Loan prediction method and device, equipment and storage medium

Abstract

The invention provides a loan prediction method and device, equipment and a storage medium, and the method comprises the steps: obtaining operator data of a target user from an operator platform, and the operator data comprises internet log data, user call data and user position data; respectively inputting the operator data into a loan willingness model and a user qualification model, and obtaining a loan probability output by the loan willingness model and a qualification probability output by the user qualification model; and according to the loan probability and the qualification probability, recommending a loan product to the user, and predicting the loan willingness of the user from multiple dimensions, so that different types of loan products are matched with different users.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a loan prediction method, device, equipment and storage medium. Background Art

[0002] At present, banks and other financial institutions are developing personal loan business. In addition to relying on offline marketing by account managers and seeking third-party financial intermediaries, they are also actively trying to place Internet advertisements, send text messages, and conduct telephone marketing to users. Some financial institutions often use factor scoring methods when screening users with loan intentions. First, the characteristic variables that affect the customer's willingness to borrow and the weights of different characteristic variables are determined, the probability of user loans is estimated, and finally the feasibility threshold is set.

[0003] This type of method requires manual determination of the scope and weight of factors, is highly subjective, and cannot guarantee the accuracy of predictions. In addition, some financial institutions have introduced machine learning or neural network models to predict potential loan populations. Financial institutions only model based on positive and negative samples of their own existing users and the characteristics of existing users of the bank to reduce manual intervention and improve prediction accuracy.

[0004] However, this method only judges the user's willingness to borrow based on a single linear indicator, and does not match different users with different types of loan products. In addition, financial institutions only make predictions based on their own existing data, and the data dimension is single, making it impossible to predict customers' willingness to borrow from multiple dimensions. Summary of the invention

[0005] The present application provides a loan prediction method, device, equipment and storage medium to solve the problems in the prior art of judging the user's loan willingness based on only a single linear indicator, failing to match different users with different types of loan products, having a single data dimension, being unable to predict the customer's loan willingness from multiple dimensions and having poor accuracy.

[0006] In a first aspect, the present application provides a loan prediction method, the method comprising:

[0007] Acquire user sample data, the user sample data including the user's operator data and a first loan tag and a second loan tag, the first loan tag is used to mark whether the user accepts the loan, and the second loan tag is used to mark whether the user is successfully granted credit after accepting the loan;

[0008] Training a loan willingness model according to the operator data of the user and the first loan label to obtain a trained loan willingness model;

[0009] The user qualification model is trained according to the operator data of the user and the second loan label to obtain a trained user qualification model.

[0010] Optionally, obtaining user sample data includes:

[0011] Obtain user credit data from financial institutions, and screen target users who have loan intentions from the user credit data, wherein the credit data is credit data authorized by both the user and the financial institution;

[0012] Acquire the operator data of the target user from the operator platform, wherein the operator data includes Internet access log data, user call data and user location data;

[0013] The user sample data is obtained according to the operator data and credit data of the target user, wherein the credit data includes whether the target user accepts the loan and whether the credit is successfully granted after accepting the loan.

[0014] Optionally, the acquiring the operator data of the target user from the operator platform includes:

[0015] Obtain the target time for successful credit granting of the target user provided by the financial institution;

[0016] The operator data within a preset time period before the target time is obtained from the operator platform, wherein the preset time periods corresponding to the Internet log data, the user call data and the user location data are different.

[0017] Optionally, obtaining the user sample data according to the operator data of the target user and whether the target user accepts the loan and whether the credit is successfully granted after accepting the loan includes:

[0018] Determine whether the target user's operator data and credit data are abnormal or missing;

[0019] If so, the K nearest neighbor method is used to determine the K samples closest to the missing or abnormal data sample, and the missing or abnormal data is filled according to the weighted average of the K samples to obtain the user sample data.

[0020] Optionally, the loan willingness model and user qualification model are trained by an XGBoost analysis method.

[0021] In a second aspect, the present application provides a loan prediction method, comprising:

[0022] Acquire the operator data of the target user from the operator platform, wherein the operator data includes Internet access log data, user call data and user location data;

[0023] Inputting the operator data into a loan willingness model and a user qualification model respectively, wherein the loan willingness model and the user qualification model are obtained by training using the method described above;

[0024] Obtaining the loan probability output by the loan willingness model and the qualification probability output by the user qualification model;

[0025] Recommend loan products to the user based on the loan probability and the qualification probability.

