the second product and the second offset as the second transformation vector. The second feature transformation matrix and the second offset are both parameters determined during the training of the probability model. The second transformation vector is the output result of the probability model, which can be output after the probability model calculates the second transformation vector, that is, the second transformation vector is determined as the underwriting result corresponding to the probability model.

[0081] In order to improve the robustness of the probability model and reduce the amount of calculation, in an embodiment of the present invention, the method for calculating the second conversion vector corresponding to the transformation vector according to the second fully connected mapping function can be specifically as follows: according to the connection matrix, the features in the transformation vector are eliminated to obtain the connection vector corresponding to the transformation vector; according to the second fully connected mapping, the second conversion vector corresponding to the connection vector is calculated.

[0082] After the transformation vector is obtained in the probability model, the features can be removed first, and then the second fully connected layer conversion can be performed. In an embodiment of the present invention, the transformation vector is feature-removed by a connection matrix, and the connection matrix is ​​a parameter determined during the probability model training. After the transformation vector is feature-removed, a connection vector is obtained, and the connection vector can be calculated according to the second fully connected mapping to obtain a second transformation vector. The specific calculation method can be: calculate the second product of the connection vector and the second feature transformation matrix, and then determine the sum of the second product and the second offset as the second transformation vector.

[0083] In the embodiment of the present invention, after the feature vector constructed in step S202 is input into the probability model, it can also be preprocessed before the first fully connected layer conversion. The specific method is: normalize the elements in the feature vector; expand the normalized feature vector, and splice the expanded feature vector into a column vector; calculate the first conversion vector corresponding to the feature vector according to the first fully connected mapping function and the column vector.

[0084] After normalizing, expanding and concatenating the eigenvector, a column vector of a column is obtained, and then the first fully connected layer conversion is performed. The normalization of the eigenvector is to convert the value of the element in the eigenvector into a floating point number between 0 and 1. Then, the normalized eigenvector is expanded, and the expansion method can be implemented by the Embedding method. After the eigenvector is expanded, it can be concatenated again, and the concatenation method can be to concatenate the elements from top to bottom and from left to right to obtain a column vector of a column. Then, the column vector of a column is mapped by the first fully connected layer, and the specific method can be: calculate the first product of the column vector and the first eigentransformation matrix, and determine the sum of the first product and the first offset as the first conversion vector.

[0085] In an embodiment of the present invention, the dimension of the underwriting results output by each probability model can be determined according to the number of underwriting levels.

[0086] S204: Determine the user's target underwriting level based on the underwriting results corresponding to each probability model.

[0087] Among them, after calculating the underwriting results of each probability model in step S203, the user's target underwriting level can be determined according to the underwriting results of each probability model.

[0088] In an embodiment of the present invention, the underwriting result may be the probability that the user belongs to each underwriting level. The method for determining the user's target underwriting level in this step may be: for each underwriting level, from the underwriting results corresponding to each probability model, filter the probability that the user belongs to the underwriting level, and obtain a probability set that the user belongs to the underwriting level; determine the user's target underwriting level based on the probability set that the user belongs to each underwriting level.

[0089] The results output by each probability model include the probability that the user belongs to each underwriting level. For each underwriting level, a corresponding probability can be determined from the results output by the probability model. Therefore, from the underwriting results corresponding to each probability model, N probabilities that the user belongs to the underwriting level can be screened out, and then the probability set of the user belonging to each underwriting level can be obtained. Then, based on the probability set of the user belonging to each underwriting level, the user's target underwriting level can be determined.

[0090] In an embodiment of the present invention, the target underwriting level of the user can be determined based on the probability set that the user belongs to each underwriting level, which can be performed as follows: calculating the sum of the probabilities in the probability set of each underwriting level, and determining the underwriting level with the highest sum of probabilities as the target underwriting level.

[0091] Assume that in the embodiment of the present invention, there are four underwriting levels, namely, rejection, extension, additional inspection, and underwriting. The probability that the user belongs to the above four levels is obtained in N probability models, and then the sum of the probabilities that the user belongs to each underwriting level in each underwriting result is calculated, and the underwriting level with the highest sum of probabilities is determined as the user's target underwriting level.

