CN112862602B - User request determining method, storage medium and electronic device - Google Patents
User request determining method, storage medium and electronic device Download PDFInfo
- Publication number
- CN112862602B CN112862602B CN202110336662.6A CN202110336662A CN112862602B CN 112862602 B CN112862602 B CN 112862602B CN 202110336662 A CN202110336662 A CN 202110336662A CN 112862602 B CN112862602 B CN 112862602B
- Authority
- CN
- China
- Prior art keywords
- user
- debt
- stock
- index
- sharing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 66
- 238000004422 calculation algorithm Methods 0.000 claims description 38
- 239000011159 matrix material Substances 0.000 claims description 19
- 238000004590 computer program Methods 0.000 claims description 13
- 238000012549 training Methods 0.000 claims description 11
- 230000014509 gene expression Effects 0.000 claims description 10
- 230000006399 behavior Effects 0.000 claims description 9
- 238000010606 normalization Methods 0.000 claims description 5
- 238000007781 pre-processing Methods 0.000 claims description 4
- 238000013523 data management Methods 0.000 abstract description 8
- 238000011156 evaluation Methods 0.000 description 15
- 230000009466 transformation Effects 0.000 description 13
- 230000008569 process Effects 0.000 description 9
- 238000013210 evaluation model Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000011002 quantification Methods 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 2
- 230000003542 behavioural effect Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000012805 post-processing Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000012854 evaluation process Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/03—Credit; Loans; Processing thereof
Landscapes
- Business, Economics & Management (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Engineering & Computer Science (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- Technology Law (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The embodiment of the invention provides a method for determining user requests, which is used for determining the security level of a user based on the determined co-debt level score of the user and determining whether the user requests of the user are passed or not based on the security level so as to avoid the situation that the data management system is failed due to unreasonable passing of the user requests.
Description
Technical Field
The present invention relates to the field of data management, and in particular, to a method for determining a user request, a storage medium, and an electronic device.
Background
With the increasing volume of the consumer credit market, the opportunity to consume finance is self-evident, but currently the main body of many markets participates in it, the greatest challenge being the risk of multi-headed debt. The same customer is faced with more and more easy loan channels, and when each institution gives credit, both its funding chain and risk resistance become very fragile. The risk management capabilities of the financial institution determine the customer's choice and trust, and also how to conduct risk pricing and customer location, as well as risks and challenges that may be encountered over long periods of time.
Therefore, based on relevant information of the clients, the degree of liability of the clients is quantified by using a machine learning algorithm, and an important basis is provided for how reasonable the bank trusts so as to maintain data security and normal operation of a trust system. Currently, the evaluation of the degree of co-debt of customers in financial industries such as banking industry is still in a more traditional mode, but the process usually only combines external data information of customers (such as loan information of customers on other financial platforms), but the process often loses the information of customers, and models with risks as targets, and finally, the output result is discrete (0, 1), which is not very accurate, but the inaccurate result can cause the data management system for granting the user request to fail or run abnormally due to the user request (such as loan request), so a new more accurate and reasonable technical scheme is needed to evaluate the degree of co-debt of customers more accurately to avoid the situation that the data management system fails due to unreasonable user request.
Disclosure of Invention
The embodiment of the invention provides a user request determining method, a storage medium and an electronic device, which at least solve the problem that the data management system is failed due to unreasonable user request in the related art.
According to an embodiment of the present invention, there is provided a method for determining a user request, including:
obtaining a debt sharing level score of the stock user based on the user characteristics of the stock user and a stock debt sharing model algorithm under the condition that the user of which the debt sharing level is to be determined is the stock user, wherein the stock user is the user after a preset expression period;
Under the condition that a user of which the debt sharing level is to be determined is an incremental user, obtaining a debt sharing level score of the incremental user based on the user characteristics of the incremental user, the debt sharing model algorithm of the stock user and the debt sharing model algorithm of the increment, wherein the debt sharing model algorithm of the stock is used for adjusting the weight of the user characteristics of the incremental user, and the incremental user is a newly added user which does not pass through a preset expression period;
a security level of a user is determined based on the determined co-debt level score of the user and a determination is made as to whether to pass a user request of the user based on the security level.
According to a further embodiment of the invention, there is also provided a computer-readable storage medium having stored therein a computer program, wherein the computer program is arranged to, when run, implement the steps of any of the method embodiments described above.