[0026] Optionally, the recommending a loan product to the user according to the loan probability and the qualification probability includes:

[0027] With respect to the loan probability, obtaining the first ranking of the target user in the target group;

[0028] Based on the qualification probability, obtaining the second ranking of the target user in the target group;

[0029] Recommending loan products to the user based on the first ranking and the second ranking.

[0030] In a third aspect, the present application provides a loan prediction device, the device comprising:

[0031] An acquisition module, which acquires user sample data, wherein the user sample data includes the user's operator data and a first loan tag and a second loan tag, wherein the first loan tag is used to mark whether the user accepts the loan, and the second loan tag is used to mark whether the user is successfully granted credit after accepting the loan;

[0032] a first training module, training a loan willingness model according to the operator data of the user and the first loan label to obtain a trained loan willingness model;

[0033] The second training module trains the user qualification model according to the operator data of the user and the second loan label to obtain a trained user qualification model.

[0034] Optionally, the acquisition module is specifically used for:

[0035] Obtaining user credit data authorized by both the user and the financial institution from the financial institution, and screening out target users with loan intentions from the user credit data;

[0036] Acquire the operator data of the target user from the operator platform, wherein the operator data includes Internet access log data, user call data and user location data;

[0037] The user sample data is obtained according to the operator data and credit data of the target user, wherein the credit data includes whether the target user accepts the loan and whether the credit is successfully granted after accepting the loan.

[0038] Optionally, the acquisition module is further used to acquire a target time for successful credit granting to the target user;

[0039] The acquisition module is further used to acquire operator data within a preset time period before the target time from the operator platform, wherein the preset time periods corresponding to the Internet log data, the user call data and the user location data are different.

[0040] Optionally, the loan prediction device further includes: a processing module;

[0041] The processing module is used to determine whether the operator data and credit data of the target user are abnormal or missing;

[0042] When there are abnormalities or missing in the operator data and credit data of the target user, the K nearest neighbor method is used to determine the K samples closest to the missing or abnormal data samples, and the missing or abnormal data is filled based on the weighted average of the K samples to obtain the user sample data.

[0043] In a fourth aspect, the present application provides a loan prediction device, characterized in that it includes:

[0044] Acquisition module: Acquisition of operator data of target users from the operator platform, wherein the operator data includes Internet access log data, user call data and user location data;

[0045] Processing module: inputting the operator data into a loan willingness model and a user qualification model respectively, wherein the loan willingness model and the user qualification model are trained by the method described in any one of the above items;

[0046] Prediction module: obtaining the loan probability output by the loan willingness model and the qualification probability output by the user qualification model;

[0047] Recommendation module: recommends loan products to users based on the loan probability and the qualification probability.

[0048] In a fifth aspect, the present application provides a loan prediction device, comprising:

[0049] Memory;

[0050] processor;

[0051] Wherein, the memory stores computer-executable instructions;

[0052] The processor executes the computer-executable instructions stored in the memory to implement the loan prediction method as described in the first aspect and various possible implementations of the first aspect.

[0053] In a sixth aspect, the present application provides a computer storage medium having computer execution instructions stored thereon, wherein the computer execution instructions are executed by a processor to implement the loan prediction method as described in the first aspect and various possible implementations of the first aspect.

[0054] The present application provides a loan prediction method, device, equipment and storage medium, the method comprising: obtaining the operator data of the target user from the operator platform, the operator data including online log data, user call data and user location data; inputting the operator data into the loan willingness model and the user qualification model respectively, obtaining the loan probability output by the loan willingness model and the qualification probability output by the user qualification model; recommending loan products to the user based on the loan probability and the qualification probability, predicting the user's loan willingness from multiple dimensions, so that different types of loan products match different users BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0056] Figure 1 This is the process of the loan prediction method provided by this application Figure 1 ;

[0057] Figure 2 This is the process of the loan prediction method provided by this application Figure 2 ;

[0058] Figure 3 This is the process of the loan prediction method provided by this application Figure 3 ;

[0059] Figure 4 This is a schematic diagram of the structure of the loan prediction device provided by this application. Figure 1 ;

[0060] Figure 5 This is a schematic diagram of the structure of the loan prediction device provided by this application. Figure 2 ;

[0061] Figure 6 It is a structural schematic diagram of the loan prediction device provided by this application.

