A user reputation evaluation method based on scorecard model
Through the user credibility evaluation method based on the score card model, the LightGBM model is used for feature learning, and the creditworthiness score is combined with the score card model, the problem of lack of objectivity in the credibility evaluation in the existing technology is solved, and more accurate user credibility scores and more accurate fraud governance is achieved.
Patent Information
- Application Number
- CN202411722007.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-28
- Publication Date
- 2025-04-01
AI Technical Summary
The existing technology lacks objectivity in user credibility assessment, and relying on expert experience, it is impossible to accurately distinguish normal users from fraudulent users, resulting in inaccurate credibility scores.
The user reputation evaluation method based on the score card model is adopted, and the communication data of old users and basic data are collected, the LightGBM model is constructed for feature learning, and the creditworthiness score is scored in combination with the score card model to achieve accurate evaluation of user reputation.
It improves the accuracy of user credibility scores and is objective, avoids the limitations brought by human subjective understanding, and enhances the accuracy of fraud governance.
Smart Images
Figure CN119741082A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of communication technology, and in particular to a user credibility evaluation method based on a scoring card model. Background Art
[0002] At present, the methods of telecommunications network fraud in various places are changing. Basic telecommunications operators continue to counter by building various fraud models, identifying suspected fraud numbers from massive signaling call data and shutting them down. However, most of the existing technologies start from the fraud characteristics of fraud numbers, build various anti-fraud models, and then uniformly shut down numbers through the operator's CRM system. It is either black or white. Due to user privacy, it cannot identify the content of user calls and Internet access. Various anti-fraud models based on behavior have limitations in accuracy, which inevitably leads to many complaints and increases the pressure on customer service to handle complaints. How to conduct differentiated evaluation and disposal based on the existing suspended numbers, such as not allowing online re-enabling for high-risk users after closure, filtering for low-risk users to avoid closure, and allowing one online re-enabling for medium-risk users, etc., is an effective way to reduce the pressure on customer service to handle complaints. Existing technologies often start from fraud numbers to mine fraud characteristics and build fraud models to identify fraud numbers. They rarely start from normal user numbers to mine normal user characteristics and evaluate whether the user is a normal user. The current business development of operators has entered the era of inventory, with more than 80% of users being old users who have been on the network for more than one year. At the same time, due to their long network time, old users have accumulated a large amount of communication behavior data, which is very helpful in mining the characteristics of normal users, identifying normal users, conducting credibility scoring, and ultimately supporting graded disposal.
[0003] The following problems exist in the existing methods for evaluating user value: the features, thresholds, weights, etc. used to calculate the existing user's credibility score often rely on expert experience, and there is no objective positive and negative samples for model training. They are not objective and cannot avoid the limitations brought by human subjective understanding, which ultimately leads to inaccurate user credibility scores; the features used in existing user risk scoring, user classification and disposal are often the fraudulent features of fraudulent users, and normal users who are mistakenly harmed by these features often exist, and it is impossible to effectively distinguish normal users from fraudulent users, which ultimately has limitations.
[0004] Therefore, how to accurately evaluate the user's credibility and screen out target users is an urgent problem that needs to be solved. Summary of the invention
[0005] The present application provides a user reputation evaluation method based on a scoring card model, which can accurately evaluate the user's reputation based on the scoring card model.
[0006] In order to achieve the above object, the present invention adopts the following technical scheme:
[0007] In a first aspect, the present invention provides a user reputation evaluation method based on a scoring card model, the method comprising:
[0008] Collect the communication data and basic information data of old users to obtain a data set;
[0009] Labeling the data set with positive and negative samples according to preset features to obtain a sample set;
[0010] Building a prediction model based on the LightGBM model, using the sample set to train the LightGBM model, performing parameter tuning and parameter selection on the prediction model according to KS and AUC values, and obtaining a trained prediction model;
[0011] Collect communication data and basic information data of users to be predicted;
[0012] Based on the communication data and basic data of the user to be predicted, using the prediction model to predict the probability that the user to be predicted is a target user;
[0013] Constructing a user reputation scoring model based on the scoring card model, using linear transformation to establish a relationship between the probability and the user reputation scoring model, and obtaining a reputation evaluation value of the user to be predicted;
[0014] Determine whether the user to be predicted is a target user according to the reputation evaluation value of the user to be predicted.