According to a further embodiment of the invention, there is also provided an electronic device comprising a memory having stored therein a computer program and a processor arranged to run the computer program to carry out the steps of any of the method embodiments described above.
According to the embodiment of the invention, the security level of the user is determined based on the determined co-debt level score of the user, and whether the user request of the user is passed or not is determined based on the security level, so that the situation that the data management system is failed due to unreasonable passing of the user request is avoided.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the application and do not constitute a limitation on the application. In the drawings:
FIG. 1 is a flow chart of a method of determining a user request according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a model building process according to an example embodiment of the invention;
fig. 3 is a schematic diagram of weight adjustment according to an example embodiment of the invention.
Detailed Description
The application will be described in detail hereinafter with reference to the drawings in conjunction with embodiments. It should be noted that, without conflict, the embodiments of the present application and features of the embodiments may be combined with each other.
It should be noted that the terms "first," "second," and the like in the description and the claims of the present invention and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order.
Example 1
The method according to the first embodiment of the present application may be performed in a mobile terminal, a computer terminal, a server, or a similar computing device. These devices have some structures known at present and are not described again.
In this embodiment, a method for determining a user request running on the computing device is provided, fig. 1 is a flowchart of a method for determining a user request according to an embodiment of the present invention, and as shown in fig. 1, the flowchart includes the following steps:
Step S101, under the condition that a user of which the debt sharing level is to be determined is an inventory user, obtaining a debt sharing level score of the inventory user based on the user characteristics of the inventory user and an inventory debt sharing model algorithm, wherein the inventory user is a user after a preset expression period;
Under the condition that the user of the co-debt level to be determined is an increment user, obtaining a co-debt level score of the increment user based on the user characteristics of the increment user, the co-debt level score evaluation result of the stock co-debt model algorithm and the increment co-debt model algorithm, wherein the stock co-debt model algorithm is used for adjusting the weight of the user characteristics of the increment user, and the increment user is a new increment user which does not pass through a preset expression period;
step S103, determining a security level of the user based on the determined co-debt level score of the user, and determining whether to pass the user request of the user based on the security level.
According to the embodiment of the invention, the security level of the user is determined based on the determined co-debt level score of the user, and whether the user request of the user is passed or not is determined based on the security level, so that the situation that the data management system is failed due to unreasonable passing of the user request is avoided. Wherein the security level may be determined according to a co-debt level score, and if the co-debt level score indicates that the user has a higher co-debt level and a lower repayment capability, the security level may be set to indicate that the user request of the user is not passed or that the user request of the user is cautiously passed, or an alarm may be given based on the user request. Vice versa, if the co-debt level score indicates that the user has a lower co-debt level and a higher repayment capacity, the security level may be set to indicate a user request by the user or a user request more loosely by the user, and a prompt may be made, for example prompting the user request.
In one example, the obtaining the credit level score for the stock user based on the user characteristics of the stock user and the stock credit model algorithm includes:
Acquiring user characteristics of the stock users, and preprocessing the user characteristics of the stock users, wherein the user characteristics of each stock user correspond to a preprocessing mode;
vector standardization is carried out based on the user characteristics of the stock users after pretreatment, and a standardized matrix is obtained;
a co-debt level score for the stock user is determined based on the standardized matrix.
In one example, the determining the co-debt level score for the stock user based on the normalization matrix includes:
Determining the distance between the stock user and the optimal scheme and the distance between the stock user and the worst scheme by using a good-bad solution distance method according to the standardized matrix;
A co-debt level score for the stock user is determined based on the distance.
In one example, determining the distance of the stock user from the optimal solution and the worst solution using a best solution distance method according to the standardized matrix includes:
The distance between the stock user and the optimal scheme and the worst scheme is determined according to the following formula:
Where i represents the ith sample client, j represents the jth index of the m indexes, Representing the distance between client i and the optimal solution,/>Representing the distance between the client i and the worst scheme, w j is the index weight,/>Is the value of the j index in the optimal scheme,/>Is the value of the j index in the worst scheme, and z ij is the value of the j index of client i;
Determining a co-debt level score for the stock user based on the distance, comprising:
determining the co-debt level score for the stock user by the following formula:
Wherein C i represents the co-debt score for customer i, Representing the distance between client i and the optimal solution,/>Representing the distance of client i from the worst scenario.
In one example, the index weight is determined by:
the entropy value of each index is calculated by the following formula:
where x ij represents the value of the j-th index of the i-th client, and e j represents the entropy value of the j-th index;
The index weight is obtained by the following formula:
Where e j represents the entropy of the jth index, and w j represents the index weight of the jth index.