[0062] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solutions and advantages of this application clearer, the technical solutions in this application will be clearly and completely described below in conjunction with the drawings in this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0064] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in sequences other than those illustrated or described herein, for example.

[0065] In the embodiments of the present application, the words "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0066] In addition to offline marketing, banks and other financial institutions are also actively trying to send relevant information to users through social media, the Internet, etc. to develop personal loan business. When screening users with loan intentions, the factor scoring method is often used. First, the characteristic variables that affect the customer's loan willingness and the weights of different characteristic variables are determined, and then the user's qualifications are evaluated, and finally the feasibility threshold is set.

[0067] The characteristic variables and weights that affect users' willingness to take out loans are often set manually and are highly subjective. On this basis, some financial institutions have introduced machine learning or neural network models to predict potential loan populations in order to reduce manual intervention and improve prediction accuracy.

[0068] However, the above method only judges the user's loan willingness based on a single linear indicator, and does not match different users with different types of loan products. In addition, financial institutions only make predictions based on their own existing data, and the data dimension is single. It is impossible to predict the customer's loan willingness from multiple dimensions and cannot guarantee the accuracy of the prediction.

[0069] The loan prediction method provided in this application obtains user credit data authorized by the user and the financial institution, including labels such as whether the user accepts the loan and whether the credit is successfully granted after the user accepts the loan, from the financial institution; and obtains user operator data from the operator platform, and the operator data includes: user Internet log data, user call data, and user location data for different preset time periods before the time when the financial institution gives the operator loan intention user and the time when the credit is successfully granted. User sample data is obtained based on the above-mentioned operator data and user credit data, and the user sample data is used to train the loan willingness model and the user qualification model. The trained loan willingness model is used to output the user's loan probability, and the trained user qualification model is used to output the user's qualification probability, and products are recommended to users in combination with the loan probability and qualification probability. This method improves the accuracy of loan prediction, matches different users with different types of loan products, and predicts the user's loan willingness from multiple dimensions, achieving a thousand-faced effect.

[0070] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0071] Figure 1 This is the process of the loan prediction method provided by this application Figure 1 This embodiment describes the model training method in the loan prediction method. Figure 1 As shown, the loan prediction method shown in this embodiment includes:

[0072] S101: Acquire user sample data, where the user sample data includes the user's operator data and a first loan tag and a second loan tag. The first loan tag is used to mark whether the user accepts the loan, and the second loan tag is used to mark whether the user is successfully granted credit after accepting the loan.

[0073] In the process of obtaining the user sample data, the operator data can be user feature labels of multiple dimensions such as Internet logs, user calls, user locations and user basic information, or it can be the operation results representing user behavior relationships obtained by substituting the above user feature labels into label expressions. For example, when the label expression Y = (A OR B) AND C, A, B, and C represent user feature labels of three dimensions: Internet log data, user call data and user location. Y is the behavioral association relationship of the user represented by the label expression in a certain time series. This application does not impose any restrictions on this.

[0074] The first loan tag and the second loan tag are obtained based on credit data, and the credit data includes: the user's loan acceptance and credit status. In order to improve the accuracy of loan prediction and match different users with different types of loan products, this application sets the first loan tag to mark whether the user accepts the loan, and sets the second loan tag to mark whether the user is successfully granted credit after accepting the loan, as user sample data representing the two dimensions of user loan willingness and user qualification during model training.

[0075] S102: Training a loan willingness model according to the operator data of the user and the first loan label to obtain a trained loan willingness model.

[0076] Among them, in order to be more in line with the loan business in the market and recommend loan products to users in a more detailed manner, the present invention divides loan products into bank loan products and consumer finance loan products, and specifically divides bank loan products into credit loan products and mortgage loan products. Correspondingly, the loan willingness model is also subdivided into a bank loan willingness sub-model and a consumer finance loan willingness sub-model according to the above-mentioned product types; the bank loan willingness sub-model includes: a credit loan willingness sub-model and a mortgage loan willingness sub-model.

[0077] The loan willingness model and its refined sub-models are trained according to the user's operator data and the first loan label in the sample data. Specifically, the sample data of users who have been marketed for loans and accepted the first loan label are used as positive sample data for model training; the sample data of users who have been marketed for loans but rejected the first loan label are used as negative sample data for model training; preferably, the positive samples and negative samples can be randomly sampled at a ratio of 10:1. The positive and negative sample data are input into the loan willingness model and its refined sub-models for training. During the training process, the recursive feature data is continuously screened, and the important feature data is re-input into the model for re-training. The overfitting degree of the model is reduced through cross-validation and other methods to achieve the training effect.