[0015] In a preferred example of the present application, it can be further configured that the communication data includes VOLTE registration signaling table, call signaling data and Internet bill data, and the basic information data includes user package value, integrated package information, ID card information and household registration place information.
[0016] In a preferred example of the present application, it can be further set that the preset features include whether the terminal is changed, whether the permanent residence is changed, whether there is a long period of disconnection from the network, whether there is normal Internet traffic, the overlap rate between the called number on the day and the historical called number, whether the monthly call volume is greater than 50 times, the value of the user package, whether the package is integrated, whether the ID card is from another place, and whether the registered residence is in a rural area.
[0017] In a preferred example of the present application, it can be further configured that the prediction model is subjected to parameter tuning and parameter selection according to the KS and AUC values to obtain a trained prediction model, including:
[0018] Set basic parameters including the number of trees, depth, learning rate, and number of leaf nodes;
[0019] The number of leaf nodes and the weight ratio of multiple classes in the basic parameters are adjusted to obtain multiple KS values and AUC values;
[0020] The parameters of the prediction model and the values corresponding to the parameters are determined based on the KS value and the AUC value to obtain a trained prediction model.
[0021] In a preferred example of the present application, it can be further configured that the relationship between the probability and the user reputation scoring model is established by using linear transformation to obtain the reputation evaluation value of the user to be predicted, including:
[0022] Assume that odds is the ratio between the probability of non-target users P and the probability of target users 1-P, and score is the reputation evaluation value, then:
[0023]
[0024] score = A + B × ln (odds);
[0025] Where A and B are constants, and the values of A and B can be obtained by linear transformation, and the linear transformation is:
[0026]
[0027] Among them, p0 is the score corresponding to the predicted probability of x, and PDO is the score that the scorecard should increase when the predicted probability doubles to 2x;
[0028] Solving the simultaneous equations, we can obtain:
[0029]
[0030] After determining the values of A and B, the score is calculated.
[0031] In a preferred example of the present application, it can be further configured that the determining whether the user to be predicted is a target user according to the reputation evaluation value of the user to be predicted includes:
[0032] When the reputation evaluation value of the user to be predicted is above 650 points, the user to be predicted is determined as a target user.
[0033] In a second aspect, the present application provides a user reputation evaluation device based on a scoring card model, the device comprising:
[0034] The data collection module is used to collect the communication data and basic information data of old users to obtain a data set; and to label the data set with positive and negative samples according to preset features to obtain a sample set;
[0035] A model training module is used to build a prediction model based on the LightGBM model, use the sample set to train the LightGBM model, perform parameter tuning and parameter selection on the prediction model according to KS and AUC values, and obtain a trained prediction model;
[0036] A reputation evaluation value module is used to construct a user reputation scoring model based on the scoring card model, use linear transformation to establish the relationship between the probability and the user reputation scoring model, and obtain the reputation evaluation value of the user to be predicted;
[0037] The target user module is used to determine whether the user to be predicted is a target user according to the reputation evaluation value of the user to be predicted.
[0038] In a third aspect, the present application provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the method for evaluating user credibility based on the scoring card model as described in any one of the above items are implemented.
[0039] In a fourth aspect, the present application provides a computer-readable storage medium having a program stored thereon, wherein when the program is executed by a processor, a user credibility evaluation method based on a scoring card model as described in any one of the above items is implemented.
[0040] In a fifth aspect, the present application provides a computer program product, comprising computer instructions, which, when executed by a processor, implement the steps of the user credibility evaluation method based on the scoring card model as described in any one of the above items.
[0041] In summary, compared with the prior art, the technical solution provided in the embodiment of the present application has at least the following beneficial effects:
[0042] 1. Compared with traditional user credibility assessment methods, the features, thresholds, weights, etc. used in this application to calculate the user's credibility score are all learned using the LightGBM model, which is objective, avoids the limitations brought by human subjective cognition, and greatly improves the accuracy of user credibility ratings.
[0043] 2. Traditional user credibility assessment methods often use the characteristics of non-target users for feature engineering construction, but many misjudged target users also happen to have some non-target user characteristics. This application deeply explores the difference between the target numbers that are mistakenly hit and the non-target numbers, and constructs a precise and comprehensive target user feature engineering, which helps to accurately distinguish the target numbers from suspected non-target numbers.
[0044] 3. Apply the mature credit scoring card model in the financial field to the field of communication technology, score communication users, support grading treatment of users, and greatly improve the accuracy of fraud management. BRIEF DESCRIPTION OF THE DRAWINGS
[0045] Figure 1 A flowchart of a method for evaluating user credibility based on a scoring card model is provided for one embodiment of the present application.