In one example, the obtaining the debt sharing level score of the incremental user based on the user characteristics of the incremental user, the debt sharing model algorithm's debt sharing level score evaluation result of the incremental user, and the incremental debt sharing model algorithm includes:
using the credited level score of the stock user as a target variable for the incremental credited model algorithm;
adjusting training sample weights of the incremental co-debt model using the co-debt level score of the stock user;
the incremental co-debt model algorithm is used to derive a co-debt level score for the incremental user based on the user characteristics of the incremental user.
In one example, the method further comprises:
and verifying the debt sharing level score of the stock user obtained by the stock debt sharing model algorithm, and/or verifying the debt sharing level score of the increment user obtained by the increment debt sharing model algorithm.
In one example, the user characteristics of the stock user include: user attributes of the stock user and behavioral characteristics of the stock user; the user characteristics of the incremental user include: the user attributes of the incremental user.
Example embodiment
The embodiment of the present invention is further explained below in conjunction with a specific implementation scenario, and the method for quantifying the risk of the liability of the client (also referred to as a user) provided in this embodiment is specifically explained by taking a banking scenario as an example, and it should be noted that the method provided in this embodiment is not only applicable to the banking scenario, but also applicable to other financial institution scenarios, and will not be described in detail below. The embodiment of the invention provides a method for quantifying the risk of sharing bonds to customers by combining unsupervised learning and supervised learning, which is used for risk identification control of financial banks. The wind control scene of the bank comprises two links, namely a pre-loan link and a loan link, the method provided by the embodiment uses a supervision learning method in the pre-loan link, and uses an unsupervised learning method in the loan link, and the two links are combined to jointly quantify the co-debt level of the client. It should be noted that, the "client" is referred to as "user" in this document, and will not be described in detail below.
In one exemplary embodiment, an inventory co-debt model may be built for an inventory customer of a bank. The newly added clients of the bank can become stock clients after a preset expression period, the clients can have more expressions of the bank behaviors in the expression period, relevant characteristic variables are extracted based on the attributes of the clients and the bank behaviors, a comprehensive evaluation model is built through an unsupervised learning algorithm, and a common debt score is calculated for the common debt level of each client. The attribute of the client may be basic information of the user, such as gender, age, industry occupation, educational background, income, etc., and may also include information such as credit inquiry result (e.g. credit rating information) of the user.
In an exemplary embodiment, in the comprehensive evaluation process, expert experience can be introduced to correct the model result, a small sample is extracted, the expert at the front end of credit giving, approval and the like is manually judged according to experience, the judging result of the expert is compared with the result of the model, and the result of the stock co-debt model is verified. That is, the score result of the co-debt model provided in the present embodiment may be compared with the result of the manual determination for verification.
In one exemplary embodiment, an incremental co-debt model may be built for an newly added customer of the bank; for the incremental debt sharing model, an algorithm of supervised learning can be used, the result of the stock debt sharing model is provided for the incremental debt sharing model to train, after the client debt sharing score obtained by the comprehensive evaluation model is classified into two categories, the two categories are used as target variables of a pre-loan link, the sample weight of the incremental client model is adjusted based on the stock debt sharing score, and the incremental debt sharing model is trained by applying xgboost algorithm.
By using the technical scheme adopted by the invention, the identification capability of the risk of the co-debt clients can be greatly improved, wherein the method of combining unsupervised and supervised learning enables the quantitative result to be more objectively close to the actual co-debt level of the clients, thereby avoiding that a data management system passes through the request of users unreasonably.
The embodiment of the invention discloses a method for quantifying the risk of co-debt of clients by combining unsupervised learning and supervised learning, which can be applied to quantification of the level of co-debt of clients before, during and after credit. Defining a scoring threshold of the co-debt customers according to the distribution of the co-debt scores of the customers, for example, in the lending strategies such as quota adjustment, management and control, etc., the customers with high co-debt level are directly given refusal; conversely, a customer with a low level of co-debt is accepted.
In an exemplary embodiment, the modeling process is largely divided into the creation of an inventory liability model (also referred to as an "inventory customer liability model") and the creation of an incremental model (also referred to as an "incremental customer liability model"), as shown in fig. 2.
