CN109754135B

CN109754135B - Credit behavior data processing method, apparatus, storage medium and computer device

Info

Publication number: CN109754135B
Application number: CN201711069348.6A
Authority: CN
Inventors: 黄引刚
Original assignee: Guangzhou Tencent Technology Co Ltd
Current assignee: Guangzhou Tencent Technology Co Ltd
Priority date: 2017-11-03
Filing date: 2017-11-03
Publication date: 2023-05-05
Anticipated expiration: 2037-11-03
Also published as: CN109754135A

Abstract

The invention relates to a credit behavior data processing method, a device, a storage medium and computer equipment, wherein the credit behavior data processing method comprises the following steps: acquiring credit behavior data reported by a corresponding user identifier; querying historical credit scores corresponding to the user identifications; determining a credit score adjustment mode corresponding to the user identifier according to the credit behavior data and the historical credit score; and determining the current credit score corresponding to the user identifier according to the credit score adjustment mode and the credit behavior data and the historical credit score. The scheme provided by the application improves the credit behavior data processing efficiency.

Description

Credit behavior data processing method, apparatus, storage medium and computer device

Technical Field

The present invention relates to the field of computer technologies, and in particular, to a credit behavior data processing method, apparatus, storage medium, and computer device.

Background

With the development of computer technology, internet platforms are gradually beginning to support the processing of credit behavior data of users in the internet and credit scoring of users based on these credit behavior data. Credit behavior data here such as user credit card usage data or user social behavior data, etc.

When credit rating is performed on a user in the conventional technology, after credit behavior data of the user is obtained, the current credit behavior data of the user is usually checked manually to perform credit rating updating result on the user. This introduces a lot of manpower and takes a long time, thereby affecting the credit behavior data processing efficiency.

Disclosure of Invention

Based on this, it is necessary to provide a credit behavior data processing method, apparatus, storage medium and computer device, aiming at the problem that the credit behavior data processing efficiency is low at present.

A credit behavioural data processing method, comprising:

acquiring credit behavior data reported by a corresponding user identifier;

querying historical credit scores corresponding to the user identifications;

determining a credit score adjustment mode corresponding to the user identifier according to the credit behavior data and the historical credit score;

and determining the current credit score corresponding to the user identifier according to the credit score adjustment mode and the credit behavior data and the historical credit score.

A credit action data processing apparatus comprising:

the acquisition module is used for acquiring credit behavior data reported by the corresponding user identifier;

The query module is used for querying historical credit scores corresponding to the user identifiers;

the determining module is used for determining a credit score adjustment mode corresponding to the user identifier according to the credit behavior data and the historical credit score;

and the mapping module is used for determining the current credit score corresponding to the user identifier according to the credit score adjustment mode and the credit behavior data and the historical credit score.

A computer readable storage medium having stored thereon a computer program which, when executed by a processor, causes the processor to perform the steps of:

acquiring credit behavior data reported by a corresponding user identifier;

querying historical credit scores corresponding to the user identifications;

A computer device comprising a memory and a processor, the memory having stored therein a computer program which, when executed by the processor, causes the processor to perform the steps of:

Acquiring credit behavior data reported by a corresponding user identifier;

querying historical credit scores corresponding to the user identifications;

According to the credit behavior data processing method, the device, the storage medium and the computer equipment, the credit behavior data reported by the corresponding user identifier is locally obtained, the historical credit score corresponding to the user identifier can be automatically inquired, and then the credit score adjustment mode is determined according to the currently obtained credit behavior data and the credit score of the user, so that the historical credit score of the user can be adjusted according to the adjustment mode, the real-time credit score of the user after the credit behavior corresponding to the credit behavior data occurs is obtained, time consumption caused by manual processing is avoided, and the credit behavior data processing efficiency is improved.

Drawings

FIG. 1 is a diagram of an application environment for a method of processing credit behavior data in one embodiment;

FIG. 2 is a flow chart of a method of processing credit behavior data in one embodiment;

FIG. 3 is a schematic diagram of a first model in one embodiment;

FIG. 4 is a schematic diagram of a second model in one embodiment;

FIG. 5 is a flow diagram of model updates in one embodiment;

FIG. 6 is a timing diagram of a method of processing credit behavioral data in one embodiment;

FIG. 7 is a flowchart of a method for processing credit action data according to another embodiment;

FIG. 8 is a block diagram of a credit behavioural data processing device in one embodiment;

FIG. 9 is a block diagram of a credit action data processing apparatus in another embodiment;

fig. 10 is an internal structural view of a computer device in one embodiment.

Detailed Description

The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.

FIG. 1 is a diagram of an application environment for a credit behavior data processing method in one embodiment. Referring to fig. 1, the credit behavior data processing method is applied to a credit behavior data processing system. The credit action data processing system includes a computer device 110 and a server 120. The server 120 includes a first server 121, a second server 122, and the like, which correspond to each of a plurality of domains. The computer device 110 is configured to obtain credit behavior data reported by servers corresponding to a plurality of domains, and then execute the credit behavior data processing method. The computer device 110 may be a server or a terminal, in particular. The terminal can be a desktop terminal or a mobile terminal, and the mobile terminal can be at least one of a mobile phone, a tablet computer, a notebook computer and the like. The server may be an independent server, or may be a server cluster formed by a plurality of independent servers.

FIG. 2 is a flow chart of a method for processing credit behavior data in one embodiment. The present embodiment is mainly exemplified by the application of the method to the computer device 110 in fig. 1. Referring to fig. 2, the credit behavior data processing method specifically includes the following steps:

s202, credit behavior data reported by the corresponding user identifier is obtained.

Wherein the user identification is used to uniquely identify a user. The user identification may be a character string comprising at least one of a number, a letter, and a symbol. The credit action data is user action data affecting the credit of the user.

The credit action data may be user credit action data in a plurality of user action scenarios. A user behavior scenario is a scenario that is related to a user and in which there is user behavior. Such as a communication behavior scenario, a payment behavior scenario, or a social behavior scenario, etc.