[0092] After determining the user's target underwriting level, it can be fed back to the underwriting terminal, which can review it. The insurance policies of users with a target underwriting level of underwriting will be automatically approved, and the insurance policies of users with the other three target underwriting levels will be executed: rejection, extension, and additional inspection. The final conclusion can only be given after manual review.

[0093] It should be noted that, in the embodiment of the present invention, before executing step S203, it is necessary to train N probability models, which are used to calculate the underwriting results in the embodiment of the present invention after the training is completed.

[0094] Since the underwriting results obtained by the N probability models are combined to determine the target probability model, the N probability models in the embodiment of the present invention can be combined with each other for model training. The specific method of training can be: according to the residuals corresponding to the N probability models, the N probability models are trained, wherein the residual corresponding to the Mth probability model among the N probability models is the difference between the expected output of the probability model and the sum of the outputs of the 1st to Mth probability models, and M is an integer greater than 0 and less than or equal to N.

[0095] In an embodiment of the present invention, the training phase includes training of each probability model and training of the combination of each model. For the training of each probability model, the residual corresponding to the probability model can be directly used as the loss function of the probability model training. For the training of the combination of each probability model, the total loss function can be calculated based on the residuals of each probability model. During training, the training of the probability model needs to be completed after the loss function of each probability model training and the total loss function meet the preset conditions or the number of training times are reached.

[0096] S205: Output the user's target underwriting level to the underwriting terminal.

[0097] After the target underwriting level is obtained, it can be output to the underwriting terminal.

[0098] In the embodiment of the present invention, a feature vector is constructed for the user's health data, and then the underwriting results corresponding to each probability model can be calculated through N probability models, and then the user's target underwriting level can be determined based on the underwriting results. In this way, in the embodiment of the present invention, the underwriting results can be calculated directly based on the feature vector of the health data based on N probability models, and then the user's target underwriting level can be determined, without the need for manual analysis of health data, reducing the cost and time consumed, and improving the efficiency of underwriting for policyholders.

[0099] The N probability models in the embodiment of the present invention can be used as a general model, which can solve the problems of large amount of model training data, complex network models, and difficulty in parallel calculation using methods such as deep learning. They can be used in many classification problems such as underwriting prediction and health risk prediction. The N probability models used in the embodiment of the present invention do not require too much training data and can meet the accuracy requirements, which can solve the problem of insufficient accuracy caused by insufficient training data during model training. At the same time, in the embodiment of the present invention, the N probability models are trained by residual cascading, and the training of the next probability model can be dependent on the residual of the previous probability model, so that the next probability model can be used to make up for the lack of accuracy of the previous probability model, thereby achieving the purpose of improving the accuracy of underwriting level determination.

[0100] It should be noted that, in the embodiment of the present invention, different values ​​of N have different precisions in determining the underwriting level, and the specific value of N can be determined based on experience or by comparing the results of different values.

[0101] Combine the following Figure 1 In the embodiment shown, the health data is taken as physical examination data, and the data processing process of the kth probability model in the N probability models in the embodiment of the present invention during model training is specifically described, where k is an integer greater than 0 and less than or equal to N. Figure 3 As shown, the method includes the following steps.

[0102] S301: constructing a feature vector according to the physical examination data of multiple samples.

[0103] Among them, the data types of the physical examination data are usually continuous variable data and categorical variable data. In an embodiment of the present invention, m physical examination data for underwriting can be screened out from all physical examination data of a sample based on an expert knowledge base (finally given by underwriter experience, feature selection methods and experiments), and the specific physical examination data can be as shown in Table 1, where m is an integer greater than 1. For example, the physical examination data may include physical examination data of continuous value types such as height, age, platelet count, urine pH, γ-glutamyl transferase, alanine aminotransferase, smoking index, weight, creatinine, and hemoglobin, hematocrit, BMI, aspartate aminotransferase, systolic blood pressure, diastolic blood pressure, fasting blood sugar, triglycerides, total cholesterol, mean corpuscular volume, mean corpuscular hemoglobin content, mean corpuscular hemoglobin concentration, neutrophil absolute value, neutrophil percentage, lymphocyte absolute value, lymphocyte percentage, etc., which are categorical physical examination data.