1) Relevant index features of the stock clients (such as the attribute information of the clients and the behavior features of the clients at the bank in the embodiment) are extracted, and particularly, the behavior performance (i.e., behavior features) of the clients at the bank is used for establishing feature variables of the stock co-debt model. The feature variables (i.e., the aforementioned "related index features") are preprocessed by methods such as the isotropic transform (which may also be referred to as the homodromous transform), the quantile transform, and the logarithmic transform to construct a normalized matrix. The attribute information of the client may be basic information of the user, such as gender, age, industry occupation, education background, income, etc., and may also include information of credit inquiry result (such as credit rating information) of the user, etc. The behavioral characteristics of the customer at the bank may be transaction information, borrowing information, deposit information, credit information, etc., but not limited thereto, and may also include other characteristic indicators suitable for analyzing the co-debt level of the customer. It should be noted that, the feature variable may be preprocessed by one algorithm, or may be preprocessed by a plurality of algorithms;
2) By an unsupervised learning method such as a good-bad solution distance method, an inventory liability model is built, gaps among customers, optimal schemes and worst schemes are considered, and an inventory customer liability degree score (which may also be called as an "inventory customer liability score") is obtained, wherein one customer corresponds to one liability degree score (which may also be called as a "liability score").
3) Extracting relevant attribute characteristics of the newly added clients, adjusting weights of training samples by using the inventory client debt score (also called as a 'debt degree score') obtained by the inventory client debt model, and classifying the inventory debt score obtained by the inventory debt model to obtain target variables of the newly added clients.
4) And establishing an incremental co-debt model for the co-debt quantification of the newly added customer based on the target variable and the training sample after weight adjustment by using xgboost algorithm.
The stock client feature process may use the following ideas:
Basic information, card information, credit inquiry data and the like of the client are extracted, namely, attribute information of the client is extracted. Because the scales and meanings of the various indexes of the clients are different, when the comprehensive evaluation method is used, the distance scale is used for quantifying the difference between the clients. The index attribute is transformed in scale using the distance scale.
Some of the data may use a homodromous transformation that applies to data where the index value itself varies in opposite directions from the co-debt level, e.g., a higher revenue index value indicates a higher ability of the customer to pay back liabilities, further indicating that the co-debt level of the customer may be less. In contrast, since the smaller the liability index value is, which indicates that the liability of the client is, in order to unify the meaning of the index, we use the following process for the anisotropic index such as income:
Where x represents pre-transformation data (e.g., revenue values) and x' represents post-processing data (e.g., post-processing revenue values).
In addition, some information can be processed by using quantile transformation, for example, the data distribution of a part of indexes presents long tail characteristics, in a model which uses distance as a core calculation, such as a comprehensive evaluation method, the distance scale is greatly influenced by extremum, the extremum of a part of clients can cause the model to be unstable, in order to increase the robustness of the model, the influence of abnormal values on the model is reduced, the index of the data presenting obvious long tail distribution is quantile transformed, for example, 95 quantile is adopted as the maximum value, abnormal data (the value larger than 95% score is assigned as 95% quantile of the variable) is removed:
x '=min (x, x 0.95), where x represents the data before transformation and x' represents the data after transformation.
In addition, some information can be processed by using logarithmic transformation, for example, the data distribution range of different indexes and the difference thereof, for example, the sizes of loan amount and loan amount are different, logarithmic transformation is performed on the indexes of amount and balance (such as various types of loan amount and balance of customers), and the transformation can reduce the influence of the difference of sizes on the whole data to a certain extent:
x '=log (1+x), where x represents data before transformation and x' represents data after transformation.
Then, based on these transformed data, a normalized matrix is constructed, and specific methods are exemplified as follows:
Let us say that n customers are total, each customer has k indexes (i.e. the above-mentioned extracted basic information of the customer (including attribute information of the customer and behavior characteristics of the customer in the bank) is processed to obtain data, one data is equivalent to an index), then the original customer index matrix is:
wherein x n1 represents the value of the 1 st index of the nth client;
the index is vector normalized, i.e. each column element divided by the L2 norm of the current column vector,
X ij represents the value of the j index of the i-th client,/>The L2 norm representing the jth index, z ij represents the value of the jth index of the ith client after vector normalization transformation.
Thereby obtaining a standardized matrix Z after standardization:
Wherein, each client corresponds to a row in the matrix, for example, the first row is each index value of the first client, the nth row is each index value of the nth client, the row corresponds to the number of clients, and the column corresponds to the number of indexes.
