Loss-of-contact mode confirmation method, electronic equipment, storage medium and product

Abstract

The invention discloses a contact loss mode confirmation method, electronic equipment, a storage medium and a product, and relates to the technical field of mode classification, and the method comprises the steps: extracting four-dimensional correlation features of a contact loss customer from each preset data source, and forming contact loss data; inputting each acquisition index under the four-dimensional association feature in the lost data into a preset lost mode classification model to obtain a model classification result; and according to the model classification result, determining the lost contact mode information of the lost contact customer, the lost contact mode being the lost contact type of the lost contact customer. The problem that it is difficult to determine the lost mode of the lost customer is solved.

Description

Technical Field

[0001] The present application relates to the technical field of pattern classification, and particularly to a method, apparatus, electronic device, storage medium and computer program product for confirming a lost contact mode. Background Art

[0002] In the financial field, financial institutions such as banks, microloan companies and financial platforms face certain credit risks in loan business, including the risk that borrowers may be unable to repay loans on time or completely lose contact for various reasons. For example, in February 2019, the Internet Finance Association of Beijing disclosed the information of 300 defaulting and lost contact customers of 5 financial platforms. The results showed that only 66 people were not lost contact, and the total lost contact ratio was as high as 78%. Due to the lack of effective means to accurately identify the lost contact mode, financial institutions often have difficulty in judging the specific lost contact situation of customers, resulting in the inability to determine which strategy to adopt to recover loans. This not only increases the difficulty of recovering loans, but also seriously affects the daily operation and capital flow of financial institutions. More seriously, since it is difficult to find these lost contact loan customers in a short time, the credit risk of financial institutions is further exacerbated. Therefore, there is a problem that financial institutions currently have difficulty in determining the lost contact mode of lost contact customers. Summary of the Invention

[0003] The main purpose of the present application is to provide a method, apparatus, electronic device, storage medium and computer program product for confirming a lost contact mode, aiming to solve the technical problem of difficulty in determining the lost contact mode of lost contact customers.

[0004] To achieve the above object, the present application proposes a method for confirming a lost contact mode, and the method for confirming a lost contact mode includes:

[0005] Extract four-dimensional correlation features of lost contact customers from each preset data source to form lost contact data;

[0006] Input each acquisition index under the four-dimensional correlation feature in the lost contact data into a preset lost contact mode classification model to obtain a model classification result;

[0007] According to the model classification result, determine the lost contact mode information of the lost contact customer, where the lost contact mode is the lost contact type of the lost contact customer.

[0008] In an embodiment, before the step of inputting each acquisition index under the four-dimensional correlation feature in the lost contact data into a preset lost contact mode classification model, the following steps are further included:

[0009] Use the historical lost contact data set of historical lost contact customers as a sample set, and the sample set includes a training set;

[0010] Divide the training set into a sub-training set and a sub-test set, and input the sub-training set into the missing connection mode classification model to be constructed for iterative training of the model, and perform classification verification on the missing connection mode classification model to be constructed based on the sub-test set;

[0011] In the case where the classification verification result of the missing connection mode classification model to be constructed does not meet the preset training conditions, calculate the loss function value of each round of training, and adjust the model parameters in the missing connection mode classification model to be constructed based on the loss function value;

[0012] In the case where the classification verification result of the trained missing connection mode classification model to be constructed meets the preset training conditions, use the trained missing connection mode classification model to be constructed as the missing connection mode classification model.

[0013] In one embodiment, the step of performing classification verification on the missing connection mode classification model to be constructed based on the sub-test set includes:

[0014] Calculate the accuracy, precision, recall rate, and F1 score of the trained missing connection mode classification model to be constructed based on the sub-test set;

[0015] In the case where the accuracy rate reaches the preset accuracy rate threshold, the precision rate reaches the preset precision rate threshold, the recall rate reaches the preset recall rate threshold, and the F1 score reaches the preset F1 score threshold, determine that the classification verification result of the trained missing connection mode classification model to be constructed meets the preset training conditions.

[0016] In one embodiment, after the step of inputting the sub-training set into the missing connection mode classification model to be constructed for iterative training of the model, it includes:

[0017] Calculate the average gain of each candidate sample feature of the samples in the sample set;

[0018] Use the candidate sample features with an average gain higher than the preset gain threshold as the acquisition indicators for missing connection data.

[0019] In one embodiment, the step of determining the missing connection mode information of the missing connection customer according to the model classification result includes:

[0020] In response to the user's mode selection operation, in the case where the mode selection operation is to turn off the rule assistance mode, use the model classification result as the missing connection mode information of the missing connection customer;

[0021] In the case where the mode selection operation is to turn on the rule assistance mode, determine and output the missing connection mode information of the missing connection customer based on the preset mode determination assistance rule and the model classification result.

[0022] In one embodiment, the step of determining the disconnection mode information of the disconnected customer based on the preset mode determination auxiliary rule and the model classification result and outputting it includes:

[0023] Determine whether the model classification result is consistent with the rule auxiliary result obtained based on the preset mode determination auxiliary rule;

[0024] When the model classification result is consistent with the rule auxiliary result, output the model classification result or the rule auxiliary result as the disconnection mode information of the disconnected customer;

[0025] When the model classification result is inconsistent with the rule auxiliary result, use the model classification result as the model prediction result, use the rule auxiliary result as the empirical prediction result, and output the model prediction result and the empirical prediction result as the disconnection mode information of the disconnected customer.

[0026] In one embodiment, before the step of determining whether the rule classification result is consistent with the auxiliary result obtained based on the preset mode determination auxiliary rule, the following steps are further included:

[0027] Compare the disconnection data of the disconnected customer with the index thresholds in the preset mode determination rule;

[0028] Record the index thresholds satisfied by the disconnection data of the disconnected customer, determine the corresponding disconnection mode of the disconnected customer based on the satisfied index thresholds, and use the corresponding disconnection mode as the rule auxiliary result.

[0029] In addition, to achieve the above object, the present application also proposes a disconnection mode confirmation device, and the disconnection mode confirmation device includes:

[0030] A data acquisition module for extracting the four-dimensional correlation features of the disconnected customer from each preset data source to form disconnection data;

[0031] A model input module for inputting each acquisition index under the four-dimensional correlation features in the disconnection data into a preset disconnection mode classification model to obtain a model classification result;

[0032] A result confirmation module for determining the disconnection mode information of the disconnected customer according to the model classification result, where the disconnection mode is the disconnection type of the disconnected customer.

[0033] In addition, to achieve the above object, the present application also proposes an electronic device, and the device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the disconnection mode confirmation method as described above.

[0034] In addition, to achieve the above object, the present application further provides 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, the steps of the disconnection mode confirmation method described above are implemented.

[0035] In addition, to achieve the above object, the present application further provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the disconnection mode confirmation method described above are implemented.

[0036] The present application provides a disconnection mode confirmation method, which includes: extracting four-dimensional association features of disconnected customers from each preset data source to form disconnected data; inputting each collection index under the four-dimensional association features in the disconnected data into a preset disconnection mode classification model to obtain a model classification result; and determining the disconnection mode information of the disconnected customers according to the model classification result, where the disconnection mode is the disconnection type of the disconnected customers.

[0037] By extracting relevant data of disconnected customers from each preset data source, the present application ensures the accuracy and reliability of the analysis, provides strong data support for the determination of the disconnection mode, and accurately predicts the disconnection mode of disconnected customers through the disconnection mode classification model by using the laws and patterns in historical data, improving the accuracy and efficiency of the disconnection mode determination. Financial institutions can adopt more targeted recovery strategies according to the disconnection mode information, and can also discover problems earlier and take actions, effectively reducing loan losses. Compared with related solutions that often rely on manual judgment or simple rule matching, which are not only inefficient but also difficult to accurately identify complex disconnection modes, the present application automatically processes the disconnected data of disconnected persons through the disconnection mode classification model to obtain the disconnection mode of disconnected customers, improving the processing efficiency and response speed, and providing a basis for subsequent loan recovery, ensuring the daily operation and capital flow of financial institutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The accompanying drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application.

[0039] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, other drawings can be obtained according to these drawings without creative efforts.

[0040] Figure 1 It is a schematic flowchart provided for Embodiment 1 of the disconnection mode confirmation method of the present application;

[0041] Figure 2 Flow chart for constructing the missing mode classification model provided for the missing mode confirmation method of this application;

[0042] Figure 3 Distribution diagram of feature importance ranking provided for the missing mode confirmation method of this application;

[0043] Figure 4 Schematic flow diagram provided for the second embodiment of the missing mode confirmation method of this application;

[0044] Figure 5 Feature map corresponding to the hide-and-seek mode provided for the missing mode confirmation method of this application;

[0045] Figure 6 Feature map corresponding to the absconding with funds mode provided for the missing mode confirmation method of this application;

[0046] Figure 7 Feature map corresponding to the fake disappearance mode provided for the missing mode confirmation method of this application;

[0047] Figure 8 Schematic diagram of the module structure of the missing mode confirmation device in the embodiment of this application;

[0048] Figure 9 Schematic diagram of the device structure of the hardware operating environment involved in the missing mode confirmation method in the embodiment of this application.

[0049] The realization of the purpose, functional features and advantages of this application will be further described in conjunction with the embodiments with reference to the accompanying drawings. Detailed implementation manners

[0050] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of this application and are not used to limit this application.