The credit action data may include credit-enhancing action data and credit-losing action data. The credit-enhancing behavioral data is behavioral data that is affecting credit. The belief-free behavior data is behavior data that negatively affects credit. For example, credit-enhancing behavior data (behavior data that enhances user credit) such as: "credit card on time", "pay charge on time" or "brave on the way", etc. The belief-losing behavior data (behavior data that reduces user credit) such as: "maliciously delinquent pays payable fees", "spread false messages" or "non-civilized behavior", etc.

Specifically, the computer device may pull credit behavior data corresponding to the user identifier from the server corresponding to each user behavior scenario. In one embodiment, the computer device may send the user identifier to the servers corresponding to the user behavior scenes, respectively, and after receiving the user identifier, the server corresponding to each user behavior scene searches the credit behavior data corresponding to the user identifier, and then reports the found credit behavior data corresponding to the user identifier to the computer device.

The server corresponding to the user behavior scene comprises a server corresponding to the communication behavior scene, such as a communication server and the like; the method also comprises a server corresponding to the social behavior scene, such as a WeChat server and the like. One or more servers corresponding to one user behavior scene may be provided. For example, a communication server corresponding to a communication behavior scene, a micro-letter server corresponding to a social behavior scene, a micro-blog server, a bean server and other social servers.

In one embodiment, the computer device may also receive credit behavior data actively reported by a server corresponding to each user behavior scene. When the server corresponding to each user behavior scene collects the credit behavior data, determining the corresponding user identification of the credit behavior data, and reporting the collected credit behavior data and the corresponding user identification to the computer equipment.

When each server corresponding to each user behavior scene collects one piece of credit behavior data, the credit behavior data and the corresponding user identification can be reported to the computer equipment in real time, so that the computer equipment can update the credit score corresponding to the corresponding user identification according to the credit behavior data in real time.

S204, inquiring historical credit scores corresponding to the user identifications.

Wherein the credit score is a value obtained by quantifying the credit of the user. The larger the credit score corresponding to the user identifier is, the better the credit of the user corresponding to the user identifier is.

In this embodiment, the credit score is changed in real time with the change of the credit behavior data corresponding to the corresponding user identifier. For example, the change process of the credit score corresponding to the user identifier and the credit behavior data corresponding to the user identifier may be expressed as < credit score 1, credit behavior data 1, credit score 2, credit behavior data 2, credit score 3, credit behavior data 3 …, credit score i, and credit behavior data i … >. The credit score 1 is an initial credit score, and after the credit behavior corresponding to the credit behavior data 1 occurs, the credit score of the user is changed into a credit score 2 in real time according to the credit behavior data 1. After the credit behavior corresponding to the credit behavior data 2 occurs, the credit score of the user is changed into a credit score 3 in real time according to the credit behavior data 2, and so on.

Specifically, after the computer equipment acquires new credit behavior data, determining a user identifier corresponding to the credit behavior data, and searching the latest credit score with a corresponding relation with the user identifier according to the user identifier to serve as the historical credit score of the current new credit behavior data.

S206, determining a credit score adjustment mode corresponding to the user identification according to the credit behavior data and the historical credit score.

The credit score adjustment mode is a mode for adjusting credit scores. The credit score adjustment means may include whether to adjust the credit score, the magnitude of the credit score adjustment, and the like. Different credit behavior data may correspond to different credit score adjustment manners. Different historical credit scores may also correspond to different credit score adjustment schemes. The credit behavior data and the historical credit score synergistically influence the credit score adjustment. Under the condition that the historical credit scores are the same, different credit behavior data can correspond to different credit score adjustment modes. For example, the higher the severity of the credit action corresponding to the credit action data, the larger the magnitude of the credit score adjustment. Under the condition that the historical credit scores are different, the same credit behavior data can correspond to different credit score adjustment modes. For example, if the credit score of the user is extremely high, the credit score of the user is not increased any more when the payment is paid on time; however, if the credit score of the user is low, the credit score of the user can be increased by repayment on time.

Specifically, the computer device may map the credit behavior data and the historical credit score to the credit score adjustment mode corresponding to the user identifier according to the mapping relationship between the credit behavior data and the historical credit score and the credit score adjustment mode. The mapping relation between the credit behavior data and the historical credit score and the credit score adjustment mode can be specifically a mapping function which takes the credit behavior data and the historical credit score as independent variables and takes the credit score adjustment mode as the dependent variables.

And S208, determining the current credit score corresponding to the user identifier according to the credit score adjustment mode and the credit behavior data and the historical credit score.

Specifically, the computer device may adjust the historical credit score according to the credit behavior data and the credit score according to the credit score adjustment mode to obtain the current credit score corresponding to the user identifier according to the credit score according to the corresponding relationship between the credit behavior data and the historical credit score and the credit score.

In one embodiment, the correspondence between the credit behavior data and the historical credit score established by the credit score adjustment method and the credit score may be a mapping relationship between the credit behavior data and the historical credit score formed by the credit score adjustment method and the credit score, specifically may be a mapping function in which the credit behavior data and the historical credit score are used as a first argument, the credit score adjustment method is used as a second argument, and the current credit score corresponding to the user identifier is used as a dependent variable.

According to the credit behavior data processing method, the historical credit scores corresponding to the user identifications can be automatically inquired when the credit behavior data reported by the corresponding user identifications are obtained locally, and then the credit score adjustment mode is determined according to the currently obtained credit behavior data and the credit scores of the users, so that the historical credit scores of the users can be adjusted according to the adjustment mode, the real-time credit scores of the users after the credit behaviors corresponding to the credit behavior data occur are obtained, time consumption caused by manual processing is avoided, and the credit behavior data processing efficiency is improved.

In one embodiment, S206 includes: and inputting the credit behavior data and the historical credit score into the first model together to obtain a credit score adjustment mode corresponding to the user identification output by the first model. S208 includes: and inputting the credit behavior data, the historical credit score and the credit score adjustment mode into the second model together to obtain the current credit score corresponding to the user identifier output by the second model.