[0078] S103: Training a user qualification model according to the operator data of the user and the second loan tag to obtain a trained user qualification model.

[0079] Among them, according to the above-mentioned classification of loan product types in this application, the principle is the same as the division of sub-models of the loan willingness model. This application divides the credit qualification model into a bank loan qualification sub-model and a consumer finance loan qualification sub-model; the bank loan qualification sub-model includes: a credit loan qualification sub-model and a mortgage loan qualification sub-model.

[0080] The credit qualification model and its refined sub-models are trained according to the operator data of the user in the sample data and the second loan label. Specifically, in the user qualification model, users who accept the loan business and are successfully granted credit are used as positive samples, and users who accept the loan business but fail to be granted credit are used as negative samples. The division of positive and negative samples and the training process steps and principles are the same as S102, and will not be described in detail here.

[0081] The loan prediction method provided in this embodiment obtains user sample data, which includes the user's operator data and a first loan tag and a second loan tag. The first loan tag is used to mark whether the user accepts the loan, and the second loan tag is used to mark whether the user is successfully granted credit after accepting the loan; the loan willingness model is trained according to the user's operator data and the first loan tag to obtain a trained loan willingness model; the user qualification model is trained according to the user's operator data and the second loan tag to obtain a trained user qualification model, so that the trained loan willingness model can output the user's loan probability, and the trained user qualification model is used to output the user's qualification probability, and the product is recommended to the user in combination with the loan probability and the qualification probability. This method improves the accuracy of loan prediction, predicts the user's loan willingness from multiple dimensions, matches different users with different types of loan products, and achieves the effect of thousands of faces for thousands of people.

[0082] Figure 2 The process of the loan prediction method provided in the embodiment of the present application is Figure 2 This embodiment is in Figure 1 Based on the embodiment, the model training method in the loan prediction method is described in detail. Figure 2 As shown, the loan prediction method shown in this embodiment includes:

[0083] S201: Obtaining user credit data authorized by the user and the financial institution from the financial institution, and screening out target users with loan intentions from the user credit data

[0084] The user credit data may include, for example: the user's historical lending data, borrowing information, repayment information, default information, credit amount, etc. Target users with loan intentions are screened out from the user credit data, and the first loan tag is set to mark whether the user accepts the loan, and the second loan tag is set to mark whether the user is successfully granted credit after accepting the loan. The first loan tag and the second loan tag are used as user sample data representing the two dimensions of user loan willingness and user qualification in model training.

[0085] S202: Obtain the target time for successful credit authorization of the target user.

[0086] Among them, the target time for successful credit granting of the target user and the relevant credit granting data are information obtained with the authorization of the user and the financial institution. It can be understood that the user behavior and the result of successful credit granting of the user have a certain degree of logical correlation. For example, if the user's qualifications are poor, the number of times the user visits bank loan apps, online loan apps and other apps in the operator's data is likely to be higher than the user group average. In the above case, recommending loan products with lower user qualification requirements to the user may lead to successful credit granting of the user. Therefore, in the process of using the model to predict loans and recommend loan products to users, the target time for successful credit granting of the target user is obtained, and the operator data before the target time is selected from the operator platform as the user sample data based on the target time. This can enable the model to have a logical correlation between user behavior and the result of successful credit granting of the user, thereby improving the accuracy and reliability of the model's prediction of loans.

[0087] S203: Acquire operator data within a preset time period before the target time from the operator platform.

[0088] This application divides operator data into multiple data dimensions such as Internet log data, user call data, user location data, and user basic information.

[0089] With respect to the Internet log data dimension, this application breaks down user Internet behaviors such as APP, WeChat applets, WeChat public accounts, web browsing, etc. into multiple types of feature tags, such as shopping APP access, financial APP access, game APP access, travel APP access, social security / provident fund APP access, and other feature tags. In addition, since loans mainly involve the financial industry, the financial APP access tags are subdivided into secondary features, specifically including feature tags such as securities APP access, bank loan APP access, online loan APP access, and consumer finance loan APP access. With respect to the user call data dimension, the present invention lists feature tags such as the number of cities where users make calls, the number of domestic roaming Internet days, and the number of international long-distance calls. With respect to user basic information, the present invention lists feature tags such as gender and age. With respect to location information, the present invention lists feature tags such as place of work, place of residence, and frequency of visits to other places.