[0051] For a better understanding of the technical solutions of this application, the following will be described in detail in conjunction with the accompanying drawings of the specification and the specific implementation manners.

[0052] The main solution of the embodiment of this application is: extracting the four-dimensional correlation features of the missing customers from each preset data source to form missing data; inputting each acquisition index under the four-dimensional correlation features in the missing data into a preset missing mode classification model to obtain a model classification result; and determining the missing mode information of the missing customers according to the model classification result, where the missing mode is the missing type of the missing customers.

[0053] In this embodiment, for the convenience of description, the following will be described with the missing mode determination system as the execution subject.

[0054] In the financial field, especially financial institutions such as banks, micro - loan companies, and financial platforms, they face the credit risk that borrowers may become out of contact. Currently, after a financial institution determines that a customer is out of contact, it often lacks effective means to accurately identify their out - of - contact patterns, resulting in increased difficulty in recovering loans, affecting daily operations and capital flow, and exacerbating credit risks.

[0055] This application provides a solution. By training a out - of - contact pattern classification model using a historical out - of - contact dataset, it can automatically learn and identify the rules and patterns in historical data, thereby more accurately predicting the out - of - contact patterns of future out - of - contact customers, assisting financial institutions in clarifying the out - of - contact types of out - of - contact customers, so as to formulate corresponding strategies to recover loans and ensure the daily operations and capital flow of financial institutions.

[0056] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication, and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device, an out - of - contact pattern determination system, etc. that can implement the above functions. Hereinafter, taking the out - of - contact pattern determination system as an example, this embodiment and the following embodiments will be described.

[0057] Based on this, the embodiment of this application provides a method for confirming out - of - contact patterns, referring to Figure 1 , Figure 1 which is a schematic flowchart of the first embodiment of the out - of - contact pattern confirmation method of this application.

[0058] In this embodiment, the out - of - contact pattern confirmation method includes steps S01 - S03:

[0059] Step S01: Extract four - dimensional correlation features of out - of - contact customers from each preset data source to form out - of - contact data;

[0060] It should be noted that the system extracts four - dimensional correlation features related to out - of - contact customers from each preset data source and uses them as the out - of - contact data of out - of - contact customers. Among them, each preset data source includes but is not limited to data providers such as the databases of financial institutions, authorized communication operators, and authorized third - party application programs (APPs). The out - of - contact data collected in the preset data source can only be collected after being authorized by the data provider. An out - of - contact customer refers to a customer who fails to repay debts or has not fully repaid debts and loses contact during the project period of a financial institution.

[0061] In addition, it should be noted that the four-dimensional correlation features include four dimensions: reach degree, loan situation, duration of being out of contact, and social relationship. Under each dimension, there are multiple collection indicators. Among them, under the dimension of reach degree, the permanent address of the out-of-contact customer, the online status of the mobile phone number, the call record of the mobile phone number, the SMS signaling characteristics, the email characteristics, other signaling characteristics, the online status of the emergency contact's mobile phone number, the call record of the emergency contact's mobile phone number, the SMS signaling characteristics of the emergency contact, and the email characteristics of the emergency contact need to be considered; under the dimension of loan situation, the loan type, overdue days, overdue loan, and estimated recoverable amount of the out-of-contact customer need to be considered; under the dimension of duration of being out of contact, it refers to the time difference between the time when the financial institution or relevant personnel find that they cannot contact the out-of-contact customer with the contact information reserved by the out-of-contact customer and the current inspection time, reflecting the length of time the out-of-contact customer has been out of contact; under the dimension of social relationship, the number of effective contacts of the out-of-contact customer and the relationship between the out-of-contact customer and the effective contacts need to be considered.

[0062] Specifically, for the collection indicators under the dimension of reach degree, it includes the following:

[0063] The permanent address is the characteristic reflected by the permanent address information reserved by the out-of-contact customer when handling loan business at the financial institution. When applying for loan business, some out-of-contact customers may fill in the address of the rented house, hotel, motel, or other temporary address as the reserved address. For these address information, it can be judged whether the permanent address is the household registration address or business address through the household registration address information on the out-of-contact customer's ID card, or the domicile information on the business license, etc. Let x1 represent the permanent address of the out-of-contact customer, then the value of the permanent address characteristic is 0 or 1. When x1 takes the value of 1, it means that the permanent address of the out-of-contact customer belongs to the household registration address or business address, and when x1 takes the value of 0, it means it is a temporary address.

[0064] The online status of the mobile phone number is the status characteristic of the mobile phone number reserved by the out-of-contact customer when handling loan business at the financial institution in the telecommunications operator's network system during the inspection period (such as one month). Generally speaking, the mobile phone number has statuses such as normal, out of service, in the network but unavailable, invalid number, not activated, abnormal, etc. Assuming x2 represents the online status characteristic of the out-of-contact customer's mobile phone number, then x2 has six status types, as shown in the following formula:

[0065]

[0066] Among them, A0, A1, A2, A3, A4, A5 are the online status categories of the mobile phone number, representing the statuses of the mobile phone number being normal, out of service, in the network but unavailable, invalid number, not activated, abnormal respectively.

[0067] The call record of the mobile phone number is the number of times that the financial institution encounters the state of the other party's mobile phone being turned off or busy when calling the mobile phone number previously reserved by the lost contact customer within the preset investigation period (for example, one month). It should be noted that the fact that the financial institution is prompted with a busy tone or the phone being turned off when calling the mobile phone number of the loan lost contact customer proves that the mobile phone number of the lost contact customer is in a normal online state. It's just that due to certain reasons, the customer has lost contact with the financial institution. However, within the investigation period, if the busy tone or the phone being turned off reaches the preset number of times, it indicates that the loan lost contact customer has no intention of communicating with the financial institution. The call record of the mobile phone number x3 can be measured by the following equation.

[0068]

[0069] Among them, the value of Call τ (s) is 0 or 1. When Call τ (s) = 1, it means that the financial institution is prompted with a busy tone or the phone being turned off when calling the mobile phone number of the lost contact customer for the s-th time within τ time. Otherwise, Call τ (s) = 0, s = 1, 2,..., z, where z is an integer.

[0070] The short message signaling feature is the situation where the other party replies to the short message within the specified time (for example, 3 days) after the financial institution successfully sends a short message to the mobile phone number previously reserved by the lost contact customer within the preset investigation period. Since the short message has less impact on the customer's normal life and the customer's short message reading rate is higher compared to the mobile phone answering rate, the short message signaling feature is also one of the associated indicators for measuring the reach. When the short message signaling feature x4 of the lost contact customer takes the value of 1, it means that the loan lost contact customer has replied to the short message of the financial institution within the specified time. When x4 takes the value of 0, it means that the lost contact customer has not replied to the short message of the financial institution within the specified time.

[0071] The email feature is the situation where the lost contact customer replies to the email within the specified time after the financial institution successfully sends an email to the email address previously reserved by the lost contact customer within the preset investigation period. When the email feature x5 of the lost contact customer takes the value of 1, it means that the lost contact customer has replied to the email of the financial institution within the specified time. Otherwise, x5 takes the value of 0.

[0072] The other signaling feature x6 is the situation where the lost contact customer replies to the message within the specified time after the financial institution successfully sends a message to the lost contact customer through other media such as third-party applications (APPs) of smart phones, mini-programs built into instant messaging software, and official accounts of financial institutions, etc., in addition to calling the mobile phone number, sending short messages, and emails, within the preset investigation period.

[0073]

[0074] Among them, Sig(τ) is the total number of times the lending institution receives reply information from the lost-contact customer using other media within time τ, and Med τ (p) is the number of times the lost-contact customer uses the p-th medium to reply to the lending institution within time τ, where p = 1, 2, ……, θ, and θ is an integer. When Sig(τ) ≥ 1 and the value of x6 is 1, it means that the lost-contact loan customer has replied to the financial institution's message using other media within the specified time; otherwise, the value of x6 is 0.

[0075] The in-network status x7 of the emergency contact's mobile phone number is similar to the in-network status x2 of the mobile phone number, and there are also six status types: normal, out of service, in-network but unavailable, invalid number, not enabled, and abnormal, as shown in the following formula:

[0076]

[0077] Among them, A0, A1, A2, A3, A4, A5 are the in-network status categories of the mobile phone number, representing the statuses that the mobile phone number is normal, out of service, in-network but unavailable, invalid number, not enabled, and abnormal, respectively.

[0078] The call record of the emergency contact's mobile phone number is the number of times that the financial institution encounters the situation of the other party's mobile phone being turned off or busy when calling the emergency contact's mobile phone number of the lost-contact customer in an emergency. The value of the call record x8 of the emergency contact's mobile phone number is based on the following:

[0079]

[0080] Among them, Emergency_Call τ (d) represents the status that the financial institution is prompted with a busy tone or the phone is turned off when calling the mobile phone number of the lost-contact customer for the d-th time within τ time. The value of Emergency_Call τ (d) is 0 or 1. When the value of Emergency_call τ (d) is 1, it means that there is a prompt of a busy tone or the phone is turned off; otherwise, the value of Emergency_Call τ (d) is 0, where d = 1, 2, ……, v, and v is an integer.