The first model is a machine learning model which specifically evaluates the credit score adjustment mode after training. The machine learning model may employ a neural network model, a support vector machine, a logistic regression model, or the like. Neural network models such as convolutional neural networks, back propagation neural networks, feedback neural networks, radial basis neural networks, or ad hoc neural networks, and the like.

Specifically, the computer device may directly use the credit behavior data and the historical credit score as input of the first model obtained by training in advance, so that the first model outputs a credit score adjustment mode corresponding to the user identifier.

In one embodiment, when the first model is trained in advance, the computer device may collect credit behavior data under each user behavior scene corresponding to a plurality of user identifiers as training samples, and determine a credit score adjustment mode corresponding to the credit behavior data according to the collected credit behavior data, so as to add training labels to the samples. The computer device may be trained to obtain a first model based on the samples and the corresponding training tags.

In one embodiment, the credit behavior data and the historical credit score are input into the first model together to obtain a credit score adjustment mode corresponding to the user identifier output by the first model, which includes: extracting credit behavior keywords included in the credit behavior data; the credit behavior keywords and the historical credit scores are input into a first model together; and transforming the credit behavior keywords and the historical credit scores according to the connection weights of all the connection layers in the first model to obtain a credit score adjustment mode corresponding to the user identification.

Wherein the credit action keyword is a character or character string reflecting characteristics of credit action. For example, the credit action data "user A borrow X, pay on demand", and the extracted credit action keywords are "X" and "pay on demand".

Specifically, the computer device may use the text-form credit behavior keywords and the historical credit scores together as input of the first model obtained by training in advance, so that the first model outputs a credit score adjustment mode corresponding to the user identifier. The computer equipment can also directly vector the credit behavior data in the text form or vector the credit behavior keywords in the text form to obtain credit behavior vectors, and then uses the credit behavior vectors and the historical credit scores as the input of the first model obtained in advance so that the first model outputs the credit score adjustment mode corresponding to the user identifier.

Where a vector is data used to express data in text form in mathematical form. For example, "pay-on-schedule" in text form is expressed as mathematical form "[0 0 0 1 0 0 0 0 0 0 0.]", and at this time "[0 0 0 1 0 0 0 0 0 0 0.]" is a vector of "pay-on-schedule". It will be appreciated that the vector into which the text form data is converted is not limited herein, as long as the text form data can be mathematically represented.

In one embodiment, the computer device may further perform standard normalization processing on the historical credit score to obtain a dimensionless credit score, and use the credit behavior keyword and the dimensionless credit score together as input of a first model obtained by training in advance, so that the first model outputs a credit score adjustment manner corresponding to the user identifier. Therefore, after the original data is subjected to data standard normalization processing, all the data are in the same order of magnitude, and the method is suitable for comprehensive comparison and evaluation.

The historical credit score is subjected to standard normalization processing, and can be calculated according to the following formula:

wherein z.score is the ratio of the difference between the credit score corresponding to the current user identifier and the average value of the credit scores corresponding to all user identifiers to the variance of the credit scores corresponding to all user identifiers.

Further, the fully connected layer is a layer for performing nonlinear processing on input data. Each processing node (neuron) in the fully connected layer inputs for processing. The pair connection weight is a model parameter reflecting the correspondence between the input and the output. Specifically, the computer equipment carries out nonlinear transformation on the credit behavior keywords and the historical credit scores according to the connection weights of all the connection layers in the first model, and a credit score adjustment mode corresponding to the user identification is obtained.

In one embodiment, the credit score output by the first model may be a credit score adjustment expressed in a multidimensional mathematical form, such as an N-dimensional vector. Different vectors correspond to different forms of credit score adjustment expressed in different text forms. For example, [0 0 0 1 0 0 0 0 0 0 0.] represents adjustment credit score and the adjustment amplitude is 0.3, [0 0 0 0 0 0 0 0 1 0 0.] represents adjustment credit score and the adjustment amplitude is 0.8, and so on.

In this embodiment, the computer device may adjust the dimension of the credit score adjustment of the mathematical form expression of the output, and the computer device may slow down the gradient of the change of the credit score adjustment by increasing the dimension. It will be appreciated that the higher the dimension of the data, the more data may be included, and the more credit score adjustment modes may be corresponded, so that the credit score adjustment may be not only integer adjustment, but also floating point adjustment. Wherein integer adjustment means adjusting the credit score with integer data. Floating point type adjustment means adjusting credit scores with floating point type data.

In the above embodiment, the credit score adjustment mode is learned and evaluated according to the credit behavior data and the historical credit score by using the strong learning and representing capability of the machine learning algorithm, and the credit score adjustment mode is evaluated for the user by using the first model obtained by training, so that the credit score effect for the user is better than that of the conventional method.

Fig. 3 shows a schematic diagram of a first model in one embodiment. Referring to fig. 3, the computer device may use the credit behavior data and the historical credit score together as inputs to the first model. Assuming that the dimension of the input data is N1, the input data in N1 dimension is nonlinear changed into intermediate data in N2 dimension through the connection weight W1 of the first layer full-connection layer, the input data in N2 dimension is nonlinear changed into intermediate data in N3 dimension through the connection weight W2 of the second layer full-connection layer, the input data in N3 dimension is nonlinear changed into output data in N4 dimension through the connection weight W3 of the third layer full-connection layer, and therefore the input data in N1 dimension is nonlinear changed into the output data in N4 dimension through the first model through each full-connection layer in a layer-by-layer processing mode, and finally the credit score adjustment mode in N4 dimension is output.

In one embodiment, the credit behavior data, the historical credit score and the credit score adjustment mode are input into the second model together to obtain the current credit score corresponding to the user identifier output by the second model, which comprises the following steps: the credit behavior keywords, the historical credit scores and the credit score adjustment mode are input into a second model together; and transforming the credit behavior keywords, the historical credit scores and the credit score adjustment modes according to the connection weight of each full connection layer in the second model to obtain the current credit scores corresponding to the user identifications.