[0104] For the above parameters, the physical examination data includes parameter values ​​corresponding to each parameter. In the embodiment of the present invention, each parameter value can be directly spliced ​​into vector data of m rows and 1 column, that is, a feature vector of a sample.

[0105] A large number of samples are usually selected during the training phase. In the embodiment of the present invention, t samples are selected for training as an example. The constructed feature vectors of t samples can be expressed as X(F,Y), where F represents the feature vector of each sample (composed of the values ​​of the physical examination data of t samples), and Y represents the desired insurance rating label vector, whose value represents the label of the feature vector of the corresponding sample. Wherein, t is an integer greater than 1.

[0106] S302: Normalize the feature vector.

[0107] Among them, since the physical examination data is directly used as the element of the feature vector in step S301, it can be found that the value range of each data in the feature vector is different, and there are discrete category values ​​and continuous floating point values, which need to be normalized.

[0108] In the embodiment of the present invention, the normalization process of the feature vector of the i-th sample in X(F,Y) is taken as an example for description.

[0109] The feature vector of the i-th sample in X(F,Y) can be expressed as F i ,i=1,2,...,t. Each F i Contains m features, i.e., the values ​​corresponding to m physical examination data, F i The jth feature can be expressed as f i,j , where j = 1, 2, ..., m, then the feature vector of the i-th sample can be expressed as: F i ={fi,1 ,f i,2 ,...,f i,j ,...,f i,m}, where F i is a column vector.

[0110] F i The jth feature f i,j Normalization can be calculated by formula 1, F i Normalized feature vector It can be expressed as: in is a column vector.

[0111]

[0112] in, represents f i,j The normalized value, μ j represents the average value, which is calculated as σ j represents the variance, which is calculated as

[0113] For each element in the eigenvector F, i,j The normalized feature vector can be obtained by normalizing the normalized process.

[0114] S303: Expand the normalized feature vector.

[0115] Among them, the Embedding method can be used to perform expansion operations on the normalized feature vector.

[0116] For example, the eigenvector of the i-th sample in X(F,Y) is normalized to obtain the eigenvector Then use the Embedding method to perform the expansion operation. For the feature vector Each 1 row * 1 column element can be expanded to a 1 row * 7 column element vector. Assume that the normalized eigenvector is m rows and 1 column, then the expanded eigenvector is expressed as is a vector of m rows and 7 columns, which can be specifically shown in Formula 2.

[0117]

[0118] S304: concatenate the expanded feature vectors to obtain a column vector.

[0119] In this step, the elements of the expanded feature vector can be concatenated in order from top to bottom and from left to right to obtain a column vector.

[0120] For example, in step S303, the feature vector of the i-th sample is expanded to obtain a two-dimensional vector of m rows and 7 columns. In this step, the two-dimensional feature vectors of m rows and 7 columns can be sequentially spliced ​​to obtain a vector X1 of 7×m rows and 1 column. The specific sequential splicing method can be: The features of are concatenated in order from top to bottom and from left to right to obtain X1, which can be specifically shown in Formula 3.

[0121]

[0122] S305: Perform a first fully connected feature mapping on the column vector to obtain a first conversion vector.

[0123] In this step, the first fully connected feature mapping can be implemented according to the first fully connected mapping function. The specific first fully connected mapping function can be shown in Formula 4.

[0124] X2=W1*X1+b1 (4)

[0125] In Formula 4, W1 represents the first feature transformation matrix, b1 represents the first offset, X1 represents the column vector obtained in step S304, and X2 represents the first conversion vector. The initial values ​​of W1 and b1 in training are random values, and specific values ​​are determined after the probability model training.

[0126] S306: Normalize the first conversion vector.

[0127] In this step, X2 is normalized to obtain a normalized first conversion vector, which is represented as X3, and all element values ​​in X3 are normalized to decimals between 0 and 1. The normalization principle of this step is the same as that of step S202.

[0128] S307: Perform nonlinear transformation on the normalized first transformation vector to obtain a transformation vector.

[0129] In this step, the ReLU function can be used to perform nonlinear transformation on each element in X3 obtained in step S306 to obtain a transformation vector X4.

[0130] S308: Perform feature elimination on the transformation vector through the connection matrix to obtain a connection vector.