[0046] Figure 2 A structural diagram of a user credibility evaluation device based on a scoring card model provided in one embodiment of the present application. DETAILED DESCRIPTION
[0047] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0048] In one embodiment of the present application, a user reputation evaluation method based on a scoring card model is provided. Figure 1 As shown, the method includes:
[0049] S100: Collect communication data and basic information data of old users to obtain a data set.
[0050] Specifically, the communication data includes:
[0051] 1. VOLTE registration signaling, key fields are shown in Table 1:
[0052]
[0053] 2. Call signaling data, key fields are shown in Table 2:
[0054]
[0055] 3. Internet call bill data, key fields are shown in Table 3:
[0056]
[0057] 4. User basic information data.
[0058] It includes the value of the user’s package, whether it is a integrated package, whether the user has an out-of-town ID card, whether the registered residence is in the rural area, etc.
[0059] The basic information data includes the value of the user's package, whether it is a fusion package, whether the user has an out-of-town ID card, whether the registered residence is in the rural area, etc.
[0060] S200: labeling the data set with positive and negative samples according to preset features to obtain a sample set.
[0061] The preset features include 10 features as shown in Table 4:
[0062]
[0063] Then, the data is labeled with positive and negative samples. The selected data samples are the numbers that have been shut down by the anti-fraud model in a certain province on a daily basis, and the numbers that have successfully signed the anti-fraud agreement and have been restored to the network for more than 1 year are used as the source of non-fraud old users. The numbers of old users with a network access time of more than 1 year in the public security work orders involved in the case are used as the source of fraud users. A comparative analysis is conducted on the characteristics of the two types of old users. Among them, there are 32,250 non-fraud users and 6,560 fraud users, totaling 38,810 number samples, of which the fraud users are the non-target users, and the non-fraud users are the target users. The data set is divided into two parts: a training set (80%) and a test set (20%). The training set is used for model training and parameter adjustment, and the test set is used for the final model evaluation. At the same time, some features are binarized, with 0 representing no and 1 representing yes. The data set samples are shown in Table 5:
[0064]
[0065] S300: Build a prediction model based on the LightGBM model, use the sample set to train the LightGBM model, perform parameter tuning and parameter selection on the prediction model according to KS and AUC values, and obtain a trained prediction model.
[0066] Specifically, the LightGBM model is a decision tree-based gradient boosting algorithm released by Microsoft in 2016. It has the characteristics of supporting parallel processing, fast training speed, high accuracy, and small memory usage. It is an optimized version of the gradient boosting tree algorithm. Before tuning the model parameters, KS and AUC values are determined as evaluation indicators of the prediction model.
[0067] The specific parameter adjustment and selection process of the prediction model is as follows:
[0068] 1. Initial parameter setting
[0069] First, a set of basic parameters were set for preliminary training, including n_estimators = 1000 (the number of trees), max_depth = 5 (the maximum depth of the tree), learning_rate = 0.27 (learning rate), and num_leaves = 61 (the number of leaf nodes). The initial selection of these parameters was based on previous experience and the results of preliminary experiments.
[0070] 2. Parameter Optimization
[0071] Adjust num_leaves and class_weights: Experiment with the number of num_leaves and different class_weights to explore the sensitivity of the model to imbalanced classes. For example, we tried num_leaves of 128 and different class weight ratios (such as [1,150] to [1,300]) to observe the changes in AUC and KS values.
[0072] Adjust learning_rate and max_depth: Find the optimal model complexity and learning speed by fine-tuning these key parameters. In some configurations, adjust learning_rate slightly (such as from 0.3 to 0.4 or 0.2) and test the impact on model performance.
[0073] Model evaluation: After each parameter adjustment, the same training set and validation set are used to evaluate the model's AUC and KS values. These indicators can indicate the model effect under different parameter configurations.