Wherein the second model is a machine learning model that is trained to specifically evaluate credit scores. Specifically, the computer device may directly use the credit behavior data, the historical credit score and the credit score adjustment manner together as input of the second model obtained by training in advance, so that the second model outputs the current credit score corresponding to the user identifier.

In one embodiment, when the computer device trains the second model in advance, credit behavior data under each user behavior scene corresponding to a plurality of user identifiers can be collected first as training samples, and credit scores corresponding to the credit behavior data are judged according to the collected credit behavior data so as to add training labels to the samples. The computer device may be trained to derive the second model based on the samples and the corresponding training tags.

In one embodiment, the computer device may use the credit action vector, the historical credit score after the standard normalization processing, and the credit score adjustment manner output by the first model together as the input of the second model obtained by training in advance, so that the second model outputs the current credit score corresponding to the user identifier. At this time, the dimension of the second model input data is the sum of the dimension of the first model input data and the dimension of the output data.

In the above embodiment, the credit scoring is learned according to the credit behavior data, the historical credit scoring and the credit scoring adjustment mode by using the strong learning and representing capability of the machine learning algorithm, and when the credit scoring is performed on the user by using the second model obtained by training, the effect of performing the credit scoring on the user is better than that of performing the credit scoring on the user by using the traditional method.

Figure 4 shows a schematic diagram of a second model in one embodiment. Referring to fig. 3, the computer device may use credit behavior data, historical credit scores, and credit score adjustment means together as inputs to the first model. Assuming that the dimension of the input data is M1 (m1=n1+n4), the input data in the M1 dimension is nonlinear changed into the intermediate data in the M2 dimension through the connection weight W4 of the first layer full-connection layer, the input data in the M2 dimension is nonlinear changed into the intermediate data in the M3 dimension through the connection weight W5 of the second layer full-connection layer, the input data in the M3 dimension is nonlinear changed into the output data in the 1 dimension through the connection weight W6 of the third layer full-connection layer, and thus the input data in the M3 dimension is nonlinear changed into the output data in the 1 dimension through the first model through each full-connection layer by layer processing, and finally the credit score is output.

In the above embodiment, the process of updating the credit score of the user is divided into two stages, the credit score adjustment mode is estimated according to the credit behavior data and the historical credit score, and then a new credit score is obtained according to the credit behavior data and the historical credit score on the basis of the estimation result, so that the credit score estimation of the user is more accurate and reasonable.

FIG. 5 is a flowchart illustrating steps for model update in a credit data processing method according to one embodiment. Referring to fig. 5, the step of updating the model specifically includes the steps of:

s502, inquiring the historical credit feedback scores corresponding to the user identifiers.

Wherein the credit feedback score is data used as a training sample label. The credit feedback score is an expected credit score calculated according to a particular algorithm. After the computer device obtains the new credit behavior data and obtains the new credit score by using the first model and the second model, a new model update sample can be formed, and the first model and the second model are updated by using the new model update sample.

S504, obtaining the current credit feedback score corresponding to the user identification according to the credit behavior data and the historical credit feedback score.

Specifically, the computer device may first calculate the feedback score adjustment corresponding to the new credit behavior data obtained. The feedback score adjustment includes a forward feedback score adjustment, a reverse feedback score adjustment, and no feedback. For example, assuming that the new credit action data is "user borrowing x and pay-off-schedule x", the feedback score adjustment amount corresponding to the credit action data is a positive feedback score adjustment amount, specifically, x. Assuming that the new credit behavior data is "x for borrowing by the user, and the amount of money cannot be paid on schedule is y", the feedback score adjustment amount corresponding to the credit behavior data is a reverse feedback score adjustment amount, specifically-y. Assuming that the new credit behavior data is "no repayment information", the feedback score adjustment amount corresponding to the credit behavior data is no feedback.

In one embodiment, S504 includes: determining a current credit feedback score corresponding to the user identifier according to the following formula:

X _i+1 ＝Reward _i +gamma*X _i (3)

wherein R is _i Obtaining the corresponding current credit feedback score after the ith credit behavior data corresponding to the user identification; x is X _i The feedback score adjustment quantity corresponding to the i-th credit behavior data acquired currently is used for adjusting the feedback score adjustment quantity; reward _i The corresponding historical credit feedback scores are obtained after the i-1 th credit behavior data are obtained; gamma is the feedback score adjustment quantity coefficient.

For example, gamma may be a fixed constant, such as 0.9. The gamma of 0.9 indicates that the feedback score adjustment decays with a loss of 0.9 over time. Such as: the feedback score adjustment determined from the currently acquired credit behavior data is 100, and can only contribute to the effect of 100 x 0.9=90.

In this embodiment, the feedback score adjustment amount corresponding to each credit behavior is accumulated in a time attenuation manner, so that the new feedback score adjustment amount obtained each time can reasonably reflect the influence of the previous credit behavior, and the calculation of the credit feedback score is more accurate through the continuity of the feedback score adjustment amount on the credit behavior.

S506, the historical credit scores, the credit behavior data, the credit score adjustment mode and the current credit feedback score are correspondingly stored as model updating samples.

Specifically, the computer device sets an initial credit score and an initial credit feedback score corresponding to the user identifier. And after receiving the reported first credit behavior data, obtaining a second credit score adjustment mode and a second credit score through the first model and the second model according to the initial credit score and the first credit behavior data. And calculates a second credit feedback score according to the above formulas (2), (3) and (4). Thus, a new model update sample 1< initial credit score, first credit behavior data, second credit score adjustment mode, second credit feedback score > is obtained, and a new model update sample can be formed according to the above processing after the computer equipment receives one credit behavior data. The computer equipment can obtain a model updating sample set, and the format of each model updating sample is < historical credit score n, credit behavior data n, credit score adjustment mode n and current credit feedback score n >, wherein the historical credit score n in the < historical credit score n, the credit behavior data n and the credit score adjustment mode n in the current credit feedback score n > is the credit score which is latest in time before the credit behavior data n is received, the credit score adjustment mode n is the credit score adjustment mode which is evaluated after the credit behavior data n is received, and the current credit feedback score n is the credit feedback score which is calculated after the credit behavior data n is received.