[0090] Furthermore, in the process of using the model to make loan predictions and recommend loan products to users, the present application obtains the time when the target user is successfully granted credit, and selects operator data within a preset time period before the target time from the operator platform as user sample data; the time and the sample data are information obtained with the authorization of the user and the financial institution. Because the Internet access log data, user call data and user location data have the characteristics of dynamic changes according to the time series, the preset time periods corresponding to the Internet access log data, user call data and user location data are different; the present application selects the time dimensions of the past seven days, the past month and the past six months as the preset time periods, and obtains the Internet access log data, user call data and user location data within different preset time periods respectively, and marks the Internet access log data, user call data and user location data obtained within different preset time periods with time series labels.

[0091] S204: Determine whether the operator data and credit data of the target user are abnormal or missing; if so, execute S206; if not, execute S205.

[0092] In order to improve the accuracy and reliability of the model prediction process, after obtaining the user's operator data and credit data, it is necessary to process the missing values ​​and outliers. The method for determining sample missing is that the sample data is incomplete within a certain range. For example, taking the Internet log data as an example, if the time series label of the user's Internet log data obtained from the operator platform is the data of the past six months, but there are actually only five months of Internet log data, this situation is determined as sample missing. The method for determining sample anomalies is the 3δ principle. In the 3δ principle, if the data follows a normal distribution: the probability of being 3δ away from the mean is P(|x-μ|>3δ)<=0.003, this is an extremely low probability event, and it is determined that the sample with a distance exceeding the mean of 3δ does not exist. Therefore, when the sample distance from the mean is greater than 3δ, the sample is considered an outlier.

[0093] S205: Using the K nearest neighbor method, determine the K samples closest to the missing or abnormal data sample, fill the missing or abnormal data according to the weighted average of the K samples, and obtain the user sample data.

[0094] Preferably, the present application uses the K nearest neighbor method to deal with missing and abnormal samples. The K nearest neighbor method can be understood as: given a training data set, for the missing or abnormal samples therein, find the K samples closest to the sample in the training data set, and use the K samples to represent the missing or abnormal samples. In this embodiment, the K nearest neighbor method is specifically: determine the K samples closest to the sample with missing samples or abnormal data according to the Euclidean distance, and estimate the actual data and abnormal data of the sample by weighted averaging these K values, fill in the missing or abnormal data, and finally obtain the user sample data.

[0095] S206: Use the operator data and credit data of the target user as user sample data.

[0096] Among them, if the operator data and credit data of the target user are not abnormal or missing, the operator data and credit data of the target user can be used as user sample data.

[0097] Specifically, the loan willingness model and the user qualification model are trained by the XGBoost analysis method.

[0098] The XGBoost (eXtremeGradient Boosting, XGBoost) analysis method is an efficient gradient boosting decision tree algorithm. The XGBoost model is a boosting tree model. This application uses the XGBoost analysis method and framework to train the loan willingness model and the user qualification model, uses the user sample data as training parameters, and inputs the loan willingness model and the user qualification model using the XGBoost model as a framework for loan prediction. In this application, according to the type of loan product, the loan willingness model is divided into a bank loan willingness sub-model and a consumer finance loan willingness sub-model; the bank loan willingness sub-model includes: a credit loan willingness sub-model and a mortgage loan willingness sub-model. The credit qualification model is divided into a bank loan qualification sub-model and a consumer finance loan qualification sub-model; the bank loan qualification sub-model includes: a credit loan qualification sub-model and a mortgage loan qualification sub-model. According to the historical marketing loan situation of the loan product, in the loan willingness model, users who have been marketed with loan products and successfully accepted them are used as positive samples, and users who have been marketed with loan products but rejected them are used as negative samples; in the user qualification model, users who have accepted loan business and have been successfully granted credit are used as positive samples, and users who have accepted loan business but failed to be granted credit are used as negative samples, wherein the positive and negative samples can be divided in a ratio of 10:1; the model is trained using the training-testing-training validation set method, wherein the division ratio of the training set, the test set, and the training validation set can be, for example, 6:2:2.