In an exemplary embodiment, further, based on the standardized matrix Z obtained above, a distance between a certain customer and an optimal solution or a worst solution is obtained by using a best solution distance method. The best and inferior solution is also called TOPSIS, and is a comprehensive evaluation method, which can make full use of information of original data and accurately reflect differences among evaluation schemes. The basic process is based on the standardized homodromous feature matrix, the optimal scheme and the worst scheme in the limited scheme are found out by adopting a cosine method, then the distances between each evaluation object and the optimal scheme and the distances between each evaluation object and the worst scheme are calculated respectively, and the relative proximity degree between each evaluation object and the optimal scheme is obtained, so that the relative proximity degree is used as the basis of the evaluation quality, and the specific algorithm flow is as follows:
the stock co-debt model algorithm flow:
procedure co-debt model:
carrying out homodromous transformation on the original index to X1;
Scaling X1 to X2 within 0 to 95% quantiles;
carrying out logarithmic transformation on X2 to obtain X3;
Constructing a normalized matrix Z= { Z1, Z2, … …, zn } after vector normalization;
each column Zi, do For Z;
the ith dimension of the worst scheme Z-, the minimum of Zi elements;
the ith dimension of the worst scheme Z+ and the maximum value of Zi elements;
End for
For zi∈Z do
proximity of Zi to optimal solution
Proximity of Zi to optimal solution
Score of Zi
End for
Sorting according to the size of C i;
End procedure；
and outputting the comprehensive evaluation result of each customer.
The standardized feature matrix has been described above as to how to determine the optimal solution, which is composed of the maximum values of each column of elements in Z, which means the ideal case where the degree of co-debt is the lowest, which does not refer to a particular customer, but the ideal case where the degree of co-debt is the lowest.
The worst scenario is composed of the minimum value of each column element in Z, which means the ideal case with the highest degree of co-debt, and as with the best scenario, this case does not refer to a specific customer, but the ideal case with the highest degree of co-debt.
Based on index variables and weights of clients, calculating the distance between a certain client and an optimal scheme and the distance between the client and a worst scheme:
Where i represents the ith sample client, j represents the jth index of the m indexes, Representing the distance between client i and the optimal solution,/>Representing the distance between the client i and the worst scheme, w j is the weight of the j index determined by the methods such as entropy weight method,Is the value of the j index in the optimal scheme,/>Is the value of the j index in the worst scheme, and z ij is the value of the j index of client i;
And finally, calculating the comprehensive co-debt degree of the customer:
C i represents the co-debt score of customer i,/> Representing the distance between client i and the optimal solution,/>Representing the distance of client i from the worst scenario.
C i measures the degree of co-debt of the clients i, and by the comprehensive evaluation method, we quantify the abstract degree of co-debt, and the quantification standard is not based on a certain fixed standard, but fully utilizes various behavior characteristic information of the clients, and the result can accurately reflect the relative level of co-debt among the clients.
In one exemplary embodiment, determining the index weight based on the entropy weight method refers to: according to the basic principle of information theory, information is a measure of the order of the system; while entropy is a measure of the degree of disorder of the system. The entropy weight method is an objective weighting method, in the specific use process, the entropy weight method calculates the entropy weight of each index by utilizing information entropy according to the variation degree of each index, and then the weight of each index is corrected by the entropy weight, so that objective index weight is obtained.
Calculating entropy values of the indexes:
x ij represents the value of the j-th index of the i-th client, and e j represents the entropy value of the j-th index.
Further, we normalize the entropy value, taking this as the weight of each index:
e j denotes the entropy value of the j-th index, and w j denotes the entropy value of the j-th index after normalization.
The larger w j, the larger the amount of information represented by the index, the more discriminative the degree of co-debt.
In one exemplary embodiment, expert experience may be used to verify the effectiveness of the supervised learning approach. In order to verify the effectiveness of the supervised learning algorithm, an expert experience judging method is introduced, and the result of the comprehensive evaluation model is corrected. And (3) manually judging the extracted data according to experience by an expert at the front end of credit giving, approval and the like, for example, labeling 1050 cases by the expert, wherein each case adopts a two-to-two crossing judgment mode. For cases with large individual differences, a re-determination method is adopted.
For a newly added customer, we build an incremental co-debt model to evaluate the customer's co-debt level. Firstly, extracting relevant index features of a newly added client (including attribute information of the client), such as basic information (such as gender, age, industry occupation, education background, income and the like), and data such as a pedestrian credit inquiry result and the like, and establishing a corresponding data set.