S508, updating the first model and the second model according to the model updating sample.

In one embodiment, S508 includes: when the number of the model updating samples reaches a preset number, obtaining the model updating samples; inputting the historical credit scores, credit behavior data and credit score adjustment modes in the acquired model updating samples into a second model; constructing a loss function according to the credit score sample output by the second model and the current credit feedback score in the acquired model updating sample; the first model and the second model are updated according to the loss function.

In one embodiment, the loss function may be determined by the following formula:

wherein m is the number of user identifications; score, z _k The difference value between the current credit score corresponding to the kth user identifier and the current credit score average value corresponding to all user identifiers and the variance ratio of the current credit score corresponding to all user identifiers are obtained; score of R. _k And (3) the difference value between the current credit feedback score corresponding to the kth user identifier and the average value of the current credit feedback scores corresponding to all the user identifiers is the ratio of the difference value between the current credit feedback score corresponding to the kth user identifier and the variance of the current credit feedback score corresponding to all the user identifiers.

Specifically, the computer device may preset a trigger condition for updating the model. The trigger condition may be that the number of model update samples reaches a preset number. It can be appreciated that the value of the preset number depends on the specific usage scenario, and the model learning is more stable as the preset number is larger.

Further, after the computer equipment judges that the triggering condition of the updated model is met currently, a model updating sample can be selected optionally, or the model updating sample is selected according to time sequence, and the historical credit score, the credit behavior data and the credit score adjustment mode in the selected model updating sample are input into a second model to obtain the corresponding current credit score of the user identifier corresponding to the training sample. At this time, the credit scores corresponding to the other user identifications are updated except the corresponding credit scores of the user identifications. The computer device may determine an average value of current credit scores corresponding to all user identifications, a variance of current credit scores corresponding to all user identifications, an average value of current credit feedback scores corresponding to all user identifications, and a variance of current credit feedback scores corresponding to all user identifications, thereby constructing a loss function, and adjust connection weights of all the connection layers in the first model and the second model according to a direction of minimizing the loss function.

In one embodiment, when storing the model training samples, the computer device may store the time of receipt of the credit behavior data included in the model training samples, and the corresponding user identification, accordingly. In this way, when computing the average value of the current credit scores corresponding to all the user identifications, the variance of the current credit scores corresponding to all the user identifications, the average value of the current credit feedback scores corresponding to all the user identifications, and the variance of the current credit feedback scores corresponding to all the user identifications, the computer device may determine the receiving time of the credit behavior data in the model update sample used by the update, thereby determining the current credit scores corresponding to all the user identifications and the current credit feedback scores corresponding to all the user identifications at the receiving time. That is, the latest credit scores corresponding to all user identifications and the latest credit feedback scores corresponding to all user identifications before the receiving time.

After the connection weight of each full connection layer in the first model and the second model is adjusted, the computer equipment can continue to select model update samples to update the model until the preset iteration times are reached, or use all the model update samples once.

In one embodiment, updating the first model and the second model according to the loss function includes: keeping the connection weight of all the connection layers in the first model unchanged, and adjusting the connection weight of all the connection layers in the second model according to the adjustment direction of the minimized loss function; and keeping the connection weight of each full connection layer in the second model unchanged, and adjusting the connection weight of each full connection layer in the first model according to the adjustment direction of the minimized loss function.

It should be understood that the adjustment sequence of the first model and the second model is not limited herein, and the computer device may fix the first model to adjust the second model first, adjust the first model after the second model is adjusted, fix the second model to adjust the first model, adjust the second model after the first model is adjusted, and may also circularly adjust the first model and the second model.

In this embodiment, in the process of updating the first model, the model parameters of the second model are kept unchanged, the model parameters of the first model are adjusted, in the process of updating the second model, the model parameters of the second model are adjusted, and the model updating efficiency can be improved.

Specifically, the computer device may adjust the connection weights of the full connection layers in the first model and the second model using a BP (Back Propagation) gradient descent algorithm to minimize the loss function.

In the embodiment, the model can be updated during the use period of the model, so that the real-time performance and the effectiveness of the use of the model are ensured.

In the above embodiment, new model update samples are continuously constructed, and the first model and the second model are updated according to the new model update samples, so that credit score evaluation performed by the first model and the second model is more accurate.

FIG. 6 illustrates a timing diagram of a method of processing credit behavioral data in one embodiment. Referring to fig. 6, after the user performs credit, a server in the corresponding domain collects credit data corresponding to the credit, and the server reports the credit data to the computer device corresponding to the user identifier.

After the computer equipment acquires the credit behavior data reported by the corresponding user identifier, inquiring the historical credit score corresponding to the user identifier, and inputting the credit behavior data and the historical credit score into the first model to obtain a credit score adjustment mode output by the first model. And the computer equipment outputs the credit behavior data, the historical credit score and the credit score adjustment mode to a second model to obtain the credit score output by the second model, so that the current credit score obtained by updating the user in real time after the credit behavior occurs is obtained. The computer device processes the credit behavior data according to the processing procedure when receiving one credit behavior data.

The computer device may also query historical credit feedback scores corresponding to the user identifications; and obtaining the current credit feedback score corresponding to the user identifier according to the credit behavior data and the historical credit feedback score. And then storing the historical credit scores, the credit behavior data, the credit score adjustment mode and the current credit feedback score as model updating samples correspondingly. And updating the first model and the second model through the model updating sample.

As shown in fig. 7, in a specific embodiment, the credit behavior data processing method specifically includes the following steps:

s702, credit behavior data reported by the corresponding user identifier is obtained.

S704, inquiring historical credit scores corresponding to the user identifications.

S706, extracting credit behavior keywords included in the credit behavior data; the credit behavior keywords and the historical credit scores are entered together into a first model.