[0074] 3. Final parameter determination
[0075] After multiple rounds of parameter adjustment and model evaluation, it was found that when n_estimators = 800, max_depth = 5, learning_rate = 0.27, num_leaves = 64, and class_weights = [1,300], the model achieved a relatively balanced optimal result in terms of AUC and KS values (AUC value 0.84, KS value 0.63). As shown in Table 7:
[0076] 1 2 3 4 5 6 7 8 n_estimators 1000 1000 1000 800 800 1000 1000 1000 max_depth 5 5 5 5 5 6 6 5 learning_rate 0.27 0.27 0.27 0.27 0.27 0.3 0.25 0.27 num_leaves 61 128 128 128 64 128 64 61 class_weights [1,150] [1,150] [1,200] [1,300] [1,300] [1,300] [1,300] [1,300] min_child_samples 100 100 100 100 100 100 100 100 min_child_weight 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 feature_fraction 1 1 1 1 1 1 1 1 verbose 0 0 0 0 0 0 0 0 random_state 2024 2024 2024 2024 2024 2024 2024 2024 AUC value 0.70 0.73 0.77 0.80 0.84 0.72 0.80 0.78 KS value 0.39 0.52 0.41 0.44 0.63 0.45 0.60 0.43
[0077] S400: Collecting communication data and basic profile data of the user to be predicted; and using the prediction model to predict the probability that the user to be predicted is a target user based on the communication data and basic profile data of the user to be predicted.
[0078] Specifically, the output results of the prediction model are shown in Table 8:
[0079] Number The probability of being a normal user 159XXXXXX87 84.53％ 159XXXXXX45 73.62％ 158XXXXXX31 36.36％ 143XXXXXX55 23.56％ 157XXXXXX66 19.61％
[0080] S500: constructing a user reputation scoring model based on the scoring card model, using linear transformation to establish a relationship between the probability and the user reputation scoring model, and obtaining a reputation evaluation value of the user to be predicted.
[0081] Specifically, the scorecard model is the internationally accepted FICO scoring system, which comprehensively evaluates the credit status of individuals in multiple dimensions and ultimately gives a score ranging from 350 to 950. Similarly, establishing a user credit scorecard model can provide great help for the hierarchical handling of fraud management.
[0082] A linear transformation is used to establish a relationship between the probability and the user reputation scoring model, and the specific process includes:
[0083] To establish a user reputation scoring model, we first use a machine learning model to predict the probability of whether a user is a normal user. After obtaining the probability of a normal user, we linearly map the probability estimate to an intuitive and easy-to-understand user reputation score, that is, through a linear transformation, we establish a corresponding relationship between the odds in the above formula and the final model score. Among them, odds is the ratio between the probability of a fraud user P and the probability of a normal user 1-P, that is:
[0084]
[0085] The score of the scorecard, that is, the reputation evaluation value of the user to be predicted, can be represented by an expression containing odds, which is defined as follows:
[0086] score = A + B × ln(odds) (2);
[0087] A and B are constants that can be calculated using the score scale that needs to be set. First, determine two values in advance: first, the corresponding score p0 when a certain probability is x; second, the score PDO that the score card needs to increase when the probability doubles to 2x. Substituting the above two values into the equation yields:
[0088]
[0089] Solving the simultaneous equations, we can obtain:
[0090]
[0091] Afterwards, the values of A and B can be calculated according to the preset requirements through linear transformation as shown in the above formula. Finally, by substituting the prediction result of the machine learning model into equation (2), the probability that the user is a normal user calculated by the prediction model can be converted into a simple and intuitive scoring card model. The normal user is the target user.
[0092] S600: Determine whether the user to be predicted is a target user according to the reputation evaluation value of the user to be predicted.
[0093] Specifically, the user is divided into different risk levels according to the reputation evaluation value of the user to be predicted, for example, a score higher than 650 is low risk, a score between 500 and 650 is medium risk, and a score lower than 500 is high risk.
[0094] Then, different handling methods are adopted for users with different risk levels. For example, high-risk users are not allowed to resume online service after being shut down, low-risk users are filtered to avoid shutdown, and medium-risk users are allowed to resume online service once, etc., so as to reduce the pressure on customer service to handle complaints, achieve refined management and control of fraud governance, and seek a balance between telecommunications anti-fraud governance and user complaints.
[0095] In this embodiment, the beneficial effects include:
[0096] 1. Compared with traditional user credibility assessment methods, the features, thresholds, weights, etc. used in this application to calculate the user's credibility score are all learned using the LightGBM model, which is objective, avoids the limitations brought by human subjective cognition, and greatly improves the accuracy of user credibility ratings.