[0131] In this step, the connection matrix H can be set during training. The initial value of H can use the Dropout function to randomly set the values ​​in the matrix H to 0 to obtain a random sparse matrix H, and then determine the value of H through probability model training.

[0132] The size of the connection matrix H is the same as that of X4, which can be expressed as Formula 5. H is a binary matrix. The values ​​of the elements in the connection matrix can be 0 or 1 according to a certain probability. The connection matrix H is multiplied by the corresponding elements of X4. After the multiplication, X4 is a feature vector with some values ​​of 0 and some values ​​of the original value. Specifically, the connection matrix H can be shown as Formula 5, and X4 can be shown as Formula 6. Then, the transformation vector is feature-removed through the connection matrix to obtain the connection vector It can be shown as formula 7.

[0133]

[0134]

[0135]

[0136] Among them, the values ​​in the connection matrix H can be randomly set to 0 or 1 in each training process.

[0137] S309: Perform a second fully connected feature mapping on the connection vector to obtain a second conversion vector.

[0138] In this step, the second fully connected feature mapping can be implemented according to the second fully connected mapping function. The specific second fully connected mapping function can be shown in Formula 8.

[0139]

[0140] In Formula 8, W2 represents the second feature transformation matrix, b2 represents the second offset, represents the connection vector obtained in step S308, and X5 represents the second conversion vector. The initial values ​​of W2 and b2 in training are random values, and the specific values ​​are determined after the probability model training.

[0141] In the embodiment of the present invention, the second conversion vector obtained by the k-th probability model is the output of the k-th probability model, which can be expressed as y k .

[0142] S310: Calculate the loss function.

[0143] It should be noted that, in the implementation of the present invention, N probability models can be connected for training by residual cascade. The connection method can be as follows: Figure 4 shown. Figure 4 The intermediate agent is the probability model in the embodiment of the present invention. Multiple probability models are cascaded together to form a training model of the entire multi-probability model. The residual of each probability model is passed to the next probability model, thereby affecting the loss function of each probability model.

[0144] The actual output of the kth probability model is represented as y k , the expected output of the probability model is expressed as The residual is denoted as e k In the embodiment of the present invention, the square of the residual of each probability model is used as the loss function of the probability model.

[0145] The residual e1 of the first probability model is expressed as:

[0146] The residual e2 of the second probability model is expressed as:

[0147] The residual e of the kth probability model k It is expressed as: Where r is an integer.

[0148] The residual e of the nth probability model n It is expressed as:

[0149] Based on the above residual representation, the loss function l of the nth probability model is n Expressed as

[0150] In this step, the loss function is calculated based on the residual, and the loss function is calculated after each training until the loss function reaches the preset condition or the number of training times is reached, thus realizing the training of a single model.

[0151] It should be noted that, combined with Figure 4 As shown in the structure diagram, while training each probability model, N probability models also need to be trained together, so Figure 4 The coordinator is used to calculate the total loss function Obj of N probability models. The total loss function Obj can be calculated by formula 9.

[0152]

[0153] In Formula 9, ω r It is the weight matrix of the rth probability model, which is formed by concatenating the first feature transformation matrix W1 and the second feature transformation matrix W2 corresponding to the probability model in series. τ is the regularization term constant.

[0154] Therefore, the coordinator is the regulator of the overall probability model training. The ultimate goal of the overall probability model training is to minimize the coordinator's objective function Obj. When the value of Obj is less than a certain threshold or the number of training times reaches the threshold, the overall probability model stops training. While the coordinator's objective function Obj is optimized, each probability model is iteratively trained in parallel, and each probability model aims to minimize its own loss function l.

[0155] It should be noted that the network structure of each probability model can be as follows Figure 5 As shown in the figure, the input feature vector is expanded (Embedding), concatenated (Flatting), the first fully connected layer (FC, full connection), normalized (Normalization), ReLU function, Dropout function, and the second fully connected layer before outputting the kernel guarantee result.

[0156] The dimension of the vector used for feature vector processing at each stage can be adjusted according to the requirements. For example, assuming that the input feature vector is 24-dimensional, the Embedding expansion dimension can be 7, the dimension of the first fully connected layer can be 16-dimensional, and the dimension of the second fully connected layer can be 4-dimensional.