And S708, converting the credit behavior keywords and the historical credit scores according to the connection weights of all the connection layers in the first model to obtain a credit score adjustment mode corresponding to the user identification.

S710, the credit behavior keywords, the historical credit scores and the credit score adjustment mode are input into the second model together.

S712, according to the connection weight of each full connection layer in the second model, the credit behavior keywords, the historical credit scores and the credit score adjustment mode are transformed to obtain the current credit scores corresponding to the user identifications.

S714, inquiring the historical credit feedback scores corresponding to the user identifications; and obtaining the current credit feedback score corresponding to the user identifier according to the credit behavior data and the historical credit feedback score.

Specifically, the current credit feedback score corresponding to the user identifier is determined according to the following formula:

X _i+1 ＝Reward _i +gamma*X _i (3)

and S716, correspondingly storing the historical credit scores, the credit behavior data, the credit score adjustment mode and the current credit feedback scores as model updating samples.

S718, judging whether the number of the model updating samples reaches the preset number, if so, jumping to S720; if not, go to S702.

S720, obtaining a model update sample; and inputting the historical credit scores, the credit behavior data and the credit score adjustment modes in the acquired model update samples into a second model.

S722, constructing a loss function according to the credit score sample output by the second model and the current credit feedback score in the acquired model update sample.

Specifically, the loss function may be determined by the following formula:

And S724, keeping the connection weight of each full connection layer in the first model unchanged, and adjusting the connection weight of each full connection layer in the second model according to the adjustment direction of the minimized loss function.

S726, keeping the connection weight of each full connection layer in the second model unchanged, and adjusting the connection weight of each full connection layer in the first model according to the adjustment direction of the minimized loss function.

In this embodiment, the historical credit score corresponding to the user identifier can be automatically queried when the credit behavior data reported by the corresponding user identifier is obtained locally, and then the adjustment mode of the credit score is determined according to the currently obtained credit behavior data and the credit score of the user, so that the historical credit score of the user can be adjusted according to the adjustment mode, the real-time credit score of the user after the credit behavior corresponding to the credit behavior data occurs is obtained, time consumption caused by manual processing is avoided, and the credit behavior data processing efficiency is improved.

It should be understood that, although the steps in the flowcharts of the above embodiments are shown in order as indicated by the arrows, these steps are not necessarily performed in order as indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in the embodiments described above may include a plurality of sub-steps or stages that are not necessarily performed at the same time, but may be performed at different times, and the order of execution of the sub-steps or stages is not necessarily sequential, but may be performed in turn or alternately with at least a portion of other steps or sub-steps of other steps.

As shown in fig. 8, in one embodiment, a credit behavior data processing apparatus 800 is provided. Referring to fig. 8, the credit action data processing apparatus 800 includes: an acquisition module 801, a query module 802, a determination module 803, and a mapping module 804.

And the obtaining module 801 is configured to obtain credit behavior data reported by the corresponding user identifier.

A query module 802 for querying historical credit scores corresponding to the user identifications.

And the determining module 803 is configured to determine, according to the credit behavior data and the historical credit score, a credit score adjustment manner corresponding to the user identifier.

The mapping module 804 is configured to determine, according to the credit score adjustment manner, a current credit score corresponding to the user identifier according to the credit behavior data and the historical credit score.

According to the credit behavior data processing device 800, the historical credit score corresponding to the user identifier can be automatically queried when the credit behavior data reported by the corresponding user identifier is acquired locally, and then the credit score adjustment mode is determined according to the currently acquired credit behavior data and the credit score of the user, so that the historical credit score of the user can be adjusted according to the adjustment mode, the real-time credit score of the user after the credit behavior corresponding to the credit behavior data occurs is obtained, time consumption caused by manual processing is avoided, and the credit behavior data processing efficiency is improved.

In one embodiment, the determining module 803 is further configured to input the credit behavior data and the historical credit score together into the first model, so as to obtain a credit score adjustment manner corresponding to the user identifier output by the first model. The mapping module 804 is further configured to input the credit behavior data, the historical credit score, and the credit score adjustment mode into the second model together, so as to obtain a current credit score corresponding to the user identifier output by the second model.

In one embodiment, the determining module 803 is further configured to extract a credit action keyword included in the credit action data; the credit behavior keywords and the historical credit scores are input into a first model together; and transforming the credit behavior keywords and the historical credit scores according to the connection weights of all the connection layers in the first model to obtain a credit score adjustment mode corresponding to the user identification.

In one embodiment, the mapping module 804 is further configured to input the credit action keyword, the historical credit score, and the credit score adjustment mode together into the second model; and transforming the credit behavior keywords, the historical credit scores and the credit score adjustment modes according to the connection weight of each full connection layer in the second model to obtain the current credit scores corresponding to the user identifications.

As shown in fig. 8, in one embodiment, the credit behavior data processing apparatus 800 further includes: model update module 805.

The model updating module 805 is configured to query a historical credit feedback score corresponding to the user identifier; obtaining a current credit feedback score corresponding to the user identifier according to the credit behavior data and the historical credit feedback score; correspondingly storing historical credit scores, credit behavior data, credit score adjustment modes and current credit feedback scores as model updating samples; and updating the first model and the second model according to the model updating sample.

In one embodiment, the model update module 805 is further configured to determine the current credit feedback score corresponding to the user identification according to the following formula:

X _i+1 ＝Reward _i +gamma*X _i (3)

In one embodiment, the model update module 805 is further configured to obtain model update samples when the number of model update samples reaches a preset number; inputting the historical credit scores, credit behavior data and credit score adjustment modes in the acquired model updating samples into a second model; constructing a loss function according to the credit score sample output by the second model and the current credit feedback score in the acquired model updating sample; the first model and the second model are updated according to the loss function.