[0081] The short message signaling feature of the emergency contact is the situation where the other party replies to the short message within the specified time after the financial institution successfully sends a short message to the emergency contact's mobile phone number of the lost-contact customer within the preset investigation time period. When the value of the short message signaling feature x9 of the emergency contact is 1, it means that the emergency contact of the lost-contact loan customer has replied to the financial institution's short message within the specified time; otherwise, the value of x9 is 0.

[0082] The emergency contact email feature refers to the situation where, after a financial institution successfully sends an email to the email address of the emergency contact of a lost-contact customer within a preset investigation period, the other party replies to the email within the specified time. When the emergency contact email feature x 10 takes the value of 1, it means that the emergency contact has replied to the financial institution's email within the specified time; otherwise, x 10 takes the value of 0.

[0083] In addition, for the collection indicators in the dimension of loan situation, they include the following:

[0084] The loan type is the result of a financial institution classifying the loan business of a lost-contact customer according to certain criteria. Referring to the classification method of bank loan types, according to different loan guarantee conditions, the loan types are divided into secured loans, mortgage loans, credit loans, discount loans, etc. The loan type x7 of the lost-contact customer includes:

[0085]

[0086] Among them, B0, B1, B2, and B3 are the loan types of the lost-contact customer, representing secured loans, mortgage loans, credit loans, and discount loans respectively.

[0087] The number of overdue days is closely related to the current date and the repayment date, and can be measured by calculating the difference between the current date and the repayment date. The number of overdue days x 12 The calculation formula is:

[0088] x 12 = M1 - M2

[0089] Among them, the measurement unit of x 12 is days, M1 is the current investigation date, and M2 is the repayment date specified by the financial institution.

[0090] The amount of overdue loans is the arrears of a lost-contact customer who fails to repay the loan on the contractually agreed date, which directly affects the motivation of the lost-contact customer to go missing. Referring to the calculation method of the bank, the indicators affecting overdue loans include the principal to be repaid, the number of overdue days, the loan interest rate, and the penalty interest. The calculation basis of the overdue loan x 13 is as follows:

[0091] x 13 = V + V×r + (V + U)×r×(M1 - M2) + G

[0092] Wherein, V is the principal to be repaid, r is the loan interest rate, U is the unpaid interest, G is the penalty interest, M1 is the current inspection date, M2 is the repayment date stipulated by the financial institution, M1 - M2 represents the number of overdue days. Since the number of overdue days is calculated in days, the loan interest rate r needs to be converted into a daily interest rate for actual calculation, and G is determined according to the contract agreement between the financial institution and the lost-contact customer.

[0093] The estimated recoverable amount is the total amount of debt repayment funds that the financial institution expects to recover from the collateral, discounted bills and other realizable items of the lost-contact customer, or the debt repayment funds that the guarantor can bear, etc. according to the contract agreement after the lost-contact customer defaults and loses contact. Among them, the collateral can be negotiable securities, or movable property, immovable property, etc. discounted bills, or bank bills, commercial bills, bonds, treasury bills, etc. Referring to the calculation method of the bank, the estimated recoverable amount x 14 The calculation formula is as follows:

[0094]

[0095] Wherein, Re_amount τ (q) represents the q-th estimated recoverable amount of the lost-contact user estimated by the financial institution within τ time, q = 1, 2,..., σ, σ is an integer, and the estimated recoverable amount of the credit loan is 0.

[0096] In addition, for the collection indicators in the dimension of lost-contact duration, it includes the following content:

[0097] The lost-contact duration is the time difference between the time when the financial institution or relevant personnel find that they cannot contact the lost-contact customer with the contact information reserved by the lost-contact customer and the current inspection time, which reflects the length of time the lost-contact customer has been out of contact. The next-level indicator of the lost-contact duration is measured by the number of lost-contact days, and the number of lost-contact days x 15 The calculation formula is as follows:

[0098] x 15 = M1 - M3

[0099] Wherein, M1 is the current inspection date, and M3 is the time when the financial institution or relevant personnel find that they cannot contact the lost-contact customer with the contact information reserved by the lost-contact customer.

[0100] In addition, for the collection indicators in the dimension of social relationship, it includes the following content:

[0101] The number of valid contacts is the number of non-empty and real existing mobile phone numbers in the mobile phone address book of the lost-contact customer within a preset inspection time period. The number of valid contacts x 16 Can be expressed as:

[0102] x 16 = |{N1, N2, ……, Nu}|

[0103] Among them, || represents the number of elements in the set, and u is an integer.

[0104] The relationship between the lost customer and the valid contact is used to determine whether the valid contact contains contacts such as parents, spouse, and direct relatives. The relationship between the lost customer and the valid contact is represented by x 17 is represented, and x 17 takes the value of 1, indicating that the valid contact contains contacts such as parents, spouse, and direct relatives. Otherwise, x 17 takes the value of 0. Then, assuming that the set of mobile phone numbers of the parents, spouse, and direct relatives of the lost customer is {Z1, Z2, ……, N e}, x 17 is calculated as follows:

[0105]

[0106] Among them, e is the number of direct relatives such as the parents of the lost customer. When the intersection of {N1, N2, ……, N u} and {Z1, Z2, ……, N e} is not empty, x 17 takes the value of 1. When the intersection of {N1, N2, ……, N u} and {Z1, Z2, ……, N e} is empty, x 17 takes the value of 0.

[0107] It can be understood that in step S01, by extracting information from multiple preset data sources, relevant information of lost customers can be comprehensively and systematically collected, avoiding information omission. At the same time, focusing on four-dimensional correlation features (reach degree, loan situation, duration of loss of contact, social relationship) helps to focus on core information, reduce data processing complexity, solve challenges such as data dispersion and incomplete information usually faced by financial institutions, effectively integrate data, and provide a solid foundation for subsequent analysis.

[0108] In step S02, each acquisition index under the four-dimensional correlation features in the lost data is input into a preset lost mode classification model to obtain a model classification result;

[0109] It should be noted that the system inputs each collection index under the four-dimensional correlation features in the lost contact data into the trained lost contact mode classification model. The lost contact mode classification model is a model based on machine learning algorithms and trained using a large number of historical lost contact data sets for predicting the lost contact modes of lost contact customers. The historical correlation data set contains the characteristics of historical lost contact customers and the corresponding lost contact modes. By learning the relationship between the characteristics of lost contact customers and the lost contact modes in the historical lost contact data, the lost contact mode classification model can automatically identify the possible lost contact modes that the lost contact customers may belong to to determine the model classification result.

[0110] It can be understood that in step S02, using the preset lost contact mode classification model for automated analysis improves the processing efficiency and reduces manual intervention. Since the preset lost contact mode classification model is trained based on historical data and algorithms, it can more accurately identify different lost contact modes and improve the accuracy of classification.

[0111] In step S03, according to the model classification result, the lost contact mode information of the lost contact customer is determined, where the lost contact mode is the lost contact type of the lost contact customer.

[0112] It should be noted that based on the obtained model classification result, the lost contact mode information of the lost contact customer is generated. The lost contact mode is the specific lost contact type of the lost contact customer, which is the basic way for the lost contact customer to deliberately create the phenomenon of losing contact with some relevant people in order to avoid repaying the debts of the lending institution. The identification of the lost contact mode is to comprehensively analyze the lost contact characteristics of the loan lost contact customers, and then be able to distinguish the type of the lost contact mode.

[0113] In addition, it should be noted that this application defines three lost contact modes, including: the hide-and-seek mode, the absconding with funds mode, and the fake disappearance mode. In fact, there can be more than three types of lost contact modes divided.

[0114] Among them, the hide-and-seek mode is that the lost contact customer deliberately avoids repaying the debts of the financial institution and refuses to contact the relevant personnel of the financial institution by means such as not answering the phone, busy tone, shutting down, frequently changing the residential address, etc., resulting in the financial institution being unable to contact the lost contact customer through the mobile phone number, residential address, and emergency contact person reserved by it. This mode of the lost contact customer avoiding the financial institution is similar to the hide-and-seek game. On the one hand, the financial institution tries to contact and find the lost contact customer through various means, and on the other hand, the lost contact customer tries to avoid and hide from the reach of the financial institution through various means.

[0115] The absconding with funds mode means that after obtaining the loan, the lost contact customer unilaterally cuts off the contact with the financial institution and relevant personnel by means such as shutting down the phone, having a disconnected number, etc., and leaves the reserved permanent address and work unit with a large amount of loan to other regions, causing the financial institution to face huge default risks.

[0116] The false disappearance mode is a mode in which a missing customer intentionally evades repayment of the debt to the lending institution and creates an appearance of being untraceable and nowhere to be found for the lending institution and related personnel.

[0117] It can be understood that step S03 clarifies the disappearance mode of the missing customer through the model classification result, improves the refinement level of the management of missing customers, helps financial institutions formulate more targeted communication and recovery strategies for different disappearance modes, and reduces the losses of financial institutions.