[0157] Through the process of the above-mentioned embodiment, after training to obtain a stable cascade probability model, it can be used to evaluate the underwriting level. For a given sample, such as the user's health data in steps S101S201, assuming that there are q underwriting levels, then after evaluation by different probability models, the underwriting results corresponding to each probability model are obtained, that is, the output X5, and then through step S204, the target underwriting level of the given sample is determined.

[0158] For example, if there are 4 underwriting levels, then the dimension of X5 output by each probability model includes 4 elements, which can be expressed as follows: Figure 6 As shown, the output of probability model 1 is (Z 11 , Z 12 , Z 13 , Z 14 ), the output of probability model 2 is (Z 21 , Z 22 , Z 23 , Z 24 ), and so on, the output of probability model n is (Z n1 , Z n2 , Z n3 , Z n4 ). By accumulating the corresponding values ​​of X5 output by different probability models, we can get the 4-dimensional output vector c={c1,c2,c3,c4} of the overall cascade network model, where c1=Z 11 +Z 21 +…+Z n1 , c2=Z 12 +Z 22 +…+Z n2 , c3=Z 13 +Z 23 +…+Z n3 , c2=Z14 +Z 24 +…+Z n4 Then determine the element with the largest value in c, and the underwriting level corresponding to this element is the target underwriting level of the given sample.

[0159] The N probability models used in the embodiment of the present invention do not require too much training data and can meet the accuracy requirements, which can solve the problem of insufficient accuracy caused by insufficient training data during model training. At the same time, in the embodiment of the present invention, the N probability models are trained through residual cascade, and the training of the next probability model can be dependent on the residual of the previous probability model, so that the next probability model can be used to make up for the lack of accuracy of the previous probability model, thereby achieving the purpose of improving the accuracy of determining the underwriting result.

[0160] In order to solve the problems existing in the prior art, an embodiment of the present invention provides a data processing system 700, such as Figure 7 As shown, the system 700 includes:

[0161] The receiving module 701 is used to receive the user's health data sent by the collection terminal;

[0162] A construction module 702, configured to construct a feature vector according to the health data received by the receiving module;

[0163] The level probability calculation module 703 is used to calculate the underwriting results corresponding to each of the probability models according to the feature vector through N probability models, wherein the underwriting results include the probability that the user belongs to each underwriting level, wherein N is an integer greater than 0;

[0164] An underwriting level calculation module 704, configured to calculate a target underwriting level of the user according to the underwriting results corresponding to each of the probability models;

[0165] The output module 705 is used to output the target underwriting level of the user to the underwriting terminal, so that the underwriting terminal underwrites the user according to the target underwriting level.

[0166] It should be understood that the manner of implementing the embodiments of the present invention is different from the Figure 2 The method of the illustrated embodiment is the same and will not be repeated here.

[0167] In an implementation of the embodiment of the present invention, the level probability calculation module 703 is specifically used for:

[0168] Through each of the probability models, according to the first fully connected mapping function, the first transformation vector corresponding to the feature vector is calculated; according to the nonlinear transformation function, the first transformation vector is transformed into a transformation vector; according to the second fully connected mapping function, the second transformation vector corresponding to the transformation vector is calculated; and the second transformation vector is determined as the underwriting result corresponding to the probability model.

[0169] In another implementation of the embodiment of the present invention, the level probability calculation module 703 is specifically used to:

[0170] Through each of the probability models, according to the connection matrix, features in the transformation vector are eliminated to obtain a connection vector corresponding to the transformation vector; and according to the second fully connected mapping, a second transformation vector corresponding to the connection vector is calculated.

[0171] In another implementation of the embodiment of the present invention, the level probability calculation module 703 is specifically used to:

[0172] By means of each of the probability models, a first product of the feature vector and a first feature transformation matrix is ​​calculated, and a sum of the first product and a first offset is determined as the first conversion vector;

[0173] And by using each of the probability models, a second product of the transformation vector and a second characteristic transformation matrix is ​​calculated, and a sum of the second product and a second offset is determined as the second transformation direction.