[0099] Furthermore, during the model training process, in order to achieve the model training effect, the feature labels in the user sample data are continuously recursively screened during the training process, and the important feature data are re-input into the model for training again. The recursive screening method can be, for example: during the training of the loan willingness model and the user qualification model, the feature labels in the operator data are sorted from high to low according to the model contribution (importance) score, and a suitable feature selection threshold is selected, and the elements with higher importance are retained for model training. Finally, the top ten feature labels ranked by importance are recursively screened for model training. In addition, during the model training process, the model is evaluated using indicators such as classification threshold analysis, learning curve, cross-validation and loss function value, and the training effect is achieved through continuous iterative optimization.

[0100] The model training method provided in this embodiment obtains user credit data authorized by the user and the financial institution from the financial institution, selects target users with loan intentions from the user credit data, takes the target time for the target user to be successfully granted credit, further obtains operator data within a preset time period before the target time from the operator platform, uses the operator data and credit data of the target user as user sample data, and trains the loan willingness model and user qualification model through the XGBoost analysis method. The trained loan willingness model, user qualification model and its sub-models can recommend loan products to different users, abandoning the subjective selection of user recommendations in the prior art, and have high accuracy.

[0101] Figure 3 The process of the loan prediction method provided in the embodiment of the present application is Figure 3 This embodiment describes a method for predicting loans based on a trained model and recommending products to users. Figure 3 As shown, the loan prediction method shown in this embodiment includes:

[0102] S301: Acquire operator data of a target user from an operator platform.

[0103] Among them, the target users are users who need loan predictions and loan products recommended to them. The operator data includes: user's Internet log data, user call data, user basic information, user location data, etc. The above-mentioned user's Internet log data, user call data dimensions, and user location data have different time dimensions, including: nearly seven days, nearly one month, and nearly six months dimensions.

[0104] S302: Inputting the operator data into the loan willingness model and the user qualification model respectively.

[0105] Among them, the operator data of users who need loan prediction and loan products recommended to them are input into the loan willingness model and the user qualification model respectively to obtain the loan probability output by the loan willingness model and the qualification probability output by the user qualification model.

[0106] S303: Acquire the loan probability output by the loan willingness model and the qualification probability output by the user qualification model.

[0107] The operator data is input into the loan willingness model and the user qualification model respectively. The loan willingness model and the user qualification model are obtained after the classification threshold analysis, learning curve, cross validation and loss function value and other indicators are evaluated to meet the training effect. The loan willingness model and the user qualification model will output the user's loan probability and user qualification probability respectively. The higher the user's loan willingness, the higher the loan probability output by the loan willingness model and its sub-model; the higher the user's credit qualification, the higher the qualification probability output by the user qualification model and its sub-model.

[0108] S304: With respect to the loan probability, obtain the first ranking of the target user in the target group.

[0109] Among them, according to the loan probability of the target user output by the loan willingness model, the target user is ranked in the target user group. Furthermore, because the present invention divides the loan willingness model into a bank loan willingness sub-model and a consumer finance loan willingness sub-model according to the type of loan products, and the bank loan willingness sub-model is divided into: a credit loan willingness sub-model and a mortgage loan willingness sub-model. It can be understood that the above sub-models will respectively output the target user's loan probability for bank loan products, consumer finance loan products, and credit loan products and mortgage loan products under bank loan products; in order to recommend loan products to different users, the loan probabilities output by the loan willingness model and its sub-models are ranked respectively, and the ranking is the first ranking. The first ranking can, for example, be a one-by-one sorting according to the user's loan probability. In order to save computing power, the loan probability can also be divided into multiple intervals. For example, the loan probabilities of 0%-10%, 10%-30%, 30%-50%, 50%-70%, and 70%-100% can be divided into five loan probability intervals. According to the probability of the target users, the target users are divided into the loan intervals in turn, and the target users are ranked according to the interval dimension. This application does not impose any restrictions on this.

[0110] S305: Based on the qualification probability, obtain the second ranking of the target user in the target group.

[0111] Among them, according to the qualification probability of the target user output by the user qualification model, the target user is ranked in the target group. Further, because the present invention divides the user qualification model into a bank loan willingness sub-model and a consumer finance loan willingness sub-model according to the type of loan products; and the bank loan willingness sub-model is divided into: a credit loan willingness sub-model and a mortgage loan willingness sub-model. It can be understood that the above sub-models will respectively output the loan probability of the target user to the target user's bank loan products, consumer finance loan products, and credit loan products and mortgage loan products under the bank loan products. In order to recommend loan products to different users, the qualification probabilities output by the user qualification model and its sub-models are ranked respectively, and the ranking is the second ranking. Its ranking method is, for example, as described in S304, and the present invention does not elaborate on this.