[0118] In a feasible implementation manner, in step S02, before the step of inputting each acquisition index under the four-dimensional correlation feature in the missing data into the preset missing mode classification model, steps A01 to A04 are further included:

[0119] Step A01: Use the historical missing data set of historical missing customers as the sample set, and the sample set includes a training set;

[0120] It should be noted that all historical missing data sets are extracted from each preset data source, including personal information of missing customers (such as name, ID number, contact information, etc.), behavior records before disappearance (such as transaction records, customer service communication records, etc.), and the corresponding disappearance mode types (such as hide-and-seek mode, absconding with funds mode, false disappearance mode). The historical missing data set is processed, features that have no effect on the classification effect are removed, and missing values and discrete features are processed. The historical missing data set of historical missing customers is used as the sample set, which is divided into a training set and a test set;

[0121] Step A02: Divide the training set into a sub-training set and a sub-test set, and input the sub-training set into the missing mode classification model to be constructed for iterative training of the model, and conduct classification verification on the missing mode classification model to be constructed based on the sub-test set;

[0122] It should be noted that according to the model parameters input by the user (such as learning rate, number of iterations, network structure, etc.), the missing mode classification model to be constructed is initialized. The missing mode classification model to be constructed is an untrained machine learning model, and its goal is to learn to extract features from historical missing data and predict the missing mode of new missing customers based on these features. The training set is further divided into a sub-training set and a sub-test set according to a certain ratio (such as 4:1, 7:3, etc.). The sub-training set is used to perform iterative training on the missing mode classification model to be constructed. During the training process, the model will try to learn the association between features and the missing mode from historical missing data, and after each round of training, the sub-test set is used to conduct classification verification on the missing mode classification model to be constructed to evaluate the performance of the model.

[0123] Exemplarily, XGBoost is selected to construct the missing connection mode classification model to be constructed. XGBoost belongs to the tree model category and is an ensemble learning method based on decision trees. It does not require standardizing the original data and can retain the original features of the data to the greatest extent. During its execution, it splits the subtrees according to the feature importance. Assume that the dataset D of missing customers is D = {(x i , y i )} with m features and n samples. x i is the feature vector of the i-th sample, and y i is the true class label of the i-th sample, where i = 1, 2,..., n. Let the number of trees in the model be K, and the predicted score of the k-th tree be f k (x i ). Then the predicted value of the model for each sample can be represented by the following additive model:

[0124]

[0125] Its objective function Obj(Φ) is:

[0126]

[0127] Among them, is the loss function, and Ω(f k ) is the regularization term, representing the complexity of the k-th decision tree. Its calculation formula is:

[0128]

[0129] Among them, T represents the number of leaf nodes, ω represents the leaf node value, and γ, λ are 's parameters, which are used to adjust the number of leaf nodes and the proportion of leaf nodes respectively. During the process of training the k-th decision tree, the objective function of XGBoost can be transformed into the following equation through iteration and second-order Taylor expansion:

[0130]

[0131] Among them, f t represents the t-th iteration of the tree, g i is the first derivative of , and h i is the second derivative of .

[0132] Exemplarily, to help understand the technical concept or technical principle of this application, please refer to Figure 2 , Figure 2The flowchart for constructing the missing contact mode classification model is provided. First, data processing is performed on the missing contact dataset. The permanent address features of loan missing contact customers, the in-network status features of mobile phone numbers, the call record features of mobile phone numbers, the SMS signaling features, the email features, the other signaling features, the in-network status features of the mobile phone numbers of emergency contacts, the call record features of mobile phone numbers, the SMS signaling features, the email features, as well as the loan type, overdue days, overdue loans, estimated recoverable amount, missing contact duration, number of valid contacts, relationship with valid contacts, and class label of loan missing contact customers are respectively represented by x1, x2, x3, x4, x5, x6, x7, x8, x9, x 10 、x 11 、x 12 、x 13 、x 14 、x 15 、x 16 、x 17 、x 18 wherein, x1, x2, x4, x5, x6, x7, x9, x 10 、x 11 、x 17 、x 18 These 11 features belong to discrete features, and the remaining 7 features belong to continuous features. Secondly, for the three features of x4, x5, x 10 in the missing contact dataset, some of the feature values are missing. The method of filling missing values with the mode is used to process the missing values of these three features. Then, since XGBoost is a supervised machine learning method and there is "good input" before there can be "good output", the data of the "hide-and-seek" mode, the absconding with funds mode, and the false disappearance mode marked by industry experts is used as the input for the XGBoost model x 18 to identify the missing contact types of loan missing contact customers. However, since XGBoost cannot recognize text of string type during the modeling process, in the data processing link, the class labels of x 18 are set to 0, 1, 2, where 0 represents the "hide-and-seek" mode, 1 represents the absconding with funds mode, and 2 represents the false disappearance mode. Since x1, x2, x4, x5, x6, x7, x9, x 10 、x 11 、x 17 These 10 features are discrete features and there is no correlation between the feature values of the variables. In order to make the decision trees of XGBoost handle discrete features in a consistent manner, the OrdinalEncoder encoding form is used to process the discrete features.

[0133] Secondly, divide the 60 sample data of the lost-contact dataset into two parts: the training set Train and the test set Test. Among them, the training set contains 36 samples, and the test set contains 24 samples. That is, the ratio of the number of samples in the training set to the test set is 6:4. The formats of the training set data and the test set data are exactly the same, but the x of the test set 18 is not given and is the variable for classification prediction. Then, divide the training set Train into a sub-training set Train' and a sub-test set Test' according to the ratio of 6:4. Use 17 features of the sub-test set Train' as the input features of XGBoost, and use the 3 lost-contact modes of loan lost-contact customers as the class labels. Part of the data of the training set Train is shown in Table 1, and part of the data of the test set Test is shown in Table 2.

[0134] Table 1

[0135]

[0136] Table 2

[0137]

[0138]

[0139] Then, set the main parameters of XGBoost. Set the learning rate to 0.1 according to the default value, set the number of trees to 1000, and set the depth of the tree to 6. Since the smaller the minimum weight of the leaf node, the easier it is to overfit, the minimum weight of the leaf node is set to 1 to control the sum and minimum value of the second derivative of the leaf node. Randomly select 80% of the samples in Train' to build the decision tree and 80% of the features in Train' to build the decision tree. Since this application aims to solve the multi-classification problem of lost-contact modes, the objective function parameter is selected as multi:softmax, and other parameters are set according to the default settings of the model.

[0140] Finally, use the XGBClassifier of the scikit-learn API to apply the XGBoost model to train Test', and then classify Test' to complete the training of the XGBoost model and obtain the classification result. The result shows that among the classification results of 15 samples, 4 out of 5 samples of class 0 are classified correctly and 1 is classified incorrectly, all 4 samples of class 1 are classified correctly, and all 6 samples of class 2 are classified correctly. After model classification, the classification result of Test' is shown in Table 3:

[0141] Table 3

[0142] TP FP FN 0 4 0 1 1 4 0 0 2 6 0 0

[0143] Among them, TP represents the samples where positive samples are predicted as positive, FP represents the samples where negative samples are predicted as positive, and FN represents the samples where positive samples are predicted as negative.

[0144] Step A03, in the case where the classification verification result of the to-be-constructed out-of-contact mode classification model does not meet the preset training conditions, calculate the loss function value of each round of training, and adjust the model parameters in the to-be-constructed out-of-contact mode classification model based on the loss function value.

[0145] It should be noted that during the model training process, the system will continuously monitor the loss function value of each round of training. The loss function value is an index to measure the difference between the model prediction result and the actual result. In machine learning, the smaller the loss function value, the stronger the prediction ability of the model. If the classification verification result of the to-be-constructed out-of-contact mode classification model does not meet the preset training conditions (such as indicators such as accuracy rate and recall rate reaching the set standards), the model parameters will be adjusted based on the loss function value to further reduce the prediction error.

[0146] Step A04, in the case where the classification verification result of the trained to-be-constructed out-of-contact mode classification model meets the preset training conditions, use the trained to-be-constructed out-of-contact mode classification model as the out-of-contact mode classification model.

[0147] It should be noted that after the classification verification result of the trained to-be-constructed out-of-contact mode classification model meets the preset training conditions, it is regarded as the final out-of-contact mode classification model.

[0148] In this embodiment, by using the historical out-of-contact dataset of historical out-of-contact customers as the sample set, the model can more accurately learn the features related to the out-of-contact mode, initialize the model based on the model parameters input by the user, input the training set and the test set into the model for training, calculate the loss function value of each round of training, and adjust the model parameters according to the loss function value. By continuously adjusting the model parameters, the prediction performance of the model is gradually improved, the loss function value is gradually reduced, and the prediction accuracy and reliability are improved.

[0149] In a feasible embodiment, in step A02, the steps of classifying and verifying the to-be-constructed out-of-contact mode classification model based on the sub-test set include steps A11 to A12:

[0150] Step A11, calculate the accuracy rate, precision rate, recall rate, and F1 score of the trained to-be-constructed out-of-contact mode classification model based on the sub-test set.

[0151] It should be noted that after the to-be-constructed lost connection mode classification model is trained, the performance of the to-be-constructed lost connection mode classification model is evaluated through the sub-test set Test', including the evaluation of accuracy, precision, recall rate, and F1 score. Accuracy is an indicator to measure the overall prediction correctness of the model. Precision is an indicator to measure the accuracy of the model in predicting lost connection customers. Recall rate is an indicator to measure the ability of the model to identify lost connection customers. F1 score is an indicator to comprehensively evaluate the precision and recall rate of the model.