[0174] In another implementation of the embodiment of the present invention, the level probability calculation module 703 is specifically used to:

[0175] Through each of the probability models, the elements in the feature vector are normalized; the normalized feature vector is expanded and the expanded feature vector is concatenated into a column vector; and according to the first fully connected mapping function and the column vector, a first conversion vector corresponding to the feature vector is calculated.

[0176] In another implementation of the embodiment of the present invention, the assurance level calculation module 704 is used to:

[0177] For each of the underwriting levels, the probability that the user belongs to the underwriting level is screened from the underwriting results corresponding to each of the probability models, and a probability set that the user belongs to the underwriting level is obtained;

[0178] A target underwriting level for the user is determined based on a probability set that the user belongs to each of the underwriting levels.

[0179] In another implementation of the embodiment of the present invention, the data processing system 700 further includes a training module;

[0180] The training module is used to:

[0181] The N probability models are trained according to the residuals corresponding to the N probability models respectively, wherein the residual corresponding to the Mth probability model among the N probability models is the difference between the expected output of the probability model and the sum of the outputs of the first to Mth probability models, and M is an integer greater than 0 and less than or equal to N.

[0182] It should be understood that the manner of implementing the embodiments of the present invention is different from the Figure 2 or Figure 3 The method of the illustrated embodiment is the same and will not be described again here.

[0183] In the embodiment of the present invention, a feature vector is constructed for the user's health data, and then the underwriting results corresponding to each probability model can be calculated through N probability models, and the user's target underwriting level can be determined based on the underwriting results. In this way, in the embodiment of the present invention, the underwriting results can be calculated directly based on the feature vector of the health data based on N probability models, and then the user's target underwriting level can be determined, without the need for manual analysis of health data, reducing the cost and time consumed, and improving the efficiency of underwriting for policyholders.

[0184] According to an embodiment of the present invention, the present invention also provides an electronic device and a readable storage medium.

[0185] The electronic device of the present invention comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the data processing method provided by an embodiment of the present invention.

[0186] Figure 8 An exemplary system architecture 800 is shown to which the data processing method or data processing apparatus according to the embodiment of the present invention can be applied.

[0187] like Figure 8 As shown, system architecture 800 may include terminal devices 801, 802, 803, a network 804, and a server 805. Network 804 is used to provide a medium for communication links between terminal devices 801, 802, 803 and server 805. Network 804 may include various connection types, such as wired, wireless communication links, or optical fiber cables, etc.

[0188] Users can use terminal devices 801, 802, and 803 to interact with server 805 through network 804 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 801, 802, and 803, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only examples).

[0189] The terminal devices 801 , 802 , and 803 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0190] The server 805 may be a server that provides various services, such as a backend management server (only an example) that provides support for shopping websites browsed by users using the terminal devices 801, 802, and 803. The backend management server may analyze and process the received data such as product information query requests, and feed back the processing results (such as product information - only an example) to the terminal device.

[0191] It should be noted that the data processing method provided in the embodiment of the present invention is generally executed by the server 805 , and accordingly, the data processing system is generally set in the server 805 .

[0192] It should be understood that Figure 8 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0193] Reference below Fig. 9 , which shows a schematic diagram of the structure of a computer system 900 suitable for implementing an embodiment of the present invention. Fig. 9 The computer system shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0194] like Fig. 9 As shown, the computer system 900 includes a central processing unit (CPU) 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage part 908 into a random access memory (RAM) 903. In the RAM 903, various programs and data required for the operation of the system 900 are also stored. The CPU 901, the ROM 902, and the RAM 903 are connected to each other via a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0195] The following components are connected to the I / O interface 905: an input section 906 including a keyboard, a mouse, etc.; an output section 907 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN card, a modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the I / O interface 905 as needed. A removable medium 911, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 910 as needed, so that a computer program read therefrom is installed into the storage section 908 as needed.

[0196] In particular, according to the embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 909, and / or installed from the removable medium 911. When the computer program is executed by the central processing unit (CPU) 901, the above-mentioned functions defined in the system of the present invention are executed.

[0197] It should be noted that the computer-readable medium shown in the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0198] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a unit, a program segment, or a part of a code, and the above-mentioned unit, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0199] The modules involved in the embodiments of the present invention may be implemented by software or hardware. The modules described may also be set in a processor, for example, it may be described as: a processor includes a receiving module, a computing unit, and a determining unit. The names of these units do not constitute a limitation on the module itself in some cases, for example, the receiving module may also be described as a "module of the function of the receiving module".