Using xgboost as the incremental co-debt model, xgboost is an optimized distributed gradient incremental learning framework that implements machine learning algorithms under the concept of Gradient Boosting, providing parallel tree promotion, which can quickly and accurately solve many data science problems. The method designs and constructs a highly extensible end-to-end lifting tree system, provides a theoretically reasonable weighted quantile sketch to calculate a candidate set, introduces a novel sparse sensing algorithm for parallel tree learning, and enables a default direction of a missing value.
Xgboost is a linear addition model that sums the results of the K trees as the final predictor, namely:
The first term to the right of the above equal sign is the loss function, where n is the number of training function samples, l is the loss for a single sample, assuming it is a convex function, y i is the true label value of the modulo training sample, For model-to-training sample predictions, f t(xi) represents the model of the ith customer, the nth tree.
The second term of the above equation Ω (f t) is a regular term that controls the complexity of the tree, preventing overfitting. Regularization terms define the complexity of the model:
wherein, gamma and lambda are regularization parameters set manually, omega is a vector formed by all leaf node values of the decision tree, and T is the number of leaf nodes.
Combining the stock and the incremental model, combining expert experience, cutting model scores of the stock debt sharing model, judging whether the client is a debt sharing client or not, and further taking the client as a target variable of the incremental debt sharing model, so that the incremental model of the newly added client can learn the characteristics of the stock client with a large number of behaviours, and the predictability of the incremental debt sharing model is improved.
In this embodiment, the target variable of the model is an index that the model wants to predict or evaluate, and in this embodiment, it is: whether the customer is a co-debt customer.
The model is trained by using input target variables, namely, a model training stage, a model using stage, a model input stage and an model output result. For example, during the training phase: the new clients become stock clients after a certain expression period, and characteristic indexes of the clients in the stock stage are selected as inputs of the stock co-debt model; and selecting characteristic indexes of the clients in the application stage as input of a newly added co-debt model. In the model use stage: the characteristic index of the new customer at the time of application is used as the input of the newly added co-debt model. In this embodiment, not only the model score of the stock debt model is used as the target variable of the incremental debt model, but also the training sample weight of the incremental debt model is adjusted by this model score, as shown in fig. 3:
for customers with stock co-debt scores greater than median, their sample weights x:
x= (d-d 0.5)/(1-d0.5), d is the inventory liability score for the sample, and d 0.5 is the median of the overall incremental liability model sample inventory liability scores.
For customers whose stock co-debt score is less than the median, their sample weights:
x= (d-d 0.5)/d0.5, d is the inventory liability score for the sample, d 0.5 is the median of the overall incremental liability model sample inventory liability scores.
Because the purpose of modeling is to identify customers with extremely high and low co-debt levels, the sample weights are adjusted by the stock co-debt score. The inventory is shared with customers with high and low scores and the sample weights are greater. This will give samples with stock debt scores more biased from the median or average, and thus will allow the xgboost algorithm to more easily identify very high and very low debt level customers.
In this embodiment, an incremental debt sharing model is built by using xgboost algorithm, and the debt sharing level of the newly added customer is predicted, so that an unsupervised learning method is used for the stock customer to build the stock debt sharing model, a supervised learning method is used for the newly added customer to build the incremental debt sharing model, and supervision and unsupervised are combined to build an evaluation system for different stages of the whole life process of the customer, and the robustness of the overall debt sharing level evaluation can be improved more.
From the description of the above embodiments, it will be clear to a person skilled in the art that the method according to the above embodiments may be implemented by means of software plus the necessary general hardware platform, but of course also by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) comprising instructions for causing a terminal device (which may be a mobile phone, a computer, a server, or a network device, etc.) to perform the method according to the embodiments of the present invention.
Embodiments of the present invention also provide a computer readable storage medium having a computer program stored therein, wherein the computer program is arranged to implement the steps of any of the method embodiments described above when run. Alternatively, specific examples in this embodiment may refer to examples described in the foregoing embodiments and optional implementations, and this embodiment is not described herein.
Alternatively, in the present embodiment, the storage medium may include, but is not limited to: a usb disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory RAM), a removable hard disk, a magnetic disk, or an optical disk, or other various media capable of storing a computer program.
An embodiment of the invention also provides an electronic device comprising a memory having stored therein a computer program and a processor arranged to run the computer program to implement the steps of any of the method embodiments described above.
Optionally, the electronic apparatus may further include a transmission device and an input/output device, where the transmission device is connected to the processor, and the input/output device is connected to the processor.