[0152] Exemplarily, assume n correct represents the number of samples correctly classified by the model. The formula for calculating the accuracy Accuracy of the model is:

[0153]

[0154] Assume that there are j types of actual lost connection modes. The samples of the jth type are positive samples, and the samples other than the samples of the jth type are negative samples. After classification by the model, TP j represents the number of positive samples predicted as positive, and FP j represents the number of negative samples predicted as positive, and FN j represents the number of positive samples predicted as negative. The results after classification by the model are shown in Table 4:

[0155] Table 4

[0156] TP FP FN 0 <![CDATA[TP0]]> <![CDATA[FP0]]> <![CDATA[FN0]]> 1 <![CDATA[TP1]]> <![CDATA[FP1]]> <![CDATA[FN1]]> 2 <![CDATA[TP2]]> <![CDATA[FP2]]> <![CDATA[FN2 <!-- 12 -->]]> … … … … j <![CDATA[TP j > <![CDATA[FP j > <![CDATA[FN j >

[0157] Considering that the model is for the comprehensive prediction performance of j categories, the overall precision of the model is calculated in the form of accumulating the precision of the classification results of the jth category and then taking the average. The precision (Precision) of the model for all modes as a whole is shown in the following equation:

[0158]

[0159] Similarly, the recall rate (Recall) of the model for all modes as a whole is:

[0160]

[0161] The F1 score (F1_score) of the model for all modes as a whole is:

[0162]

[0163] Among them, F1_socre represents the harmonic mean of Precision and Recall.

[0164] Step A12: When the accuracy rate reaches the preset accuracy rate threshold, the precision rate reaches the preset precision rate threshold, the recall rate reaches the preset recall rate threshold, and the F1 score reaches the preset F1 score threshold, it is determined that the classification verification result of the out-of-contact mode classification model to be constructed after training reaches the preset training conditions.

[0165] It should be noted that the preset accuracy rate threshold, preset precision rate threshold, preset recall rate threshold, and preset F1 score threshold are the minimum levels that the model accuracy rate, model precision rate, model recall rate, and model F1 score need to reach, which are preset according to business requirements. Compare the model performance indicators (accuracy rate, precision rate, recall rate, F1 score) calculated in step A11 with the corresponding preset thresholds. If the accuracy rate, precision rate, recall rate, and F1 score of the model all reach the preset thresholds, it is determined that the out-of-contact mode classification model after training reaches the preset training conditions and can be used in actual business.

[0166] Exemplarily, to facilitate understanding of the technical concept or technical principle of this application, please refer to Figure 2 , Figure 2 A flowchart for constructing an out-of-contact mode classification model is provided. After training the XGBoost model, based on the classification results in Table 3, it can be seen that there are 14 correctly classified samples and 15 total samples. Therefore, the classification accuracy rate of the model is And the precision rate, recall rate, and F1 score of the model for the three modes of 0, 1, 2 and the overall are calculated, as shown in Table 4.

[0167] Table 5

[0168] Precision Recall F1_score 0 1.000000 0.800000 0.888888 1 1.000000 1.000000 1.000000 2 0.857143 1.000000 0.923077 All mode 0.952381 0.933333 0.937322

[0169] From this, it can be seen that XGBoost has the best classification effect on the first type of mode in Train', followed by the second type of mode, and finally the third type of mode. For the overall classification effect of Train', the precision rate is 0.952381, the recall rate is 0.933333, and the F1_score is 0.937322. Generally speaking, the classification effect of XGBoost on Test' is relatively good and there is no need to adjust the parameters.

[0170] In this embodiment, the accuracy rate measures the proportion of correct classifications of the model among all test samples, providing an overview of the overall performance. The precision reflects the proportion of truly positive samples among the samples predicted as positive by the model. The recall rate measures the proportion of truly positive samples that are correctly identified by the model among all truly positive samples. The F1 score provides a comprehensive index of the performance of both precision and recall rate. Based on the calculation of these metrics, financial institutions can comprehensively evaluate the performance of the model, thereby ensuring that the model has sufficient accuracy and reliability in practical applications. Only when the model meets the preset thresholds for all metrics is it determined to meet the preset training conditions, ensuring that the model performs well in multiple aspects, thus improving its practicality and reliability.

[0171] In a feasible embodiment, after the step of inputting the sub-training set into the to-be-constructed out-of-contact mode classification model for iterative training of the model in step A02, steps A21 to A22 are included:

[0172] Step A21, calculating the average gain of each candidate sample feature in the sample set;

[0173] It should be noted that all candidate sample features in the sample set are traversed, and their gains are calculated at each possible splitting point. The splitting point refers to the threshold for dividing the data into two subsets on a certain feature. The goal of the decision tree is to find the optimal splitting point so that the split subsets have higher purity (i.e., samples of the same category or similar values are grouped together). The average gain of all candidate sample features is calculated to obtain the average gain of each feature. The average gain reflects the average contribution degree of the feature to the improvement of the prediction ability during the model training process. The higher the average gain of a feature, the more important it is in the model and the greater the impact on the accuracy of the prediction result.

[0174] In addition, it should be noted that in order to measure the importance of each feature in out-of-contact mode recognition, Gain is used as an evaluation index for measuring feature importance. Gain refers to the average gain after splitting the tree using a certain feature. The average gain calculation formula is:

[0175]

[0176] where Gain is the average gain value, L represents the left subtree, R represents the right subtree, I is the unsplit subtree, g i is the first derivative of, h i is the second derivative of, is the loss function, and λ is a parameter of, used to adjust the proportion of leaf nodes.

[0177] Exemplarily, for the purpose of facilitating the understanding of the technical concept or technical principle of the present application, please refer to Figure 2 , Figure 2 A flowchart for constructing a missing connection mode classification model is provided. After evaluating the classification results in Table 3, the feature importance of each candidate sample feature is sorted. According to the average gain calculation formula, the importance of each feature for the missing connection mode classification is different. As Figure 3 shown, Figure 3 is a distribution diagram of the feature importance ranking, where the horizontal axis is the Gain value of the feature, representing the importance of each feature, and the vertical axis is the label of each candidate sample feature. Among the 17 features, the 13 Gain value of x is the largest, which is 0.22389846. Immediately following is x 12 , with a Gain value of 0.18830772. The importance of x2 and x 14 for the missing connection mode classification is relatively close, with values of 0.15527190 and 0.14787544 respectively. The importance of x1, x3, x 15 , x8, x 16 , x 11 these six features has a certain effect on the classification of the missing connection mode, but their importance decreases in turn. In the Train' and Test' datasets, after learning and testing, x4, x5, x6, x7, x9, x 10 , x 17 these seven features have no impact on the recognition of the missing connection mode. Therefore, it can be considered to remove them from the sample set.

[0178] In addition, it should be noted that after sorting the feature importance of each candidate sample feature, the final missing connection mode classification model is output, and the trained model is used to test Test. The parameter settings remain unchanged. The results show that the classification of XGBoost for 24 test samples Test is: 0, 0, 0, 1, 0, 2, 1, 0, 2, 0, 0, 0, 0, 0, 0, 0, 2, 0, 0, 1, 0, 0, 0, 2. From this, it can be seen that in the test set Test, there are 17 samples in the "hide and seek" mode, 3 samples in the absconding with funds mode, and 4 samples in the false disappearance mode.

[0179] Step A22: Use the candidate sample features with an average gain higher than the preset gain threshold as the acquisition indicators for the missing connection data.

[0180] It should be noted that after obtaining the average gain of each sample feature, high-weight and low-weight features are distinguished according to a preset gain threshold. This threshold is usually set according to the performance requirements of the model, the characteristics of the data, and empirical rules. Features with an average gain lower than this threshold are regarded as features that contribute less or are redundant to the model prediction performance, that is, low-weight features. Traverse the average gains of all features, mark the features with an average gain lower than the preset gain threshold as low-weight features, and remove these metrics from the sample set. The remaining in the sample set are the candidate sample features with an average gain higher than the preset gain threshold, which are used as the collection metrics for lost connection data. Subsequent data collection is based on these collection metrics.

[0181] Additionally, after identifying low-weight features, they can be output to the user. After the user determines to delete them, the low-weight features in the sample set are then deleted to obtain the collection metrics for lost connection data.

[0182] In this embodiment, by calculating the average gain of each feature after the model splitting node, the contribution of each feature to the model prediction performance is quantified, which helps to identify the most critical features for model prediction and provides an objective basis for the subsequent optimization of the feature set. By removing the candidate sample features with an average gain lower than the preset gain threshold from the sample set, it helps to identify and eliminate the features that contribute less to the model prediction performance, realize the optimization of the feature set, and construct a more concise and efficient model, thereby improving the prediction accuracy and training efficiency of the model. At the same time, by reducing the feature dimension, the complexity of the model can be reduced, and the training speed and prediction accuracy of the model can be improved.

[0183] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar content as in the above-mentioned first embodiment can be referred to the above introduction and will not be repeated hereinafter. On this basis, please refer to Figure 4 , in step S03, the steps of determining the lost connection mode information of the lost connection customers according to the model classification result include steps S11 to S12:

[0184] Step S11, in response to the user's mode selection operation, when the mode selection operation is to turn off the rule assistance mode, use the model classification result as the lost connection mode information of the lost connection customers;

[0185] It should be noted that when the user selects to turn off the rule-assisted mode through the interface or other means, the system directly uses the model classification result of the preset out-of-contact mode classification model as the out-of-contact mode information of the out-of-contact customer. The mode selection operation refers to the explicit selection made by the user through interface interaction or other means on whether to enable the rule-assisted mode. Turning off the rule-assisted mode means that the user hopes that the system determines the out-of-contact mode information only based on the model classification result without introducing other rules or logics. In the closed state, the rule-assisted mode does not participate in the process of determining the out-of-contact mode information. The rule-assisted mode is usually used to further adjust or refine the result through preset rules on the basis of model classification.