In one embodiment, the model updating module 805 is further configured to keep the connection weights of all the connection layers in the first model unchanged, and adjust the connection weights of all the connection layers in the second model according to an adjustment direction that minimizes the loss function; and keeping the connection weight of each full connection layer in the second model unchanged, and adjusting the connection weight of each full connection layer in the first model according to the adjustment direction of the minimized loss function.

In one embodiment, the model update module 805 is further configured to determine the loss function by:

the loss function may be determined by the following formula:

FIG. 10 illustrates an internal block diagram of a computer device in one embodiment. The computer device may be specifically the computer device 110 of fig. 1. As shown in fig. 10, the computer device includes a processor, a memory, and a network interface connected by a system bus. The memory includes a nonvolatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system, and may also store a computer program that, when executed by a processor, causes the processor to implement a credit behavior data processing method. The internal memory may also have stored therein a computer program which, when executed by the processor, causes the processor to perform the credit action data processing method. It will be appreciated by those skilled in the art that the structure shown in fig. 10 is merely a block diagram of some of the structures associated with the present application and is not limiting of the computer device to which the present application may be applied, and that a particular computer device may include more or fewer components than shown, or may combine certain components, or have a different arrangement of components.

In one embodiment, the credit behavior data processing apparatus provided in the present application may be implemented in the form of a computer program, where the computer program may run on a computer device as shown in fig. 10, and a nonvolatile storage medium of the computer device may store respective program modules that make up the credit behavior data processing apparatus, for example, an acquisition module 801, a query module 802, a determination module 803, a mapping module 804, and the like shown in fig. 8. The computer program comprising the individual program modules causes the processor to carry out the steps in the credit behavior data processing method of the individual embodiments of the present application described in the present specification.

For example, the computer device shown in fig. 10 may acquire the credit action data reported by the corresponding user identifier through the acquisition module 801 in the credit action data processing apparatus 800 shown in fig. 8. Historical credit scores corresponding to the user identifications are queried via query module 802. And determining a credit score adjustment mode corresponding to the user identification according to the credit behavior data and the historical credit score through a determination module 803. And determining a current credit score corresponding to the user identification according to the credit score adjustment mode through the mapping module 804 and the credit behavior data and the historical credit score.

In one embodiment, a computer readable storage medium is provided, having a computer program stored thereon, which when executed by a processor causes the processor to perform the steps of: acquiring credit behavior data reported by a corresponding user identifier; querying historical credit scores corresponding to the user identifications; determining a credit score adjustment mode corresponding to the user identifier according to the credit behavior data and the historical credit score; and determining the current credit score corresponding to the user identifier according to the credit score adjustment mode and the credit behavior data and the historical credit score.

In one embodiment, determining a credit score adjustment mode corresponding to the user identifier according to the credit behavior data and the historical credit score includes: and inputting the credit behavior data and the historical credit score into the first model together to obtain a credit score adjustment mode corresponding to the user identification output by the first model. According to the credit score adjustment mode, determining the current credit score corresponding to the user identifier according to the credit behavior data and the historical credit score, wherein the method comprises the following steps: and inputting the credit behavior data, the historical credit score and the credit score adjustment mode into the second model together to obtain the current credit score corresponding to the user identifier output by the second model.

In one embodiment, the computer program, when executed by the processor, further causes the processor to perform the steps of: inquiring historical credit feedback scores corresponding to the user identifications; obtaining a current credit feedback score corresponding to the user identifier according to the credit behavior data and the historical credit feedback score; correspondingly storing historical credit scores, credit behavior data, credit score adjustment modes and current credit feedback scores as model updating samples; and updating the first model and the second model according to the model updating sample.

In one embodiment, obtaining the current credit feedback score corresponding to the user identifier according to the credit behavior data and the historical credit feedback score includes: determining a current credit feedback score corresponding to the user identifier according to the following formula:

X _i+1 ＝Reward _i +gamma*X _i (3)

In one embodiment, updating the first model and the second model based on the model update samples includes: when the number of the model updating samples reaches a preset number, obtaining the model updating samples; inputting the historical credit scores, credit behavior data and credit score adjustment modes in the acquired model updating samples into a second model; constructing a loss function according to the credit score sample output by the second model and the current credit feedback score in the acquired model updating sample; the first model and the second model are updated according to the loss function.

In one embodiment, the loss function is determined by the following formula:

the loss function may be determined by the following formula:

According to the storage medium, the historical credit score corresponding to the user identifier can be automatically inquired when the credit behavior data reported by the corresponding user identifier is obtained locally, and then the mode of adjusting the credit score is determined in real time according to the currently obtained credit behavior data and the credit score of the user, so that the historical credit score of the user can be adjusted according to the adjustment mode, the real-time credit score of the user after the credit behavior corresponding to the credit behavior data occurs is obtained, time consumption caused by manual processing is avoided, and the credit behavior data processing efficiency is improved.

In one embodiment, a computer device is provided that includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of: acquiring credit behavior data reported by a corresponding user identifier; querying historical credit scores corresponding to the user identifications; determining a credit score adjustment mode corresponding to the user identifier according to the credit behavior data and the historical credit score; and determining the current credit score corresponding to the user identifier according to the credit score adjustment mode and the credit behavior data and the historical credit score.

X _i+1 ＝Reward _i +gamma*X _i (3)

In one embodiment, the loss function is determined by the following formula:

the loss function may be determined by the following formula:

According to the computer equipment, the historical credit score corresponding to the user identifier can be automatically inquired when the credit behavior data reported by the corresponding user identifier is obtained locally, and then the mode of adjusting the credit score is determined in real time according to the currently obtained credit behavior data and the credit score of the user, so that the historical credit score of the user can be adjusted according to the adjustment mode, the real-time credit score of the user after the credit behavior corresponding to the credit behavior data occurs is obtained, time consumption caused by manual processing is avoided, and the credit behavior data processing efficiency is improved.

Those skilled in the art will appreciate that implementing all or part of the above-described methods may be accomplished by way of computer programs, which may be stored on a non-transitory computer readable storage medium, which when executed may comprise the steps of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the various embodiments provided herein may include non-volatile and/or volatile memory. The nonvolatile memory can 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. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), memory bus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), among others.