[0112] S306: Recommending loan products to the user based on the first ranking and the second ranking.

[0113] Among them, different loan products have different bank interest rates and user qualification requirements. When recommending loan products to users, in order to achieve the effect of matching different users with different types of loan products and recommending loan products to them, it is necessary to combine the first ranking and second ranking of the target user for recommendation. For example, among the credit loan products under bank loan products, for credit loan products with higher interest rates and lower user qualification requirements, select the target users who rank in the top 5% in the first ranking and the top 10%-30% in the second ranking; for loan products with lower bank interest rates and higher user qualification requirements, select the target users who rank in the top 5%-10% in the first ranking and the top 10% in the second ranking.

[0114] The model training method provided in this embodiment obtains the operator data of the target user from the operator platform, inputs the operator data into the loan willingness model and the user qualification model respectively, and obtains the loan probability output by the loan willingness model and the qualification probability output by the user qualification model. According to the loan probability, the first ranking of the target user in the target group is obtained; according to the qualification probability, the second ranking of the target user in the target group is obtained, and the loan product is recommended to the user in combination with the first ranking and the second ranking. In the current situation where there are many types of loan products with different bank interest rates and user qualification requirements, it is possible to match different users with different types of loan products and recommend loan products to them, achieving the effect of one thousand faces for one thousand people.

[0115] Figure 4 This is a schematic diagram of the structure of the loan prediction device provided by this application. Figure 1 The device comprises:

[0116] An acquisition module 41 acquires user sample data, wherein the user sample data includes the user's operator data and a first loan tag and a second loan tag, wherein the first loan tag is used to mark whether the user accepts the loan, and the second loan tag is used to mark whether the user is successfully granted credit after accepting the loan;

[0117] A first training module 42 is configured to train a loan willingness model according to the operator data of the user and the first loan label to obtain a trained loan willingness model;

[0118] The second training module 43 trains the user qualification model according to the operator data of the user and the second loan tag to obtain a trained user qualification model.

[0119] Optionally, the acquisition module 41 is specifically used for:

[0120] Obtaining user credit data authorized by both the user and the financial institution from the financial institution, and screening out target users with loan intentions from the user credit data;

[0121] Acquire the operator data of the target user from the operator platform, wherein the operator data includes Internet access log data, user call data and user location data;

[0122] The user sample data is obtained according to the operator data and credit data of the target user, wherein the credit data includes whether the target user accepts the loan and whether the credit is successfully granted after accepting the loan.

[0123] Optionally, the acquisition module 41 is further used to acquire the target time for successful credit granting of the target user;

[0124] The acquisition module 41 is further used to acquire operator data within a preset time period before the target time from the operator platform, wherein the preset time periods corresponding to the Internet log data, the user call data and the user location data are different.

[0125] Optionally, the loan prediction device further includes: a processing module 44;

[0126] The processing module 44 is used to determine whether the operator data and credit data of the target user are abnormal or missing;

[0127] When there are abnormalities or missing in the operator data and credit data of the target user, the K nearest neighbor method is used to determine the K samples closest to the missing or abnormal data samples, and the missing or abnormal data is filled based on the weighted average of the K samples to obtain the user sample data.

[0128] Figure 5 This is a schematic diagram of the structure of the loan prediction device provided by this application. Figure 2 The device comprises:

[0129] Acquisition module 51: Acquisition of operator data of a target user from an operator platform, wherein the operator data includes Internet access log data, user call data and user location data;

[0130] Processing module 52: inputting the operator data into a loan willingness model and a user qualification model respectively, wherein the loan willingness model and the user qualification model are obtained by training using the method described in any one of the above items;

[0131] Prediction module 53: obtaining the loan probability output by the loan willingness model and the qualification probability output by the user qualification model;

[0132] Recommendation module 54: recommends loan products to the user based on the loan probability and the qualification probability.

[0133] Figure 6 : is a schematic diagram of the structure of the loan prediction device provided by the present application. The loan prediction device comprises:

[0134] Memory;

[0135] processor;

[0136] Wherein, the memory stores computer-executable instructions;

[0137] The processor executes the computer-executable instructions stored in the memory to implement the loan prediction method as described in the first aspect and various possible implementations of the first aspect.