[0097] 2. Traditional user credibility assessment methods often use the characteristics of non-target users for feature engineering construction, but many misjudged target users also happen to have some non-target user characteristics. This application deeply explores the difference between the target numbers that are mistakenly hit and the non-target numbers, and constructs a precise and comprehensive target user feature engineering, which helps to accurately distinguish the target numbers from suspected non-target numbers.
[0098] 3. Apply the mature credit scoring card model in the financial field to the field of communication technology, score the users of communication numbers, support user classification and treatment, and greatly improve the accuracy of fraud management.
[0099] In some embodiments, the communication data includes VOLTE registration signaling table, call signaling data and Internet bill data, and the basic information data includes user package value, integrated package information, ID card information and household registration information.
[0100] In this embodiment, by using comprehensive and multi-dimensional sample data, the accuracy of the predicted user reputation score is improved.
[0101] In some embodiments, the preset features include whether the terminal is changed, whether the permanent residence is changed, whether there is a long period of disconnection from the network, whether there is normal Internet traffic, the overlap rate between the called number on the day and the historical called number, whether the monthly call volume is greater than 50 times, the value of the user package, whether the package is integrated, whether the identity card is from another place, and whether the registered residence is in a rural area.
[0102] In this embodiment, by selecting comprehensive and multi-dimensional features, the accuracy of the predicted user reputation score is improved.
[0103] In some embodiments, the prediction model is subjected to parameter tuning and parameter selection according to the KS and AUC values to obtain a trained prediction model, including:
[0104] Set basic parameters including the number of trees, depth, learning rate, and number of leaf nodes;
[0105] The number of leaf nodes and the weight ratio of multiple classes in the basic parameters are adjusted to obtain multiple KS values and AUC values;
[0106] The parameters of the prediction model and the values corresponding to the parameters are determined based on the KS value and the AUC value to obtain a trained prediction model.
[0107] In this embodiment, the KS value and AUC value are used to evaluate the training effect of the model, thereby improving the ability of the pre-trained prediction model to distinguish between positive and negative samples.
[0108] In some embodiments, determining whether the user to be predicted is a target user according to the reputation evaluation value of the user to be predicted includes:
[0109] When the reputation evaluation value of the user to be predicted is above 650 points, the user to be predicted is determined as a target user.
[0110] In this embodiment, the probability of misjudging the user to be predicted as the target user is reduced.
[0111] This application also provides a user reputation evaluation device based on a scoring card model. Figure 2 As shown, the device comprises:
[0112] The data collection module 100 is used to collect the communication data and basic information data of old users to obtain a data set; and to label the data set with positive and negative samples according to preset features to obtain a sample set;
[0113] The model training module 200 is used to build a prediction model based on the LightGBM model, use the sample set to train the LightGBM model, perform parameter tuning and parameter selection on the prediction model according to the KS and AUC values, and obtain a trained prediction model;
[0114] Prediction module 300, used to collect communication data and basic data of the user to be predicted; based on the communication data and basic data of the user to be predicted, use the prediction model to predict the probability that the user to be predicted is the target user;
[0115] The reputation evaluation value module 400 is used to construct a user reputation scoring model based on the scoring card model, use linear transformation to establish the relationship between the probability and the user reputation scoring model, and obtain the reputation evaluation value of the user to be predicted;
[0116] The target user module 500 is used to determine whether the user to be predicted is a target user according to the reputation evaluation value of the user to be predicted.
[0117] The functional implementation of each module in the above-mentioned user credibility evaluation device based on the scoring card model corresponds to the steps in the above-mentioned user credibility evaluation method embodiment based on the scoring card model, and its functions and implementation processes are not repeated here one by one.
[0118] The present application also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the method for evaluating user credibility based on the scoring card model as described in any of the above embodiments are implemented.
[0119] The present application also provides a computer-readable storage medium, on which a program is stored, wherein the computer-readable storage medium refers to a carrier for storing data, which may include but is not limited to a floppy disk, an optical disk, a hard disk, a flash memory, a USB flash drive, and / or a memory stick, etc., and the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The working process, working details, and technical effects of the computer-readable storage medium provided in this embodiment can be found in the above embodiment of a user reputation evaluation method based on a scoring card model, which will not be repeated here.
[0120] The application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the user reputation evaluation method based on the scoring card model as described in any of the above embodiments.
[0121] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
[0122] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification. The above-mentioned embodiments only express several implementation methods of the present application, and their descriptions are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, without departing from the concept of the present application, several deformations and improvements can be made, which all belong to the protection scope of the present application. Therefore, the protection scope of the patent of this application shall be based on the attached claims.