[0186] Step S12, in the case where the mode selection operation is to turn on the rule-assisted mode, determine and output the out-of-contact mode information of the out-of-contact customer based on the preset mode determination auxiliary rule and the model classification result.

[0187] It should be noted that when the user selects to turn on the rule-assisted mode, the system not only considers the model classification result but also combines the preset mode determination auxiliary rule to determine the out-of-contact mode information of the out-of-contact customer and outputs the final out-of-contact mode information. The out-of-contact mode information is the result determined by comprehensively combining the model classification result and the preset rule, reflecting a more refined and accurate customer out-of-contact state. When the mode selection operation is that the user selects to enable the rule-assisted mode, it means that the user hopes that the system can further optimize or adjust the model classification result by combining the preset mode determination auxiliary rule. The preset mode determination auxiliary rule is formulated according to business requirements, historical data or expert knowledge and is used to further determine the model classification result in the rule-assisted mode, including logical judgments based on factors such as customer historical behavior, transaction records, and credit scores.

[0188] In this embodiment, when the user's mode selection operation is to turn off the rule-assisted mode, the system directly uses the result of model classification as the out-of-contact mode information of the out-of-contact customer, improving the processing efficiency, reducing the time cost brought by manual intervention or rule judgment, ensuring the direct application of the model classification result, and maintaining the consistency and accuracy of classification. When the user's mode selection operation is to turn on the rule-assisted mode, the system obtains the result of model classification and further determines the classification result according to the preset mode determination auxiliary rule, and finally determines and outputs the out-of-contact mode information of the out-of-contact customer, making the finally output out-of-contact mode information more reliable and useful, more in line with the actual business requirements, and improving the accuracy and applicability of the classification result.

[0189] In a feasible embodiment, in step S12, the steps of determining and outputting the out-of-contact mode information of the out-of-contact customer based on the preset mode determination auxiliary rule and the model classification result include steps B01 to B03:

[0190] Step B01: Determine whether the model classification result is consistent with the rule assistance result obtained based on the preset mode determination assistance rules;

[0191] It should be noted that the system calculates the rule assistance result according to a series of preset determination assistance rules (these rules may be based on historical experience, business logic or industry knowledge to determine which type of classification mode it is), and compares the rule assistance result with the model classification result to check whether the two results are the same.

[0192] Step B02: When the model classification result is consistent with the rule assistance result, output the model classification result or the rule assistance result as the disconnection mode information of the disconnected customer;

[0193] It should be noted that when the rule assistance result is consistent with the model classification result, the system believes that both methods effectively capture the characteristics of the data, and selects one of them (the model classification result or the rule assistance result, which are actually the same at this time) as the final disconnection mode information output.

[0194] Step B03: When the model classification result is inconsistent with the rule assistance result, take the model classification result as the model prediction result, take the rule assistance result as the experience prediction result, and output the model prediction result and the experience prediction result as the disconnection mode information of the disconnected customer.

[0195] It should be noted that if the model classification result is inconsistent with the rule assistance result, the system marks the model classification result as the "model prediction result" and the rule assistance result as the "experience prediction result". Subsequently, the system outputs these two results together as the disconnection mode information of the disconnected customer. Among them, the model prediction result is the prediction obtained by analyzing the input data based on the machine learning algorithm, reflecting the potential laws and trends in the data, which is the result of algorithmic automated processing. The experience prediction result is the prediction obtained according to the preset rules, historical experience or domain knowledge, and more relies on the prior knowledge of industry experts and business logic.

[0196] In this embodiment, by comparing the model classification result with the rule-assisted result, the system can initially identify the consistency or difference between the two methods, providing an important basis for decision-making in subsequent steps. If the results are consistent, it indicates that the two methods are complementary to a certain extent, enhancing the credibility of the determination and helping to reduce the situations of misjudgment and missed judgment. If they are inconsistent, it prompts the need for further analysis to determine which result is more reliable or whether it is necessary to comprehensively consider both to improve customer satisfaction and service quality. By outputting the model prediction result and the empirical prediction result together as the loss-of-contact mode information of the loss-of-contact customers, both the original information of the two methods is retained, and a more comprehensive perspective is provided for decision-makers to refer to, which helps to enhance the flexibility and accuracy of decision-making and at the same time enhances the adaptability and robustness of the system.

[0197] In a feasible embodiment, before the step of determining whether the rule classification result is consistent with the assisted result obtained based on the preset mode determination assistance rule in step B01, there are also steps B11 to B12:

[0198] Step B11, comparing the loss-of-contact data of the loss-of-contact customers with each index threshold in the preset mode determination rule;

[0199] It should be noted that the detailed loss-of-contact data of the loss-of-contact customers is extracted from the database. These data usually include information in four dimensions: the degree of contact, loan situation, loss-of-contact duration, and social relationship. The preset loss-of-contact mode determination rules are loaded. These rules contain multiple indicators for judging the loss-of-contact mode and their corresponding thresholds. Each piece of data of the loss-of-contact customers is compared with the corresponding index threshold in the preset rules to determine whether the data meets the threshold requirements of the index.

[0200] In addition, it should be noted that for the determination rules corresponding to the hide-and-seek mode, they include:

[0201] ① The reserved address is a temporary address: When applying for a loan, the loss-of-contact customer reserves a non-household registration address or business address. That is, in order to avoid contact with financial institutions and relevant personnel, the loss-of-contact customer uses the address of a rented house, the address of a temporary work unit, the address of a hotel or other temporary addresses as the reserved address, that is, x1 = 0.

[0202] ② The mobile phone number is in a normal in-network state: The mobile phone number of the lost-contact customer is in a normal in-network state. However, the lost-contact customer refuses to contact the relevant personnel of the financial institution by not answering, busy tone, power-off, etc. In addition, the financial institution is also unable to contact the lost-contact customer through channels such as text messages, emails, Apps, mini-programs, or official accounts. Specifically, the status of the lost-contact customer's mobile phone number is normal, the call record of the mobile phone number is prompted as busy tone or power-off, the text message signaling feature is not replied, the email feature is not replied, and other signaling features are not replied, that is, x2 = A0, x3 ≠ 0, x4 = 0, x5 = 0, x6 = 0.

[0203] ③ It is difficult to contact the emergency contact: The lost-contact customer reserved the contact information of their emergency contact when handling the loan business. However, when the financial institution contacts this emergency contact, it encounters a situation of an empty number, or power-off, busy tone, not answering. The financial institution is also difficult to contact this emergency contact through channels such as text messages and emails. Specifically, when the mobile phone number of the emergency contact is an empty number or the status of the mobile phone number of the emergency contact is normal, the call record of the mobile phone number of the emergency contact is prompted as busy tone or power-off, the text message signaling feature of the emergency contact is not replied, the email feature of the emergency contact is not replied, that is, (x7 = A3) ∪ (x7 = A0) ∩ (x8 ≠ 0), x9 = 0, x 10 = 0.

[0204] ④ The credit loan accounts for a relatively large proportion: In the hide-and-seek mode, the loan type is a credit loan, and the lost-contact customers without guarantors and collateral account for a relatively large proportion, that is, x 11 = B2.

[0205] ⑤ Weak social relations: According to the mobile phone number reserved by the lost-contact customer, it is found that there are few communication contacts of them, and it is deduced that their social relations are relatively weak. Comparing the number of contacts in their address book in different time periods, if the number fluctuates greatly, it means that the social relations of the lost-contact customer are unstable. When the set of mobile phone numbers of the effective contacts in the address book and the contacts such as the parents, spouse, and direct relatives of the lost-contact customer is an empty set, there is no one similar to a direct relative in the address book of the loan lost-contact customer, and it is difficult for the financial institution to learn about their whereabouts from the effective contacts of the lost-contact customer, that is, x 16 ≤ the lower limit of the number of effective contacts (for example, 5), x 17 = 0, σ 16 ≥ the upper limit of the standard deviation (for example, 1), where σ 16 is the standard deviation of x 16 and represents the fluctuation of the number of effective contacts within a period of time. When σ 16 is greater than the upper limit of the standard deviation, it means that the number fluctuates greatly and the social relations of the lost-contact customer are unstable.

[0206] Exemplarily, for the purpose of facilitating the understanding of the technical concept or technical principle of the present application, please refer to Figure 5 , Figure 5 A feature map corresponding to the hide-and-seek mode is provided. In the hide-and-seek mode, there may be a connection between the lost customer and the emergency contact and the valid contact. The lost customer reserves a temporary address, but has changed their residential address. The call record of the lost customer's mobile phone number is prompted as busy tone, empty number or not answered. The call record of the lost customer's emergency contact's mobile phone number is prompted as busy tone, turned off, empty number or not answered. And most of the loan types of the lost customer are unsecured loans, without a guarantor, without collateral or with little collateral value, which is not enough to repay the loan amount. It is difficult for financial institutions to reach both the lost customer and the lost customer's emergency contact. The number of valid contacts of the lost customer is small, unstable, and the social relationship is weak. And most of their valid contacts do not know or do not disclose the information of the lost customer. The financial institution has a poor reach effect on the valid contacts of the lost customer.