Alternatively, specific examples in this embodiment may refer to examples described in the foregoing embodiments and optional implementations, and this embodiment is not described herein.
It will be appreciated by those skilled in the art that the modules or steps of the invention described above may be implemented in a general purpose computing device, they may be concentrated on a single computing device, or distributed across a network of computing devices, they may alternatively be implemented in program code executable by computing devices, so that they may be stored in a memory device for execution by computing devices, and in some cases, the steps shown or described may be performed in a different order than that shown or described, or they may be separately fabricated into individual integrated circuit modules, or multiple modules or steps within them may be fabricated into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the principle of the present invention should be included in the protection scope of the present invention.
Claims (9)
1. A method for determining a user request, comprising:
obtaining a debt sharing level score of the stock user based on the user characteristics of the stock user and a stock debt sharing model algorithm under the condition that the user of which the debt sharing level is to be determined is the stock user, wherein the stock user is the user after a preset expression period;
obtaining a debt sharing level score of the increment user based on the user characteristics of the increment user, the debt sharing level score of the stock debt sharing model algorithm and the increment debt sharing model algorithm when the user of which the debt sharing level is to be determined is the increment user, wherein the debt sharing level score of the stock user is used as a target variable of the increment debt sharing model algorithm; adjusting training sample weights of the incremental co-debt model using the co-debt level score of the stock user; applying xgboost algorithm to establish an incremental co-debt model based on the target variable and the training sample after weight adjustment; obtaining a debt sharing level score of the incremental user by using the incremental debt sharing model algorithm based on the user characteristics of the incremental user, wherein the incremental user is a new user which does not pass through a preset expression period;
a security level of a user is determined based on the determined co-debt level score of the user and a determination is made as to whether to pass a user request of the user based on the security level.
2. The method of determining a user request according to claim 1, the obtaining a debt level score for the stock user based on user characteristics of the stock user and a stock debt model algorithm, comprising:
Acquiring user characteristics of the stock users, and preprocessing the user characteristics of the stock users, wherein the user characteristics of each stock user correspond to a preprocessing mode;
vector standardization is carried out based on the user characteristics of the stock users after pretreatment, and a standardized matrix is obtained;
and determining the debt sharing level score of the stock user according to the standardized matrix.
3. The method of determining a user request according to claim 2, wherein said determining a co-debt level score for the stock user from the normalization matrix comprises:
Determining the distance between the stock users and the optimal scheme and the worst scheme by using a good-bad solution distance method according to the standardized matrix;
and determining the co-debt level score of the stock user according to the distance.
4. A method for determining a user request according to claim 3, wherein determining the distance between the stock user and the optimal solution and the worst solution using a best solution distance method according to the standardized matrix comprises:
and determining the distance between the stock users and the optimal scheme and the worst scheme according to the following formula:
Where i represents the ith sample client, j represents the jth index of the m indexes, Representing the distance between client i and the optimal solution,/>Representing the distance between the client i and the worst scheme, w j is the index weight,/>Is the value of the j index in the optimal scheme,Is the value of the j index in the worst scheme, and z ij is the value of the j index of client i;
determining a co-debt level score for the stock user based on the distance, comprising:
Determining a co-debt level score for the stock user by:
Wherein C i represents the co-debt score for customer i, Representing the distance between client i and the optimal solution,/>Representing the distance of client i from the worst scenario.
5. The method of determining a user request according to claim 4, wherein the index weight is determined by:
the entropy value of each index is calculated by the following formula:
Where x ij represents the value of the j-th index of the i-th client, and e j represents the entropy value of the j-th index;
The index weight is obtained by the following formula:
Where e j represents the entropy of the jth index, and w j represents the index weight of the jth index.
6. The method of determining a user request according to claim 1, further comprising:
and verifying the debt sharing level score of the stock user obtained by the stock debt sharing model algorithm, and/or verifying the debt sharing level score of the increment user obtained by the increment debt sharing model algorithm.
7. The method of claim 1, wherein the user characteristics of the stock user include: user attributes of the stock users and behavior characteristics of the stock users; the user characteristics of the incremental user include: and the user attribute of the increment user.
8. A computer-readable storage medium, characterized in that the storage medium has stored therein a computer program, wherein the computer program is arranged to implement the method of any of the claims 1 to 7 when run.