Claims
1. A user reputation evaluation method based on a scoring card model, characterized in that: include: Collect the communication data and basic information data of old users to obtain a data set; Labeling the data set with positive and negative samples according to preset features to obtain a sample set; Building a prediction model based on the LightGBM model, using the sample set to train the LightGBM model, performing parameter tuning and parameter selection on the prediction model according to KS and AUC values, and obtaining a trained prediction model; Collect communication data and basic information data of users to be predicted; Based on the communication data and basic data of the user to be predicted, using the prediction model to predict the probability that the user to be predicted is a target user; Constructing a user reputation scoring model based on the scoring card model, using linear transformation to establish a relationship between the probability and the user reputation scoring model, and obtaining a reputation evaluation value of the user to be predicted; Determine whether the user to be predicted is a target user according to the reputation evaluation value of the user to be predicted.
2. The user reputation evaluation method based on the scoring card model according to claim 1 is characterized in that: The communication data includes VOLTE registration signaling table, call signaling data and Internet bill data, and the basic information data includes user package value, integrated package information, ID card information and household registration information.
3. The user reputation evaluation method based on the scoring card model according to claim 2 is characterized in that: The preset features include whether the terminal is changed, whether the permanent residence is changed, whether there is a long period of disconnection from the network, whether there is normal Internet traffic, the overlap rate between the called number on the day and the historical called number, whether the monthly call volume is greater than 50 times, the value of the user package, whether the package is integrated, whether the ID card is from another place, and whether the registered residence is in the rural area.
4. The user reputation evaluation method based on the scoring card model according to claim 3 is characterized in that: The method of performing parameter tuning and parameter selection on the prediction model according to the KS and AUC values to obtain a trained prediction model includes: Set basic parameters including the number of trees, depth, learning rate, and number of leaf nodes; The number of leaf nodes and the weight ratio of multiple classes in the basic parameters are adjusted to obtain multiple KS values and AUC values; The parameters of the prediction model and the values corresponding to the parameters are determined based on the KS value and the AUC value to obtain a trained prediction model.
5. The user reputation evaluation method based on the scoring card model according to claim 4 is characterized in that: The use of linear transformation to establish the relationship between the probability and the user reputation scoring model to obtain the reputation evaluation value of the user to be predicted includes: Assume that odds is the ratio between the probability of non-target users P and the probability of target users 1-P, and score is the reputation evaluation value, then: score = A + B × ln (odds); Where A and B are constants, and the values of A and B can be obtained by linear transformation, and the linear transformation is: Among them, p0 is the score corresponding to the predicted probability of x, and PDO is the score that the scorecard should increase when the predicted probability doubles to 2x; Solving the simultaneous equations, we can obtain: After determining the values of A and B, the score is calculated.
6. The user reputation evaluation method based on the scoring card model according to claim 5 is characterized in that: The determining whether the user to be predicted is a target user according to the reputation evaluation value of the user to be predicted includes: When the reputation evaluation value of the user to be predicted is above 650 points, the user to be predicted is determined as a target user.
7. A user reputation evaluation device based on a scoring card model, characterized in that: include: The data collection module is used to collect the communication data and basic information data of old users to obtain a data set; Labeling the data set with positive and negative samples according to preset features to obtain a sample set; A model training module is used to build a prediction model based on the LightGBM model, use the sample set to train the LightGBM model, perform parameter tuning and parameter selection on the prediction model according to KS and AUC values, and obtain a trained prediction model; A prediction module, used to collect communication data and basic information data of the user to be predicted; based on the communication data and basic information data of the user to be predicted, use the prediction model to predict the probability that the user to be predicted is the target user; A reputation evaluation value module is used to construct a user reputation scoring model based on the scoring card model, use linear transformation to establish the relationship between the probability and the user reputation scoring model, and obtain the reputation evaluation value of the user to be predicted; The target user module is used to determine whether the user to be predicted is a target user according to the reputation evaluation value of the user to be predicted.
8. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method for evaluating user credibility based on a scoring card model according to any one of claims 1 to 6 are implemented.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program, wherein when the program is executed by the processor, the user credibility evaluation method based on the scoring card model according to any one of claims 1 to 6 is implemented.
10. A computer program product comprising computer instructions, characterized in that: When executed by a processor, the computer instructions implement the steps of the user credibility evaluation method based on the scoring card model described in claims 1 to 6.