[0207] In addition, it should be noted that the determination rules corresponding to the mode of absconding with funds include:

[0208] ① The overdue loan amount is relatively large: The lost customers in this lost mode generally have a relatively large overdue loan amount. The loan types are mostly mortgage loans, guaranteed loans or discount loans, with a guarantor, collateral or discount bills. However, after the lost customer is lost, the value of the realizable assets such as collateral and discount bills is less than the overdue loan. The guarantor can also not bear or can only bear part of the responsibility for repaying the loan. In some cases, even the guarantor is lost, that is, x 13 ≥ the preset amount upper limit, x 11 = B0, B1, B3, x 14 <x 13 .

[0209] ② Unable to reach: In the mode of absconding with funds, the mobile phone numbers of the lost loan customers are in a state of being out of service, empty number, unavailable while on the network or other abnormal states in many cases. There are also some lost loan customers whose mobile phone numbers are in a normal on-network state, but the lost customers ignore the contact of relevant personnel of the financial institution by turning off the phone or not answering. It is also difficult for the financial institution to contact the lost customer and the lost customer's emergency contact through channels such as text messages, emails, apps, mini-programs or official accounts, that is, x2 = A1, A2, A3, A5, x3≠0, x4 = 0, x5 = 0, x6 = 0, (x7 = A3) ∪ (x7 = A0) ∩ (x8≠0), x9 = 0, x 10 = 0.

[0210] ③No valid address: The addresses reserved by the lost-contact customers at financial institutions are mostly the household registration addresses on their ID cards or the residential addresses of their business licenses. However, in order to avoid contact with financial institutions and relevant personnel, the lost-contact customers do not live at the reserved household registration addresses or business addresses, nor do they work at their original work units. It is impossible to contact the lost-contact customers through valid addresses, that is, x1 = 1.

[0211] ④The social relationships are relatively complex and extensive. In the mode of absconding with funds, there are many communication contacts of the lost-contact customers. Most of the lost-contact customers have relatively complex and extensive social relationships. There were relatively stable mobile contacts before they lost contact. However, it is also difficult for financial institutions to learn about the whereabouts of the loan lost-contact customers from these mobile contacts, that is, x 16 ≥ upper limit of the number of valid contacts (for example, 100), σ 16 < upper limit of the standard deviation.

[0212] Exemplarily, to help understand the technical concept or technical principle of the present application, please refer to Figure 6 , Figure 6 A characteristic diagram corresponding to the mode of absconding with funds is provided. There may be connections between the lost-contact customers and their emergency contacts and valid contacts. The lost-contact customers reserved household registration addresses or business addresses, but they have left the place where their work units are located. The call records of their mobile phone numbers are prompted as out of service, empty number, powered off, busy tone, abnormal or not answered. The call records of the mobile phone numbers of the lost-contact customers' emergency contacts are also prompted as powered off, busy tone, empty number or not answered. Moreover, most of the loan types of the lost-contact customers are mortgage loans, guaranteed loans or discount loans, and there are also credit loans but the proportion is relatively small. The amount of overdue loans is relatively large. It is difficult for financial institutions to reach both the lost-contact customers and their emergency contacts. The number of valid contacts of the lost-contact customers is large, their social relationships are complex and extensive, and most of their valid contacts do not know or do not disclose information about the lost-contact customers. The effect of financial institutions reaching the valid contacts of the lost-contact customers is poor.

[0213] In addition, it should be noted that the determination rules corresponding to the false disappearance mode include:

[0214] ①Unable to reach: In the false disappearance mode, the mobile phone numbers of the lost-contact customers mostly show the states of out of service, empty number, and cancelled mobile phone numbers in many cases. The lending institutions can no longer contact the customers themselves through means such as mobile phone numbers, text messages, emails, apps, mini-programs or official accounts, that is, x2 = A1, A3, A4, x3 ≠ 0, x4 = 0, x5 = 0, x6 = 0.

[0215] ②No valid address: In the false disappearance mode, most of the lost-contact customers avoid repaying the debts of the lending institutions by leaving their original residences, original work units, etc. Therefore, the permanent addresses once reserved by the loan lost-contact customers at the lending institutions are invalid, that is, x1 = 1.

[0216] ③The time of losing contact exceeds the preset duration (e.g., three months): Compared with other modes, the customers who lose contact in the false disappearance mode generally have more overdue days, and at the same time, the time of losing contact is longer, that is, x 15 ≥ the preset duration.

[0217] ④Cut off the original social relationships: The customers who lose contact in the false disappearance mode actively cut off their original social relationships. For example, the loan customers who lose contact do not contact their relatives, friends, colleagues, etc. during the investigation period. Even if there are many valid contacts in the original mobile phone address book before their loss of contact, the lending institutions cannot learn their whereabouts from these valid contacts, that is, x 16 ≥ the upper limit of the number of contacts (e.g., 100), σ 16 < the upper limit of the standard deviation.

[0218] Exemplarily, to help understand the technical concept or technical principle of the present application, please refer to Figure 7 , Figure 7 a characteristic diagram corresponding to the false disappearance mode is provided. In the false disappearance mode, the customers who lose contact cut off the connection with the emergency contacts and valid contacts, leave their permanent addresses or work units, and the call records of the mobile phone numbers are prompted as out of service, invalid numbers, or cancelled mobile phone numbers (not enabled). The time of losing contact of the customers who lose contact is long and the expected number of days is large. It is difficult for financial institutions to reach both the customers who lose contact and their emergency contacts. Moreover, in the false disappearance mode, although the number of valid contacts of the customers who lose contact is large and the social relationships are stable, neither the emergency contacts nor the valid contacts of the customers who lose contact know the news of the customers who lose contact, and the financial institutions have no effect of reaching the valid contacts of the customers who lose contact.

[0219] Step B12, record the index thresholds satisfied by the data of the customers who lose contact, determine the corresponding loss-of-contact mode of the customers who lose contact based on the satisfied index thresholds, and use the corresponding loss-of-contact mode as the rule assistance result.

[0220] It should be noted that all the index thresholds satisfied by the data of the customers who lose contact are recorded. These thresholds are the key basis for judging the loss-of-contact mode. Based on the satisfied thresholds, multiple indexes that meet the thresholds are comprehensively considered, and according to the logic in the preset rules, the loss-of-contact mode that the customers who lose contact most conform to is judged, and the determined loss-of-contact mode is output as the rule assistance result for use in subsequent steps.

[0221] In this embodiment, by comparing the lost-contact data with the multi-dimensional index thresholds, the part of the lost-contact customer data that matches the preset rules can be quickly and accurately identified, providing a basis for subsequent pattern determination. This not only improves the efficiency of determination but also ensures the accuracy of determination, helping to reduce the situations of misjudgment and missed judgment. According to the index thresholds met by the lost-contact customers, their corresponding lost-contact patterns are determined and output as the rule auxiliary results, which not only provides important reference information for subsequent comprehensive determination but also helps to improve the pertinence and accuracy of business decisions.

[0222] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the method for confirming the lost-contact pattern of this application. Based on this technical concept, more forms of simple transformation are within the protection scope of this application.

[0223] This application also provides a lost-contact pattern confirmation device. Please refer to Figure 8 , and the lost-contact pattern confirmation device includes:

[0224] A data acquisition module 10, which is used to extract the four-dimensional correlation features of lost-contact customers from each preset data source to form lost-contact data;

[0225] A model input module 20, which is used to input each acquisition index under the four-dimensional correlation features in the lost-contact data into a preset lost-contact pattern classification model to obtain a model classification result;

[0226] A result confirmation module 30, which is used to determine the lost-contact pattern information of lost-contact customers according to the model classification result, where the lost-contact pattern is the lost-contact type of lost-contact customers.

[0227] Optionally, the lost-contact pattern confirmation device includes a model training module 40, and the model training module 40 is used for:

[0228] Taking the historical lost-contact data set of historical lost-contact customers as a sample set, and the sample set includes a training set;

[0229] Dividing the training set into a sub-training set and a sub-test set, and inputting the sub-training set into the to-be-built lost-contact pattern classification model for iterative training of the model, and performing classification verification on the to-be-built lost-contact pattern classification model based on the sub-test set;

[0230] In the case that the classification verification result of the to-be-built lost-contact pattern classification model does not meet the preset training conditions, calculating the loss function value of each round of training, and adjusting the model parameters in the to-be-built lost-contact pattern classification model based on the loss function value;

[0231] In the case that the classification verification result of the trained to-be-built lost-contact pattern classification model meets the preset training conditions, taking the trained to-be-built lost-contact pattern classification model as the lost-contact pattern classification model.

[0232] Optionally, the model training module 40 is further configured to:

[0233] Calculate the accuracy, precision, recall rate, and F1 score of the to-be-built missing connection pattern classification model after training based on the sub-test set;

[0234] When the accuracy reaches the preset accuracy threshold, the precision reaches the preset precision threshold, the recall rate reaches the preset recall rate threshold, and the F1 score reaches the preset F1 score threshold, determine that the classification verification result of the to-be-built missing connection pattern classification model after training meets the preset training conditions.