For the sake of brevity, all of the possible combinations of the features of the above embodiments are not described, however, the scope of the description should be considered as if there are no contradictions between the combinations of the features.

The foregoing examples illustrate only a few embodiments of the invention, which are described in detail and are not to be construed as limiting the scope of the invention. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the invention, which are all within the scope of the invention. Accordingly, the scope of protection of the present invention is to be determined by the appended claims.

Claims

1. A credit behavioural data processing method, comprising:

acquiring credit behavior data reported by a corresponding user identifier, extracting credit behavior keywords included in the credit behavior data, vectorizing the credit behavior keywords, and obtaining a credit behavior vector;

querying historical credit scores corresponding to the user identifications, and carrying out standard normalization processing on the historical credit scores to obtain dimensionless credit scores, wherein the dimensionless credit scores are positively correlated with the difference value between the credit scores corresponding to the user identifications and the credit score average value corresponding to all the user identifications and the ratio of the variances of the credit scores corresponding to all the user identifications;

The credit action vector and the dimensionless credit score are input into a first model together, the credit action vector and the dimensionless credit score are transformed according to the connection weight of each full connection layer in the first model, a credit score adjustment mode corresponding to the user identifier is obtained, and the first model is a machine learning model which is trained and then specifically evaluates the credit score adjustment mode;

and commonly inputting the credit behavior vector, the dimensionless credit score and the credit score adjustment mode output by the first model into a second model, and transforming the credit behavior vector, the dimensionless credit score and the credit score adjustment mode according to the connection weight of each full-connection layer in the second model to obtain the current credit score corresponding to the user identifier, wherein the second model is a machine learning model for specifically evaluating the credit score after training.

2. The method according to claim 1, wherein the method further comprises:

inquiring historical credit feedback scores corresponding to the user identifications;

obtaining a current credit feedback score corresponding to the user identifier according to the credit behavior data and the historical credit feedback score;

Correspondingly storing the historical credit scores, the credit behavior data, the credit score adjustment modes and the current credit feedback scores as model updating samples;

updating the first model and the second model according to the model updating sample.

3. The method according to claim 2, wherein the obtaining the current credit feedback score corresponding to the user identifier according to the credit behavior data and the historical credit feedback score includes:

determining a current credit feedback score corresponding to the user identifier according to the following formula:

X _i+1 ＝Reward _i +gamma*X _i

4. The method of claim 2, wherein updating the first model and the second model from the model update samples comprises:

when the number of the model updating samples reaches a preset number, obtaining model updating samples;

Inputting the historical credit scores, credit behavior data and credit score adjustment modes in the acquired model updating samples into a second model;

constructing a loss function according to the credit score sample output by the second model and the current credit feedback score in the acquired model updating sample;

updating the first model and the second model according to the loss function.

5. The method of claim 4, wherein updating the first model and the second model according to the loss function comprises:

keeping the connection weight of each full connection layer in the first model unchanged, and adjusting the connection weight of each full connection layer in the second model according to the adjustment direction of minimizing the loss function;

and keeping the connection weight of each full connection layer in the second model unchanged, and adjusting the connection weight of each full connection layer in the first model according to the adjustment direction of minimizing the loss function.

6. The method of claim 4, wherein the loss function is determined by the following formula:

7. A credit action data processing apparatus comprising:

the acquisition module is used for acquiring credit behavior data reported by the corresponding user identifier, extracting credit behavior keywords included in the credit behavior data, vectorizing the credit behavior keywords, and obtaining a credit behavior vector;

the query module is used for querying the historical credit scores corresponding to the user identifications, carrying out standard normalization processing on the historical credit scores to obtain dimensionless credit scores, and forming positive correlation with the difference value between the credit scores corresponding to the user identifications and the credit score average value corresponding to all user identifications and the ratio of the variances of the credit scores corresponding to all user identifications;

the determining module is used for inputting the credit action vector and the dimensionless credit score into a first model together, transforming the credit action vector and the dimensionless credit score according to the connection weight of each full-connection layer in the first model to obtain a credit score adjustment mode corresponding to the user identifier, wherein the first model is a machine learning model for specifically evaluating the credit score adjustment mode after training;

The mapping module is used for inputting the credit action vector, the dimensionless credit score and the credit score adjustment mode output by the first model into a second model together, transforming the credit action vector, the dimensionless credit score and the credit score adjustment mode according to the connection weight of each full-connection layer in the second model to obtain the current credit score corresponding to the user identifier, wherein the second model is a machine learning model for specifically evaluating the credit score after training.

8. The apparatus of claim 7, wherein the apparatus further comprises:

the model updating module is used for inquiring the historical credit feedback scores corresponding to the user identifiers; obtaining a current credit feedback score corresponding to the user identifier according to the credit behavior data and the historical credit feedback score; correspondingly storing the historical credit scores, the credit behavior data, the credit score adjustment modes and the current credit feedback scores as model updating samples; updating the first model and the second model according to the model updating sample.

9. The apparatus of claim 8, wherein the formula for determining the current credit feedback score corresponding to the user identification is:

X _i+1 ＝Reward _i +gamma*X _i

10. The apparatus of claim 8, wherein the model update module is further configured to obtain model update samples when the number of model update samples reaches a preset number; inputting the historical credit scores, credit behavior data and credit score adjustment modes in the acquired model updating samples into a second model; constructing a loss function according to the credit score sample output by the second model and the current credit feedback score in the acquired model updating sample; updating the first model and the second model according to the loss function.

11. The apparatus of claim 10, wherein the model update module is further configured to

12. The apparatus of claim 10, wherein the formula for determining the loss function is:

13. A computer readable storage medium having stored thereon a computer program which, when executed by a processor, causes the processor to perform the steps of the method according to any of claims 1 to 6.

14. A computer device comprising a memory and a processor, the memory having stored therein a computer program which, when executed by the processor, causes the processor to perform the steps of the method of any of claims 1 to 6.