9. An electronic device comprising a memory and a processor, characterized in that the memory has stored therein a computer program, the processor being arranged to run the computer program to implement the method of any of the claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110336662.6A CN112862602B (en) | 2021-03-29 | 2021-03-29 | User request determining method, storage medium and electronic device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110336662.6A CN112862602B (en) | 2021-03-29 | 2021-03-29 | User request determining method, storage medium and electronic device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112862602A CN112862602A (en) | 2021-05-28 |
| CN112862602B true CN112862602B (en) | 2024-05-24 |
Family
ID=75993132
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110336662.6A Active CN112862602B (en) | 2021-03-29 | 2021-03-29 | User request determining method, storage medium and electronic device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112862602B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108648074A (en) * | 2018-05-18 | 2018-10-12 | 深圳壹账通智能科技有限公司 | Loan valuation method, apparatus based on support vector machines and equipment |
| CN109727123A (en) * | 2019-01-04 | 2019-05-07 | 平安科技（深圳）有限公司 | User's collage-credit data construction method, device and computer readable storage medium |
| CN110135972A (en) * | 2019-04-23 | 2019-08-16 | 上海淇玥信息技术有限公司 | A kind of method, apparatus, system and recording medium for improving user and moving branch rate |
| CN110728301A (en) * | 2019-09-09 | 2020-01-24 | 北京镭文科技有限公司 | Credit scoring method, device, terminal and storage medium for individual user |
-
2021
- 2021-03-29 CN CN202110336662.6A patent/CN112862602B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108648074A (en) * | 2018-05-18 | 2018-10-12 | 深圳壹账通智能科技有限公司 | Loan valuation method, apparatus based on support vector machines and equipment |
| CN109727123A (en) * | 2019-01-04 | 2019-05-07 | 平安科技（深圳）有限公司 | User's collage-credit data construction method, device and computer readable storage medium |
| CN110135972A (en) * | 2019-04-23 | 2019-08-16 | 上海淇玥信息技术有限公司 | A kind of method, apparatus, system and recording medium for improving user and moving branch rate |
| CN110728301A (en) * | 2019-09-09 | 2020-01-24 | 北京镭文科技有限公司 | Credit scoring method, device, terminal and storage medium for individual user |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112862602A (en) | 2021-05-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kozodoi et al. | Fairness in credit scoring: Assessment, implementation and profit implications | |
| US10572885B1 (en) | Training method, apparatus for loan fraud detection model and computer device | |
| Caruso et al. | Cluster Analysis for mixed data: An application to credit risk evaluation | |
| Verbraken et al. | Development and application of consumer credit scoring models using profit-based classification measures | |
| US20180260891A1 (en) | Systems and methods for generating and using optimized ensemble models | |
| Ala’raj et al. | Modelling customers credit card behaviour using bidirectional LSTM neural networks | |
| Callejón et al. | A System of Insolvency Prediction for industrial companies using a financial alternative model with neural networks | |
| CN112102073A (en) | Credit risk control method and system, electronic device and readable storage medium | |
| CN110852881B (en) | Risk account identification method and device, electronic equipment and medium | |
| US20190130481A1 (en) | Entity Segmentation for Analysis of Sensitivities to Potential Disruptions | |
| US20190385100A1 (en) | System And Method For Predicting Organizational Outcomes | |
| Eddy et al. | Credit scoring models: Techniques and issues | |
| CN112836750A (en) | System resource allocation method, device and equipment | |
| Bartoloni et al. | Financial performance in manufacturing firms: a comparison between parametric and non-parametric approaches | |
| CN116468273A (en) | Customer risk identification method and device | |
| CN116596659A (en) | Enterprise intelligent credit approval method, system and medium based on big data wind control | |
| CN114881485A (en) | Enterprise fund risk assessment method based on analytic hierarchy process and cloud model | |
| CN113344692A (en) | Method for establishing network loan credit risk assessment model with multi-information-source fusion | |
| CN117575773A (en) | Method, device, computer equipment and storage medium for determining service data | |
| Fu et al. | Applying DEA–BPN to enhance the explanatory power of performance measurement | |
| CN112508684A (en) | Joint convolutional neural network-based collection risk rating method and system | |
| CN112862602B (en) | User request determining method, storage medium and electronic device | |
| CN116128339A (en) | Client credit evaluation method and device, storage medium and electronic equipment | |
| CN116091200A (en) | Scene credit granting system and method based on machine learning, electronic equipment and medium | |
| CN116912016A (en) | Bill auditing method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |