High-value user identification method and apparatus, electronic device, medium, and program product

Abstract

This application discloses a method, apparatus, electronic device, medium, and program product for identifying high-value users, relating to the field of financial technology. The high-value user identification method includes: acquiring the original feature data of a target user; inputting the original feature data into a preset first user identification model to obtain a first identification result corresponding to the target user; inputting the original feature data into a preset second user identification model to obtain a second identification result corresponding to the target user; inputting the original feature data into a preset third user identification model to obtain a third identification result corresponding to the target user; and determining the target identification result of the target user based on the first, second, and third identification results. The technical solution of this application solves the problem of low comprehensiveness and accuracy in customer identification when mining high-value customers in financial scenarios.

Description

Technical Field

[0001] This application relates to the field of financial technology, and in particular to methods, devices, electronic devices, computer-readable storage media, and computer program products for identifying high-value users. Background Technology

[0002] Currently, when financial institutions target high-value customers in credit scenarios, they primarily rely on basic customer attributes, consumption behavior, and historical credit behavior, using expert experience rules for segmentation or traditional machine learning models for high-value customer identification. However, traditional machine learning models and custom expert experience rules often focus on certain category attributes in customer data, resulting in limitations in the definition of high-value customers. They cannot generalize to all behavioral scenarios and therefore cannot identify all the full characteristics of high-value customers, leading to low comprehensiveness and accuracy in identifying high-value customers.

[0003] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main purpose of this application is to provide a method, device, electronic device, computer-readable storage medium, and computer program product for identifying high-value users, aiming to solve the technical problem of low comprehensiveness and accuracy in customer identification when mining high-value customers in financial scenarios.

[0005] To achieve the above objectives, this application proposes a method for identifying high-value users, the method comprising:

[0006] The original feature data of the target user is obtained, and the original feature data is input into a preset first user identification model to obtain a first identification result corresponding to the target user. The first user identification model is trained from the location data, device data, communication behavior data and access behavior data of historical users.

[0007] The original feature data is input into a preset second user identification model to obtain a second identification result corresponding to the target user. The second user identification model is trained from the network relationship behavior feature information of historical users.

[0008] The original feature data is input into a preset third user identification model to obtain the third identification result corresponding to the target user. The third user identification model is trained by predefined association feature information corresponding to high-value users and non-high-value users respectively.

[0009] Based on the first identification result, the second identification result, and the third identification result of the target user, the target identification result of the target user is determined.

[0010] In one embodiment, before the step of inputting the original feature data into a preset second user identification model to obtain a second identification result corresponding to the target user, the method further includes:

[0011] Collect basic behavioral data of high-value historical users from the primary carrier.

[0012] Based on the basic behavioral data of the first operator, a network relationship behavior feature map of high-value customer groups is constructed.

[0013] The preset graph network model is updated based on the network relationship behavior feature map to obtain the graph network user identification model;

[0014] Based on the network relationship behavior feature map, obtain the network relationship behavior feature information of historical users;

[0015] The network relationship behavior feature information of the historical users is input into the graph network user identification model, and the node information in the graph network user identification model is updated until the preset training stopping condition is reached, thus obtaining the second user identification model.

[0016] In one embodiment, the second user identification model includes at least a first node that is interconnected with each other and a next node that is associated with the first node;

[0017] The step of inputting the network relationship behavior feature information of the historical users into the graph network user identification model and updating the node information in the graph network user identification model includes:

[0018] Update the first node in the second user identification model according to the network relationship behavior feature information to obtain the node information corresponding to the first node;

[0019] The first node sends the node information of the first node to the next node associated with the first node.

[0020] After the next node receives the node information, it performs average label merging processing on the node information and its own node information to obtain label merging feature information;

[0021] The tag merging feature information is merged with its own node information to obtain the updated node information;

[0022] The next node sends the updated node information to the next node associated with it, and then returns to execute the following steps: after the next node receives the node information, it performs average label merging processing on the node information and its own node information until the node information of all nodes in the graph network user identification model is updated.

[0023] In one embodiment, the third user identification model is a clustering analysis model;

[0024] Before the step of inputting the original feature data into a preset third user identification model to obtain the third identification result corresponding to the target user, the method further includes:

[0025] Feature data corresponding to high-value samples and non-high-value samples are collected, wherein the high-value samples and non-high-value samples are determined by preset business rules, and the feature data includes communication data, access data and behavioral trajectory data;

[0026] Based on the feature data corresponding to the high-value samples and non-high-value samples respectively, a high-value sample correction feature matrix is ​​generated;

[0027] The rows in the high-value sample correction feature matrix are used as samples of one dimension to obtain multiple samples;

[0028] Cluster analysis is performed on each sample based on multiple clustering dimensions to obtain multiple clusters, wherein the clustering dimensions include at least credit channel, interest rate, credit limit and number of periods;

[0029] Based on the identified clusters, a corresponding clustering analysis model is generated, wherein the clustering analysis model is used to determine the probability that the sample belongs to each of the clusters based on the feature data of the input sample.

[0030] In one embodiment, the step of generating a high-value sample correction feature matrix based on the feature data corresponding to the high-value samples and non-high-value samples respectively includes:

[0031] Construct a similarity matrix for each of the samples based on the feature data;

[0032] Generate the adjacency matrix and angle matrix of the samples based on the similarity matrix corresponding to each sample;

[0033] Calculate the high-value sample correction matrix for each sample based on the adjacency matrix and angle matrix of each sample;

[0034] Determine the corrected feature vector of each of the high-value sample correction matrices;

[0035] The correction feature vectors of each high-value sample correction matrix are subjected to matrix standardization to obtain the high-value sample correction feature matrix.

[0036] In one embodiment, the first identification result, the second identification result, and the third identification result are all in the form of user value scores;

[0037] The step of determining the target user's target identification result based on the first identification result, second identification result, and third identification result includes:

[0038] Obtain the first weight, second weight, and third weight corresponding to the first user identification model, the second user identification model, and the third user identification model, respectively;

[0039] The product of the first weight and the first recognition result, the product of the second weight and the second recognition result, and the product of the third weight and the third recognition result are calculated respectively. Finally, the sum of each product is calculated to obtain the target recognition result of the target user, wherein the target recognition result is in the form of a user value score.

[0040] Furthermore, to achieve the above objectives, this application also proposes a high-value user identification device, which includes:

[0041] The first identification module is used to acquire the original feature data of the target user, input the original feature data into the preset first user identification model, and obtain the first identification result corresponding to the target user. The first user identification model is trained by historical user location data, device data, communication behavior data and access behavior data.

[0042] The second identification module is used to input the original feature data into a preset second user identification model to obtain the second identification result corresponding to the target user, wherein the second user identification model is trained by the network relationship behavior feature information of historical users;

[0043] The third identification module is used to input the original feature data into a preset third user identification model to obtain the third identification result corresponding to the target user. The third user identification model is trained by predefined association feature information corresponding to high-value users and non-high-value users respectively.

[0044] The target recognition module is used to determine the target recognition result of the target user based on the first recognition result, the second recognition result, and the third recognition result of the target user.

[0045] In addition, to achieve the above objectives, this application also proposes an electronic device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the high-value user identification method as described above.

[0046] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the high-value user identification method described above.

[0047] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the high-value user identification method described above.

[0048] This application proposes a method for identifying high-value users. The method includes: first, acquiring the original feature data of a target user; inputting the original feature data into a preset first user identification model to obtain a first identification result corresponding to the target user, wherein the first user identification model is trained from historical user location data, device data, communication behavior data, and access behavior data; inputting the original feature data into a preset second user identification model to obtain a second identification result corresponding to the target user, wherein the second user identification model is trained from historical user network relationship behavior feature information; inputting the original feature data into a preset third user identification model to obtain a third identification result corresponding to the target user, wherein the third user identification model is trained from predefined association relationship feature information corresponding to high-value users and non-high-value users respectively; and finally, determining the target user's target identification result based on the first identification result, the second identification result, and the third identification result. This application's high-value user identification method not only employs traditional machine learning models based on multidimensional user data for identification, but also combines models trained using user relationship behavior features for identification. Furthermore, it effectively learns the characteristics of high-value customers by learning the relationship features between high-value and non-high-value customers. Finally, by comprehensively considering the identification results output by the three different models, a target identification result that fully identifies multiple user characteristics is obtained. Compared with traditional expert experience rules and traditional machine learning model identification schemes, this application's technical solution combines the identification results of multiple models, more comprehensively covers user characteristics, overcomes limitations, and thus improves the comprehensiveness and accuracy of high-value identification results. Attached Figure Description

[0049] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0050] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a flowchart illustrating the high-value user identification method in Embodiment 1 of this application.

[0052] Figure 2 This is a schematic diagram illustrating the improvement in AUC metrics of test samples after applying the embodiments of this application compared to before applying the embodiments of this application;

[0053] Figure 3 This diagram illustrates the improvement in KS index of test samples after applying the embodiments of this application compared to before applying the embodiments of this application.

[0054] Figure 4 This is a schematic diagram of the structural layering of the first user identification model in the embodiments of this application;

[0055] Figure 5 This is a schematic diagram of the structural composition of a high-value user identification device according to an embodiment of this application;

[0056] Figure 6 This is a schematic diagram of the device structure of the hardware operating environment involved in the high-value user identification method in this application embodiment.

[0057] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0058] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0059] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0060] The executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, mobile phone, computer, server, etc., or an electronic device or control device capable of performing the above functions. The following description uses a computer as the executing entity to illustrate this embodiment and the subsequent embodiments.

[0061] In the financial sector, financial institutions primarily rely on customer attributes, consumption behavior, and historical credit history to identify high-value customers in lending scenarios. They then use expert-based rule combinations for customer segmentation or traditional machine learning models for high-value customer identification. However, traditional machine learning models and expert-based rules often focus on specific attribute categories, resulting in limitations in defining high-value customers. They cannot generalize to all user behavior scenarios and therefore cannot identify all the characteristics of high-value customers. Furthermore, traditional machine learning models mainly rely on basic attributes and statistical features to train classification and prediction models, lacking effective methods for uncovering network relationship characteristics and group association characteristics of high-value customers, thus failing to comprehensively assess and identify them.

[0062] To improve the comprehensiveness and accuracy of high-value customer assessment and identification, embodiments of this application provide a high-value user identification method, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the high-value user identification method of this application. The high-value user identification method includes:

[0063] Step S10: Obtain the original feature data of the target user, input the original feature data into the preset first user identification model, and obtain the first identification result corresponding to the target user. The first user identification model is trained by the location data, device data, communication behavior data and access behavior data of historical users.

[0064] In this context, "target users" refers to users who need to be identified as high-value users, and their mobile phone numbers can be used as their identifiers. In the context of telecommunications operators, various dimensions of raw characteristic data associated with the target user's mobile phone number can be obtained, such as location data, device data, communication behavior data, and access behavior data. Location data can include the user's location information; device data includes the model and price of the user's mobile device (such as a mobile phone); communication behavior data includes the frequency and duration of communication between the user and other users, as well as information about other users; and access behavior data includes information about the applications and websites accessed by the user using their mobile device, which may reflect the user's consumption habits and financial needs.

[0065] In addition, the preset first user identification model can be a traditional machine learning model, which is mainly trained by multi-dimensional feature data of users collected in history. It can make a prediction and identification of the high value of users based on the notification data of these dimensions.

[0066] For example, the first identification result can be presented as a score or percentage, with a higher score or percentage indicating a higher value of the user in the personal loan scenario.

[0067] Step S20: Input the original feature data into the preset second user identification model to obtain the second identification result corresponding to the target user. The second user identification model is trained by the network relationship behavior feature information of historical users.

[0068] In addition to using the aforementioned first user identification model to predict and identify high-value target users, this application embodiment also uses another second user identification model to predict and identify target users.

[0069] It should be noted that this second user identification model is a graph network behavioral relationship model obtained by improving and optimizing the propagation process of the traditional graph network model. Based on this model and the target user's original feature data, it accurately identifies whether the target user is a high-value personal loan customer. Compared to the first user identification model, the second model is trained using the user's network relationship behavioral feature information. It can accurately predict and identify the user's loan value from another perspective, based on the network relationship behavioral features contained in the target user's original feature data. This enriches the dimensions of the feature data when predicting and identifying high-value users, making the final identification result more comprehensive.

[0070] For example, the second identification result can be presented as a score or percentage, with a higher score or percentage indicating a higher value of the user in the personal loan scenario.

[0071] Step S30: Input the original feature data into the preset third user identification model to obtain the third identification result corresponding to the target user. The third user identification model is trained by the predefined association feature information corresponding to high-value users and non-high-value users respectively.

[0072] The technical solution of this application uses three different models. Two of these models have been described in steps S10 to S20. The third model is a user identification model trained by learning the relationship characteristics between high-value and non-high-value credit customer groups.

[0073] The third user identification model utilizes predefined data on the correlation features between high-value and non-high-value users during training. This model mines correlation features between non-high-value credit customers and high-value customers, performing differentiated mining on these two groups. It learns strong correlation features among high-value credit customers, uncovering correlation features that are difficult to detect, thus making the sample processing features of the high-value credit customer identification model more comprehensive and ensuring that the model can identify more high-value credit customer correlation features that are not detected by existing technologies. Furthermore, when predicting the third identification result, the third user identification model uses clustering technology, taking the score or percentage corresponding to the highest probability category of the target user as the third identification result.

[0074] For example, the third identification result can be presented as a score or percentage, with a higher score or percentage indicating a higher value of the user in the personal loan scenario.

[0075] Step S40: Determine the target identification result of the target user based on the first identification result, the second identification result, and the third identification result of the target user.

[0076] After obtaining the target user's prediction results in three different models, the pre-set weights corresponding to these three models can be combined to obtain a comprehensive target recognition result. This target recognition can be in the form of a score or a percentage. The higher the score or percentage, the higher the value of the user in the personal loan scenario.

[0077] For example, after determining the target user's target identification result, a pre-set rule for determining whether the target user is a high-value user can be used, such as a score threshold or a percentage threshold. If the target identification result is greater than the score threshold or percentage threshold, the target user is determined to be a high-value user; otherwise, the target user is a non-high-value user. This achieves intuitive identification and judgment of the target user.

[0078] Furthermore, if the first identification result, the second identification result, and the third identification result are all in the form of user value scores, step S40 may include:

[0079] Step S41: Obtain the first weight, second weight, and third weight corresponding to the first user identification model, the second user identification model, and the third user identification model, respectively.

[0080] Step S42: Calculate the product of the first weight and the first recognition result, the product of the second weight and the second recognition result, and the product of the third weight and the third recognition result respectively. Finally, calculate the sum of each product to obtain the target recognition result of the target user. The target recognition result is in the form of user value score.

[0081] For example, the target user's mobile phone number and user value score (i.e., recognition result) output by each model are respectively: (phone, V1_score), (phone, V2_score), and (phone, V3_score), where phone is the target user's mobile phone number, V1_score is the first recognition result output by the first user recognition model, V2_score is the first recognition result output by the first user recognition model, and V3_score is the first recognition result output by the first user recognition model.

[0082] The user's final high-value credit score is calculated based on a weighted average of the scores output by the three models, resulting in a total credit value score (phone, V_score). For example, the calculation rules are as follows:

[0083] V_score=0.2*V1_score+0.45*V2_score+0.35*V3_score;

[0084] Where V_score represents the target recognition result, and 0.2, 0.45, and 0.35 represent the first, second, and third weights of the first, second, and third user recognition models, respectively.

[0085] The high-value user identification method in this application is based on an improved graph network model and clustering technology combined with multiple model combinations to identify high-value credit customers. Finally, the AUC (Area Under Curve, ROC (receiver operating characteristic curve) and KS (Kolmogorov-Smirnov statistic, the degree of difference between two samples or samples and the theoretical distribution) indices are statistically analyzed for the three model combinations (V4) and the single model (V1), respectively. Figure 2 and Figure 3 (The horizontal axis in the figure represents the false positive rate, and the vertical axis represents the true positive rate.) As shown, on the test sample, the model's AUC index improved by 55%, and the ks index (0.23 for V1 and 0.41 for V4) improved by 78%. The overall prediction effect of the model has been significantly improved, enabling more accurate identification of high-value users in personal loan scenarios in the financial field.

[0086] Furthermore, in a feasible embodiment, prior to step S10, the high-value user identification method further includes a method for training a first user identification model, such as... Figure 4 As shown, the first user identification model provided by the technical solution of this application embodiment can be a high-value credit customer identification model, including an input layer, an intermediate layer, and a prediction layer. The input layer consists of four main categories of data: user location data, device data, communication behavior data, and access behavior data, used to generate various basic features for identifying high-value credit customers. The intermediate layer is used to numerically encode and standardize the various full features integrated by the input layer, processing them into feature vector data (i.e., train data) for model prediction input. Due to the wide dimensionality of operator data, most of the original features are categorical and continuous features; the feature processing adopts a chi-square binning method.

[0087] In addition, the prediction layer consists of two prediction layers (prediction layer 1 and prediction layer 12). Since credit scenarios are divided into many categories according to dimensions such as channel, interest rate, amount, and term, and the high-value target customer groups under each sub-category are also different, in order to achieve accurate identification of high-value people under each category, it is necessary to make predictions for each category in the prediction layer. The first sub-layer is used to predict the probability that the user to be identified belongs to the high-value user under each credit category. The second sub-layer predicts the value level of high-value credit customers based on the multi-task learning mechanism and the probability output by the first sub-layer.

[0088] Furthermore, in a feasible embodiment, before the step of inputting the original feature data into a preset second user identification model to obtain the second identification result corresponding to the target user, the method may further include:

[0089] Step A10: Collect basic behavioral data of the first operator for historical users tagged as high-value;

[0090] Step A20: Based on the basic behavioral data of the first operator, construct a network relationship behavior feature map of high-value customer groups;

[0091] Step A30: Update the preset graph network model based on the network relationship behavior feature graph to obtain the graph network user identification model;

[0092] Step A40: Obtain historical users' network relationship behavior feature information based on the network relationship behavior feature graph;

[0093] Step A50: Input the network relationship behavior feature information of historical users into the graph network user identification model, update the node information in the graph network user identification model until the preset training stopping condition is reached, and obtain the second user identification model.

[0094] The first operator's basic behavioral data includes network relationship behavior data of users tagged as high-value, including device behavior information, location behavior information, access behavior information, and communication behavior information. These tags can be assigned by staff according to business rules. Specifically, device behavior information includes device type information, device replacement information, and operating system information; location behavior information includes travel information, base station access information, workplace / residence information, and Wi-Fi connection information; communication behavior data includes specific call and SMS behavior information; and access behavior data includes specific website and app access information.

[0095] Based on the mobile operator data above, a network relationship behavior feature graph of corresponding high-value customer groups can be constructed. Specifically, network relationship behavior features include: high-end device usage behavior, high-end venue stay behavior, high-value APP and website access behavior, and high-end institution communication behavior, etc. Simultaneously, to identify high-value customer groups, it is necessary to mine graph association information and construct a relationship graph of high-value credit users using various user network behaviors. Then, based on the graph network model, the network relationship behavior feature information of high-value customer groups in the graph can be quickly and accurately located.

[0096] Subsequently, the network relationship and behavioral characteristics of high-value customer groups from historical users are input into the graph network user identification model to update the node information of the network nodes, thereby training the model until the model reaches a preset training stopping condition, at which point training stops, resulting in a trained second user identification model. The training stopping condition can be one or more of the following: a preset number of training iterations, the model accuracy reaching a preset threshold, or the model loss converging, ensuring that the model's performance meets the requirements for high-value user identification.

[0097] Furthermore, the second user identification model includes at least a first node that is interconnected with the first node and a next node that is associated with the first node; the step of inputting the network relationship behavior feature information of historical users into the graph network user identification model and updating the node information in the graph network user identification model may include:

[0098] Step A51: Update the first node in the second user identification model according to the network relationship behavior feature information to obtain the node information corresponding to the first node;

[0099] Step A52: Send the node information of the first node to the next node associated with the first node through the first node;

[0100] Step A53: After the next node receives the node information, it performs average label merging processing on the node information and its own node information to obtain the label merging feature information.

[0101] Step A54: Merge the label merged feature information with its own node information to obtain the updated node information;

[0102] Step A55: The next node sends the updated node information to the next node associated with it, and then returns to execute step A53: After the next node receives the node information, it performs average label merging processing on the node information and its own node information until the node information of all nodes in the graph network user identification model is updated.

[0103] It should be noted that during the training of a graph network model, gradient misalignment may occur during the backpropagation of error from the output layer to the input layer, which may cause the model performance to fall short of expectations. To address this issue, the technical solution in this application improves the propagation process of the graph network model by employing a gradient propagation method based on label merging. For example, the label merging formula is as follows:

[0104]

[0105] in, For the node information before its own update, Its associated node information, This is the node information after tag merging. For example, the node information is illustrated in the table below:

[0106]

[0107] Specifically, the steps for updating the information of each node are as follows: First, update the node information of the first node directly connected to the input port based on the obtained network relationship behavior feature information. Then, the first node (i.e., this node) sends its own feature information to its associated next node. After receiving the relevant information, the associated next node will update its own feature information. The associated node processes the received message through the label merging formula to generate a new feature after label merging. Then, the feature after label merging is merged with the node's own feature by addition. Update the node features, and so on, until the node information of all nodes in the model is updated.

[0108] In this embodiment, an improved algorithm is used on the basis of the graph network model to perform supervised correction on the feature information propagation path, thereby achieving accurate calculation of the propagation gradient. This effectively improves the performance of the graph network model and makes the second user identification model more accurate in predicting high-value user situations.

[0109] Furthermore, in a feasible embodiment, the third user identification model can be a clustering analysis model. Before the step of inputting the original feature data into the preset third user identification model to obtain the third identification result corresponding to the target user, the method may further include:

[0110] Step B10: Collect feature data corresponding to high-value samples and non-high-value samples respectively. High-value samples and non-high-value samples are determined by preset business rules. Feature data includes communication data, access data and behavioral trajectory data.

[0111] Step B20: Generate a high-value sample correction feature matrix based on the feature data corresponding to high-value samples and non-high-value samples respectively.

[0112] Step B30: Take the row in the high-value sample correction feature matrix as a sample of one dimension to obtain multiple samples;

[0113] Step B40: Perform cluster analysis on each sample based on multiple clustering dimensions to obtain multiple clusters. The clustering dimensions include at least credit channel, interest rate, amount, and number of periods.

[0114] Step B50: Generate the corresponding clustering analysis model based on the determined clusters. The clustering analysis model is used to determine the probability that a sample belongs to each cluster based on the feature data of the input sample.

[0115] This application also provides a method for training a third-party user identification model. The training method for the third-party user identification model takes into account the sparse nature of operator data in specific scenarios, improves and optimizes clustering techniques, and learns the relationship characteristics between high-value and non-high-value credit customer groups to output effective group association behavior characteristics of high-value customer groups, thereby further and more accurately identifying high-value credit customers.

[0116] First, in the process of training the third user identification model in this application embodiment, the correlation characteristics of non-credit high-value customer groups and high-value customer groups are mined. By performing differentiated mining on the two types of customer groups, the strong correlation characteristics of credit high-value customers are learned, and a credit high-value customer identification model is trained. This customer identification model is the third user identification model.

[0117] Specifically, firstly, based on predefined high-value user and non-high-value user sample data according to business rules, improved clustering techniques are applied to construct basic association features. The basic data for constructing association features includes three major categories: communication, access, and behavioral trajectory data. Then, a similarity matrix of samples is constructed based on the feature data to further generate the corresponding high-value sample corrected feature matrix. The high-value sample corrected feature matrix reflects the corresponding features of the feature data corresponding to high-value users and can be used for cluster analysis. Further, each row of the high-value sample corrected feature matrix U is taken as a k-dimensional sample, with a total of n samples. Then, clustering is performed using the input clustering method. For example, after subdividing by multiple dimensions such as credit channel, interest rate, amount, and term, m high-value subcategories are defined, and the clustering dimension is m. The new sample points Y = {y1, y2, ..., y3} are then clustered. n Clustering into clusters C1, C2, ..., C n Finally, using the probability values ​​P1, P2, ..., P of each sample belonging to each cluster output in the previous step... m The highest probability category of the sample output is used as the predicted value of the V3 model. The probability category can be one of low value, medium value, and high value, or it can be in the form of a score or percentage, for example, 20% for low value, 50% for medium value, and 80% for high value.

[0118] Furthermore, the step of generating a high-value sample correction feature matrix based on the feature data corresponding to high-value samples and non-high-value samples respectively may include:

[0119] Step A21: Construct similarity matrices for each of the samples based on the feature data;

[0120] Step A23: Generate the adjacency matrix and angle matrix of the samples based on the similarity matrix corresponding to each sample;

[0121] Step A24: Calculate the high-value sample correction matrix for each sample based on the adjacency matrix and angle matrix of each sample.

[0122] Step A25: Determine the correction feature vector of the correction matrix for each high-value sample;

[0123] Step A26: Perform matrix standardization on the correction feature vectors of each high-value sample correction matrix to obtain the high-value sample correction feature matrix.

[0124] For example, in the feature data, the total number of high-value samples is n, each sample has m1 corresponding communication features, m2 corresponding DPI (Deep Packet Inspection) access features, and m3 corresponding behavioral trajectory features. The total number of features after merging is m = m1 + m2 + m3, which constitutes a single sample vector x.i ={x i1 ,x i2 ,…,x im Then, the similarity matrix of the samples can be represented as: X = {x1, x2, ..., x} n For example, an instance of the sample similarity matrix X can be:

[0125]

[0126] In this matrix, the values ​​from left to right represent: number of contacts, travel distance, number of top-ups, and number of packages. Next, based on the high-value sample similarity matrix X, an adjacency matrix W for the samples is generated. First, the following formula is applied to calculate the n*n adjacency matrix W.

[0127]

[0128] Where W is S ij The high-value sample adjacency matrix is ​​formed. Then, the high-value sample adjacency matrix is ​​calculated using the following formula.

[0129]

[0130] That is, the sum of the elements in each row of the adjacency matrix W is equal to the angle matrix D. i This forms an n*n pair of angle matrices. Further, the corrected inverse matrix Ln, corresponding to both the angle matrix D and the adjacency matrix W, is constructed. Specifically, the calculation expression for the corrected inverse matrix is:

[0131] L n =DW (-1) ;

[0132] After the angle matrix and adjacency matrix are constructed, the high-value sample correction matrix can be obtained by adding the angle matrix and adjacency matrix, multiplying by the aforementioned correction inverse matrix, and finally taking the square root.

[0133]

[0134] Where Lt is the high-value sample correction matrix. Next, the k1 corrected eigenvalues ​​and corrected eigenvectors u of the high-value sample correction matrix Lt are calculated. Specifically, the eigenvalues ​​of Lt are calculated, sorted from smallest to largest, and the first k eigenvalues ​​are selected. The eigenvectors u1, u2, ..., u of the first k eigenvalues ​​are then calculated. k .

[0135] Finally, the k column vectors above are used to form a matrix {u1, u2, ..., u k Then, the matrix is ​​standardized row by row, and finally an n*k dimensional high-value sample correction feature matrix U, U∈R is formed. n*k.

[0136] In conjunction with the aforementioned embodiments, the high-value user identification method provided in this application improves and optimizes the gradient propagation process of the graph network model based on historical high-value customer network behavior characteristics, training a high-value credit customer identification model (i.e., the second user identification model). The network behavior characteristics refer to features generated by combining customer spatial, temporal, and event-related behaviors based on network relationships. Furthermore, it processes specific behavioral characteristics based on mobile operator data. Considering the sparse nature of high-value credit customer feature data, it improves and optimizes existing clustering methods by constructing a feature correction matrix. The resulting third user identification model performs clustering based on the transformed correction matrix rather than on the original feature data. By learning the correlation behaviors between high-value and non-high-value credit customer groups, it outputs effective correlation features of high-value customer groups to mine high-value credit customers, improving the comprehensiveness and accuracy of the identification results. This application's high-value user identification method combines the characteristics of operator basic data with the needs of identifying high-value credit users. It uses a traditional machine learning model, an improved graph network model, and an improved clustering model for prediction and identification, respectively, and finally combines the identification results to obtain target identification results with more comprehensive and broader feature coverage and higher accuracy.

[0137] Because the technical solution of this application embodiment is based on mobile operator-specific data to construct a network behavior relationship model, and considering the characteristics of missing order relationships in various dimensions of data, a label merging gradient propagation method is adopted for the gradient propagation process of the graph network model. By pre-simulating the feature information propagation path, the hidden correlation information in the dynamic and static data of the mobile network is fully mined, improving the performance of the graph network model and accurately identifying high-value credit customers from the massive customer base of mobile operators. In addition, the technical solution of this application embodiment addresses the disadvantage of sparse operator-specific behavior data by constructing a corrective feature matrix to improve and optimize existing clustering techniques, further enhancing the clustering effect and effectively mining the value of operator data in identifying high-value credit customers. Moreover, by mining the correlation features of high-value credit customer groups, the accuracy of the high-value credit customer identification model can be improved on the one hand; on the other hand, some new high-value group features that do not exist in the present can be discovered, thus making the sample features of the high-value identification model more extensive, thereby ensuring that the model can identify more high-value specific cluster group correlation features that existing technologies have failed to identify.

[0138] According to the foregoing Figure 2 and Figure 3As can be seen, the technical solution of this application combines a traditional machine learning model (i.e., the first user identification model), an improved graph network model (i.e., the second user identification model), and an improved clustering model (i.e., the third user identification model) to identify the original feature data of target customers in credit scenarios. Compared with the single-model prediction scheme, the combined model has a greatly improved effect.

[0139] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the high-value user identification method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0140] This application also provides a high-value user identification device, referring to... Figure 5 The high-value user identification device includes:

[0141] The first identification module 10 is used to acquire the original feature data of the target user, input the original feature data into a preset first user identification model, and obtain the first identification result corresponding to the target user. The first user identification model is trained by historical user location data, device data, communication behavior data and access behavior data.

[0142] The second identification module 20 is used to input the original feature data into a preset second user identification model to obtain a second identification result corresponding to the target user, wherein the second user identification model is trained by the network relationship behavior feature information of historical users;

[0143] The third identification module 30 is used to input the original feature data into a preset third user identification model to obtain the third identification result corresponding to the target user. The third user identification model is trained by predefined association feature information corresponding to high-value users and non-high-value users respectively.

[0144] The target recognition module 40 is used to determine the target recognition result of the target user based on the first recognition result, the second recognition result and the third recognition result of the target user.

[0145] In one embodiment, the second identification module 20 is further configured to:

[0146] Collect basic behavioral data of high-value historical users from the primary carrier.

[0147] Based on the basic behavioral data of the first operator, a network relationship behavior feature map of high-value customer groups is constructed.

[0148] The preset graph network model is updated based on the network relationship behavior feature map to obtain the graph network user identification model;

[0149] Based on the network relationship behavior feature map, obtain the network relationship behavior feature information of historical users;

[0150] The network relationship behavior feature information of the historical users is input into the graph network user identification model, and the node information in the graph network user identification model is updated until the preset training stopping condition is reached, thus obtaining the second user identification model.

[0151] In one embodiment, the second user identification model includes at least a first node that is interconnected with the first node and a next node that is associated with the first node; the second identification module 20 is further configured to:

[0152] Update the first node in the second user identification model according to the network relationship behavior feature information to obtain the node information corresponding to the first node;

[0153] The first node sends the node information of the first node to the next node associated with the first node.

[0154] After the next node receives the node information, it performs average label merging processing on the node information and its own node information to obtain label merging feature information;

[0155] The tag merging feature information is merged with its own node information to obtain the updated node information;

[0156] The next node sends the updated node information to the next node associated with it, and then returns to execute the following steps: after the next node receives the node information, it performs average label merging processing on the node information and its own node information until the node information of all nodes in the graph network user identification model is updated.

[0157] In one embodiment, the third user identification model is a clustering analysis model; the third identification module is further used for:

[0158] Feature data corresponding to high-value samples and non-high-value samples are collected, wherein the high-value samples and non-high-value samples are determined by preset business rules, and the feature data includes communication data, access data and behavioral trajectory data;

[0159] Based on the feature data corresponding to the high-value samples and non-high-value samples respectively, a high-value sample correction feature matrix is ​​generated;

[0160] The rows in the high-value sample correction feature matrix are used as samples of one dimension to obtain multiple samples;

[0161] Cluster analysis is performed on each sample based on multiple clustering dimensions to obtain multiple clusters, wherein the clustering dimensions include at least credit channel, interest rate, credit limit and number of periods;

[0162] Based on the identified clusters, a corresponding clustering analysis model is generated, wherein the clustering analysis model is used to determine the probability that the sample belongs to each of the clusters based on the feature data of the input sample.

[0163] In one embodiment, the third identification module is further configured to:

[0164] Construct a similarity matrix for each of the samples based on the feature data;

[0165] Generate the adjacency matrix and angle matrix of the samples based on the similarity matrix corresponding to each sample;

[0166] Calculate the high-value sample correction matrix for each sample based on the adjacency matrix and angle matrix of each sample;

[0167] Determine the corrected feature vector of each of the high-value sample correction matrices;

[0168] The correction feature vectors of each high-value sample correction matrix are subjected to matrix standardization to obtain the high-value sample correction feature matrix.

[0169] In one embodiment, the first identification result, the second identification result, and the third identification result are all in the form of user value ratings; the target identification module 40 is further configured to:

[0170] Obtain the first weight, second weight, and third weight corresponding to the first user identification model, the second user identification model, and the third user identification model, respectively;

[0171] The product of the first weight and the first recognition result, the product of the second weight and the second recognition result, and the product of the third weight and the third recognition result are calculated respectively. Finally, the sum of each product is calculated to obtain the target recognition result of the target user, wherein the target recognition result is in the form of a user value score.

[0172] The high-value user identification device provided in this application, employing the high-value user identification method described in the above embodiments, can solve the technical problem of low comprehensiveness and accuracy in customer identification when mining high-value customers in financial scenarios. Compared with the prior art, the beneficial effects of the high-value user identification device provided in this application are the same as those of the high-value user identification method provided in the above embodiments, and other technical features in this high-value user identification device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0173] This application provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the high-value user identification method described above.

[0174] The following is for reference. Figure 6 The diagram illustrates a structural schematic of an electronic device suitable for implementing embodiments of this application. The electronic devices in these embodiments may include, but are not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 6 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0175] like Figure 6 As shown, the electronic device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the electronic device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. While electronic devices with various systems are shown in the figures, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0176] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0177] The electronic device provided in this application, employing the high-value user identification method described in the above embodiments, can solve the technical problem of low comprehensiveness and accuracy in customer identification when mining high-value customers in financial scenarios. Compared with the prior art, the beneficial effects of the electronic device provided in this application are the same as those of the high-value user identification method provided in the above embodiments, and other technical features of this electronic device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0178] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0179] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0180] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to perform the high-value user identification method described in the above embodiments.

[0181] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0182] The aforementioned computer-readable storage medium may be included in an electronic device or may exist independently without being assembled into an electronic device.

[0183] The aforementioned computer-readable storage medium carries one or more programs. When the aforementioned one or more programs are executed by an electronic device, the electronic device causes the following actions: to acquire raw feature data of a target user; to input the raw feature data into a preset first user identification model to obtain a first identification result corresponding to the target user, wherein the first user identification model is trained from historical user location data, device data, communication behavior data, and access behavior data; to input the raw feature data into a preset second user identification model to obtain a second identification result corresponding to the target user, wherein the second user identification model is trained from historical user network relationship behavior feature information; to input the raw feature data into a preset third user identification model to obtain a third identification result corresponding to the target user, wherein the third user identification model is trained from predefined association relationship feature information corresponding to high-value users and non-high-value users respectively; and to determine the target identification result of the target user based on the first identification result, the second identification result, and the third identification result.

[0184] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0185] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0186] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0187] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described high-value user identification method. This addresses the technical problem of low comprehensiveness and accuracy in customer identification when mining high-value customers in financial scenarios. Compared with existing technologies, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the high-value user identification method provided in the above embodiments, and will not be elaborated upon here.

[0188] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the high-value user identification method described above.

[0189] The computer program product provided in this application can solve the technical problem of low comprehensiveness and accuracy in customer identification when mining high-value customers in financial scenarios. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the high-value user identification method provided in the above embodiments, and will not be repeated here.

[0190] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for identifying high-value users, characterized in that, The high-value user identification method includes: The original feature data of the target user is obtained, and the original feature data is input into a preset first user identification model to obtain a first identification result corresponding to the target user. The first user identification model is trained from the location data, device data, communication behavior data and access behavior data of historical users. The original feature data is input into a preset second user identification model to obtain a second identification result corresponding to the target user. The second user identification model is trained from the network relationship behavior feature information of historical users. The original feature data is input into a preset third user identification model to obtain the third identification result corresponding to the target user. The third user identification model is trained by predefined association feature information corresponding to high-value users and non-high-value users respectively. Based on the first identification result, the second identification result, and the third identification result of the target user, the target identification result of the target user is determined.

2. The high-value user identification method as described in claim 1, characterized in that, Before the step of inputting the original feature data into a preset second user identification model to obtain the second identification result corresponding to the target user, the method further includes: Collect basic behavioral data of high-value historical users from the primary carrier. Based on the basic behavioral data of the first operator, a network relationship behavior feature map of high-value customer groups is constructed. The preset graph network model is updated based on the network relationship behavior feature map to obtain the graph network user identification model; Based on the network relationship behavior feature map, obtain the network relationship behavior feature information of historical users; The network relationship behavior feature information of the historical users is input into the graph network user identification model, and the node information in the graph network user identification model is updated until the preset training stopping condition is reached, thus obtaining the second user identification model.

3. The high-value user identification method as described in claim 2, characterized in that, The second user identification model includes at least a first node that is interconnected with each other and a next node that is associated with the first node; The step of inputting the network relationship behavior feature information of the historical users into the graph network user identification model and updating the node information in the graph network user identification model includes: Update the first node in the second user identification model according to the network relationship behavior feature information to obtain the node information corresponding to the first node; The first node sends the node information of the first node to the next node associated with the first node. After the next node receives the node information, it performs average label merging processing on the node information and its own node information to obtain label merging feature information; The tag merging feature information is merged with its own node information to obtain the updated node information; The next node sends the updated node information to the next node associated with it, and then returns to execute the following steps: after the next node receives the node information, it performs average label merging processing on the node information and its own node information until the node information of all nodes in the graph network user identification model is updated.

4. The high-value user identification method as described in claim 3, characterized in that, The third user identification model is a cluster analysis model; Before the step of inputting the original feature data into a preset third user identification model to obtain the third identification result corresponding to the target user, the method further includes: Feature data corresponding to high-value samples and non-high-value samples are collected, wherein the high-value samples and non-high-value samples are determined by preset business rules, and the feature data includes communication data, access data and behavioral trajectory data; Based on the feature data corresponding to the high-value samples and non-high-value samples respectively, a high-value sample correction feature matrix is ​​generated; The rows in the high-value sample correction feature matrix are used as samples of one dimension to obtain multiple samples; Cluster analysis is performed on each sample based on multiple clustering dimensions to obtain multiple clusters, wherein the clustering dimensions include at least credit channel, interest rate, credit limit and number of periods; Based on the identified clusters, a corresponding clustering analysis model is generated, wherein the clustering analysis model is used to determine the probability that the sample belongs to each of the clusters based on the feature data of the input sample.

5. The high-value user identification method as described in claim 1, characterized in that, The step of generating the high-value sample correction feature matrix based on the feature data corresponding to the high-value samples and non-high-value samples respectively includes: Construct a similarity matrix for each of the samples based on the feature data; Generate the adjacency matrix and angle matrix of the samples based on the similarity matrix corresponding to each sample; Calculate the high-value sample correction matrix for each sample based on the adjacency matrix and angle matrix of each sample; Determine the corrected feature vector of each of the high-value sample correction matrices; The correction feature vectors of each high-value sample correction matrix are subjected to matrix standardization to obtain the high-value sample correction feature matrix.

6. The high-value user identification method as described in claim 5, characterized in that, The first identification result, the second identification result, and the third identification result are all in the form of user value scoring; The step of determining the target user's target identification result based on the first identification result, second identification result, and third identification result includes: Obtain the first weight, second weight, and third weight corresponding to the first user identification model, the second user identification model, and the third user identification model, respectively; The product of the first weight and the first recognition result, the product of the second weight and the second recognition result, and the product of the third weight and the third recognition result are calculated respectively. Finally, the sum of each product is calculated to obtain the target recognition result of the target user, wherein the target recognition result is in the form of a user value score.

7. A high-value user identification device, characterized in that, The high-value user identification device includes: The first identification module is used to acquire the original feature data of the target user, input the original feature data into the preset first user identification model, and obtain the first identification result corresponding to the target user. The first user identification model is trained by historical user location data, device data, communication behavior data and access behavior data. The second identification module is used to input the original feature data into a preset second user identification model to obtain the second identification result corresponding to the target user, wherein the second user identification model is trained by the network relationship behavior feature information of historical users; The third identification module is used to input the original feature data into a preset third user identification model to obtain the third identification result corresponding to the target user. The third user identification model is trained by predefined association feature information corresponding to high-value users and non-high-value users respectively. The target recognition module is used to determine the target recognition result of the target user based on the first recognition result, the second recognition result, and the third recognition result of the target user.

8. An electronic device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the high-value user identification method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the high-value user identification method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the high-value user identification method as described in any one of claims 1 to 6.

Images