[0200] As another aspect, the present invention further provides a computer-readable medium, which may be included in the device described in the above embodiment; or may exist independently without being assembled into the device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by a device, the device executes the data processing method provided by the present invention.

[0201] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A data processing system, characterized in that: include: A receiving module, used for receiving the user's health data sent by the collection terminal; A construction module, used to construct a feature vector according to the health data received by the receiving module; A level probability calculation module, configured to calculate, according to the feature vector, through N probability models, respectively an underwriting result corresponding to each of the probability models, wherein the underwriting result includes a probability that the user belongs to each underwriting level, wherein N is an integer greater than 0; An underwriting level calculation module, used to calculate the target underwriting level of the user according to the underwriting results corresponding to each of the probability models; an output module, configured to output the target underwriting level of the user to an underwriting terminal, so that the underwriting terminal underwrites the user according to the target underwriting level; The data processing system also includes a training module; The training module is used to: The N probability models are trained according to the residuals corresponding to the N identical probability models, wherein the residual corresponding to the Mth probability model among the N probability models is the difference between the expected output of the probability model and the sum of the outputs of the first to Mth probability models, M is an integer greater than 0 and less than or equal to N, and the N probability models are trained by residual cascading.

2. The system according to claim 1, characterized in that The level probability calculation module is specifically used for: Through each of the probability models, according to the first fully connected mapping function, the first transformation vector corresponding to the feature vector is calculated; according to the nonlinear transformation function, the first transformation vector is transformed into a transformation vector; according to the second fully connected mapping function, the second transformation vector corresponding to the transformation vector is calculated; and the second transformation vector is determined as the underwriting result corresponding to the probability model.

3. The system according to claim 2, characterized in that The level probability calculation module is specifically used for: Through each of the probability models, according to the connection matrix, features in the transformation vector are eliminated to obtain a connection vector corresponding to the transformation vector; and according to the second fully connected mapping, a second transformation vector corresponding to the connection vector is calculated.

4. The system according to claim 2, characterized in that The level probability calculation module is specifically used for: By means of each of the probability models, a first product of the feature vector and a first feature transformation matrix is ​​calculated, and a sum of the first product and a first offset is determined as the first conversion vector; And by using each of the probability models, a second product of the transformation vector and a second characteristic transformation matrix is ​​calculated, and a sum of the second product and a second offset is determined as the second conversion vector.

5. The system according to claim 2, characterized in that The level probability calculation module is specifically used for: Through each of the probability models, the elements in the feature vector are normalized; the normalized feature vector is expanded and the expanded feature vector is concatenated into a column vector; and according to the first fully connected mapping function and the column vector, a first conversion vector corresponding to the feature vector is calculated.

6. The system according to claim 1, characterized in that The underwriting level calculation module is used to: For each of the underwriting levels, the probability that the user belongs to the underwriting level is screened from the underwriting results corresponding to each of the probability models, and a probability set that the user belongs to the underwriting level is obtained; A target underwriting level for the user is determined according to a probability set that the user belongs to each of the underwriting levels.

7. A method of data processing, characterized in that: include: Receiving health data of the user collected by a collection device; constructing a feature vector according to the health data; According to the feature vector, using N probability models, respectively calculating the underwriting results corresponding to each of the probability models, the underwriting results including the probability that the user belongs to each underwriting level, wherein N is an integer greater than 0; Calculating the user's insurance level according to the insurance results corresponding to each of the probability models; Outputting the insurance level of the user to an insurance terminal so that the insurance terminal underwrites the user according to the insurance level; Before respectively calculating the core guarantee knot corresponding to each probability model through N probability models according to the feature vector, the method further includes: The N probability models are trained according to the residuals corresponding to the N identical probability models, wherein the residual corresponding to the Mth probability model among the N probability models is the difference between the expected output of the probability model and the sum of the outputs of the first to Mth probability models, M is an integer greater than 0 and less than or equal to N, and the N probability models are trained by residual cascading.

8. An electronic device, characterized in that: include: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method as claimed in claim 7.

9. A computer readable medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to claim 7 is implemented.

Images