[0138] The present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the loan prediction method performed by the above-mentioned loan prediction device is implemented.

[0139] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0140] It will be appreciated by those skilled in the art that all or some of the steps, systems, and functional modules / units in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. In hardware implementations, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed by several physical components in cooperation. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is well known to those of ordinary skill in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0141] So far, the technical solution of the present application has been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present application is obviously not limited to these specific embodiments, and the above embodiments are only used to illustrate the technical solution of the present application rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A loan prediction method, characterized in that: The method comprises: Acquire user sample data, the user sample data including the user's operator data and a first loan tag and a second loan tag, the first loan tag is used to mark whether the user accepts the loan, and the second loan tag is used to mark whether the user is successfully granted credit after accepting the loan; Training a loan willingness model according to the operator data of the user and the first loan label to obtain a trained loan willingness model; The user qualification model is trained according to the operator data of the user and the second loan label to obtain a trained user qualification model.

2. The method according to claim 1, characterized in that The obtaining of user sample data includes: Obtain user credit data from financial institutions, and screen target users with loan intentions from the user credit data, wherein the credit data is credit data authorized by both the user and the financial institution; Acquire the operator data of the target user from the operator platform, wherein the operator data includes Internet access log data, user call data and user location data; The user sample data is obtained according to the operator data and credit data of the target user, wherein the credit data includes whether the target user accepts the loan and whether the credit is successfully granted after accepting the loan.

3. The method according to claim 1, characterized in that The acquiring the operator data of the target user from the operator platform includes: acquiring the target time for successful credit granting of the target user provided by a financial institution; The operator data within a preset time period before the target time is obtained from the operator platform, wherein the preset time periods corresponding to the Internet log data, the user call data and the user location data are different.

4. The method according to claim 1, characterized in that The acquiring the user sample data according to the operator data and credit data of the target user includes: Determine whether the target user's operator data and credit data are abnormal or missing; If so, the K nearest neighbor method is used to determine the K samples closest to the missing or abnormal data sample, and the missing or abnormal data is filled according to the weighted average of the K samples to obtain the user sample data.

5. The method according to claim 1, characterized in that The loan willingness model and user qualification model are trained by the XGBoost analysis method.

6. A loan prediction method, characterized in that: The method comprises: Acquire the operator data of the target user from the operator platform, wherein the operator data includes Internet access log data, user call data and user location data; Inputting the operator data into a loan willingness model and a user qualification model respectively, wherein the loan willingness model and the user qualification model are obtained by training according to the method according to any one of claims 1 to 5; Obtaining the loan probability output by the loan willingness model and the qualification probability output by the user qualification model; Recommend loan products to the user based on the loan probability and the qualification probability.

7. The method according to claim 6, characterized in that The recommending a loan product to the user according to the loan probability and the qualification probability includes: With respect to the loan probability, obtaining the first ranking of the target user in the target group; Based on the qualification probability, obtaining the second ranking of the target user in the target group; Recommending loan products to the user based on the first ranking and the second ranking.

8. A loan prediction device, characterized in that: include: Acquisition module: Acquisition of user sample data, wherein the user sample data includes the user's operator data and a first loan tag and a second loan tag, wherein the first loan tag is used to mark whether the user accepts the loan, and the second loan tag is used to mark whether the user is successfully granted credit after accepting the loan; A first training module: training a loan willingness model according to the operator data of the user and the first loan label to obtain a trained loan willingness model; The second training module: trains the user qualification model according to the operator data of the user and the second loan label to obtain a trained user qualification model.

9. A loan prediction device, characterized in that: include: Acquisition module: Acquisition of operator data of target users from the operator platform, wherein the operator data includes Internet access log data, user call data and user location data; Processing module: inputting the operator data into a loan willingness model and a user qualification model respectively, wherein the loan willingness model and the user qualification model are obtained by training according to the method according to any one of claims 1 to 5; Prediction module: obtaining the loan probability output by the loan willingness model and the qualification probability output by the user qualification model; Recommendation module: recommends loan products to users based on the loan probability and the qualification probability.

10. A loan prediction device, characterized in that: include: Memory; processor; Wherein, the memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the loan prediction method according to any one of claims 1 to 7.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, they are used to implement the loan prediction method according to any one of claims 1 to 7.

Images