[0235] Optionally, the model training module 40 is further configured to:

[0236] Calculate the average gain of each candidate sample feature of the samples in the sample set;

[0237] Use the candidate sample features with an average gain higher than the preset gain threshold as the acquisition metrics for missing connection data.

[0238] Optionally, the result confirmation module 30 is further configured to:

[0239] In response to the user's mode selection operation, when the mode selection operation is to turn off the rule assistance mode, use the model classification result as the missing connection mode information of the missing connection customer;

[0240] When the mode selection operation is to turn on the rule assistance mode, determine and output the missing connection mode information of the missing connection customer based on the preset mode determination assistance rules and the model classification result.

[0241] Optionally, the result confirmation module 30 is further configured to:

[0242] Determine whether the model classification result is consistent with the rule assistance result obtained based on the preset mode determination assistance rules;

[0243] When the model classification result is consistent with the rule assistance result, output the model classification result or the rule assistance result as the missing connection mode information of the missing connection customer;

[0244] When the model classification result is inconsistent with the rule assistance result, use the model classification result as the model prediction result, use the rule assistance result as the empirical prediction result, and output the model prediction result and the empirical prediction result as the missing connection mode information of the missing connection customer.

[0245] Optionally, the result confirmation module 30 is further configured to:

[0246] Compare the missing connection data of the missing connection customer with the various index thresholds in the preset mode determination rules;

[0247] Record the metric thresholds that the disconnection data of the disconnected customers satisfies, determine the corresponding disconnection mode of the disconnected customers based on the satisfied metric thresholds, and use the corresponding disconnection mode as the rule assistance result.

[0248] The disconnection mode confirmation device provided in this application adopts the disconnection mode confirmation method in the above-mentioned embodiment, which can solve the technical problem of difficultly determining the disconnection mode of customers. Compared with the prior art, the beneficial effects of the disconnection mode confirmation device provided in this application are the same as those of the disconnection mode confirmation method provided in the above-mentioned embodiment, and other technical features in the disconnection mode confirmation device are the same as the features disclosed in the method of the above-mentioned embodiment, which will not be elaborated here.

[0249] 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, and the instructions are executed by the at least one processor so that the at least one processor can execute the disconnection mode confirmation method in the first embodiment above.

[0250] The following refers to Figure 9 , which shows a schematic structural diagram of an electronic device suitable for implementing the embodiments of the present application. The electronic device in the embodiments of the present application may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, PADs (Portable Application Description: tablet computers), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 9 The electronic device shown is only an example and should not impose any limitations on the functions and usage scope of the embodiments of the present application.

[0251] As Figure 9As shown, the electronic device may include a processing device 1001 (such as a central processing unit, a graphics processing unit, etc.), which may perform various appropriate actions and processes according to a program stored in the read-only memory 1002 or a program loaded from the storage device 1003 into the random access memory 1004. In the random access memory 1004, various programs and data required for the operation of the electronic device are also stored. The processing device 1001, the read-only memory 1002, and the random access memory 1004 are connected to each other through a bus 1005. The input / output interface 1006 is also connected to the bus. Generally, the following systems may be connected to the input / output interface 1006: an input device 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a microphone, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 may allow the electronic device to communicate with other devices wirelessly or wiredly to exchange data. Although the figure shows an electronic device having various systems, it should be understood that it is not required to implement or have all the systems shown. Instead, more or fewer systems may be implemented or had.

[0252] Specifically, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts may be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program codes for executing the methods shown in the flowcharts. In such an embodiment, the computer program may be downloaded and installed from a network through the communication device, or installed from the storage device 1003, or installed from the read-only memory 1002. When the computer program is executed by the processing device 1001, the above functions defined in the methods of the embodiments disclosed in the present application are executed.

[0253] The electronic device provided by the present application adopts the disconnection mode confirmation method in the above embodiments, and can solve the technical problem of difficultly determining the disconnection mode of customers. Compared with the prior art, the beneficial effects of the electronic device provided by the present application are the same as those of the disconnection mode confirmation method provided by the above embodiments, and other technical features in the electronic device are the same as those disclosed in the method of the previous embodiment, and will not be elaborated here.

[0254] It should be understood that the various parts disclosed in the present application may be implemented by hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in a suitable manner in any one or more embodiments or examples.

[0255] As described above, this is only the specific implementation manner of the present application. However, the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present application, and all should be covered by the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

[0256] The present application provides a computer-readable storage medium having computer-readable program instructions (i.e., computer programs) stored thereon, and the computer-readable program instructions are used to execute the out-of-contact mode confirmation method in the above embodiments.

[0257] The computer-readable storage medium provided by the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination of the above. More specific examples of the computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM) or a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium that contains or stores a program, and the program can be used by or combined with an instruction execution system or device. The program code contained on the computer-readable storage medium can be transmitted by any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0258] The above computer-readable storage medium may be included in an electronic device; or it may exist separately without being assembled into the electronic device.

[0259] The above computer-readable storage medium carries one or more programs. When the one or more programs are executed by an electronic device, the out-of-contact mode confirmation device is enabled to: extract four-dimensional correlation features of out-of-contact customers from each preset data source to form out-of-contact data; input each collection index under the four-dimensional correlation features in the out-of-contact data into a preset out-of-contact mode classification model to obtain a model classification result; and determine the out-of-contact mode information of the out-of-contact customers according to the model classification result, where the out-of-contact mode is the out-of-contact type of the out-of-contact customers.

[0260] Computer program code for performing the operations of this application can be written in one or more programming languages or combinations thereof. The above-mentioned programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, executed as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (for example, by connecting through the Internet using an Internet service provider).

[0261] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code that contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0262] The modules described in the embodiments of this application can be implemented in software or in hardware. Among them, the name of the module does not constitute a limitation to the unit itself in some cases.

[0263] The readable storage medium provided by this application is a computer-readable storage medium. The computer-readable storage medium stores computer-readable program instructions (i.e., computer programs) for performing the above-mentioned method for confirming the out-of-contact mode, which can solve the technical problem of difficultly determining the out-of-contact mode of customers. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by this application are the same as those of the method for confirming the out-of-contact mode provided in the above embodiments, and will not be elaborated here.

[0264] The present application also provides a computer program product, including a computer program, which when executed by a processor, implements the steps of the method for confirming the out-of-contact mode as described above.

[0265] The computer program product provided by the present application can solve the technical problem of difficult determination of the out-of-contact mode of customers. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as those of the method for confirming the out-of-contact mode provided in the above embodiments, and will not be elaborated here.

[0266] The above are only partial embodiments of the present application, and do not limit the patent scope of the present application accordingly. Any equivalent structural transformation made by using the content of the specification and drawings of the present application under the technical concept of the present application, or any direct / indirect application in other related technical fields, is included in the patent protection scope of the present application.

Claims

1. A method for confirming a lost connection mode, characterized in that: The lost connection mode confirmation method comprises: Extract the four-dimensional correlation features of the lost customers from each preset data source to form lost data; Inputting each acquisition index under the four-dimensional correlation feature in the lost connection data into a preset lost connection mode classification model to obtain a model classification result; According to the model classification result, the loss mode information of the lost customer is determined, wherein the loss mode is the loss type of the lost customer.

2. The method for confirming the lost connection mode according to claim 1, characterized in that: Before the step of inputting each acquisition index under the four-dimensional correlation feature in the lost connection data into the preset lost connection mode classification model, the step further includes: Using a historical lost contact data set of historically lost contact customers as a sample set, wherein the sample set includes a training set; Dividing the training set into a sub-training set and a sub-testing set, inputting the sub-training set into the loss mode classification model to be constructed to perform iterative training of the model, and performing classification verification on the loss mode classification model to be constructed based on the sub-testing set; When the classification verification result of the loss mode classification model to be constructed does not meet the preset training conditions, calculating the loss function value of each round of training, and adjusting the model parameters in the loss mode classification model to be constructed based on the loss function value; When the classification verification result of the trained loss mode classification model to be constructed meets the preset training conditions, the trained loss mode classification model to be constructed is used as the loss mode classification model.

3. The method for confirming the lost connection mode according to claim 2, characterized in that: The step of performing classification verification on the disconnection mode classification model to be constructed based on the sub-test set includes: Calculating the accuracy, precision, recall, and F1 score of the trained loss mode classification model to be constructed based on the subtest set; When the accuracy rate reaches the preset accuracy rate threshold, the precision rate reaches the preset precision rate threshold, the recall rate reaches the preset recall rate threshold, and the F1 score reaches the preset F1 score threshold, it is determined that the classification verification result of the loss mode classification model to be constructed after training meets the preset training conditions.

4. The method for confirming the lost connection mode according to claim 2, characterized in that: After the step of inputting the sub-training set into the loss mode classification model to be constructed for iterative training of the model, the following steps are included: Calculate the average gain of each candidate sample feature of the samples in the sample set; The candidate sample features whose average gain is higher than the preset gain threshold are used as the collection indicators of the lost connection data.

5. The method for confirming the lost connection mode according to claim 1, characterized in that: The step of determining the lost connection mode information of the lost customer according to the model classification result comprises: In response to a mode selection operation of a user, when the mode selection operation is to close the rule-assisted mode, using the model classification result as the lost connection mode information of the lost connection customer; In the case where the mode selection operation is to turn on the rule-assisted mode, the lost connection mode information of the lost connection customer is determined and output based on the preset mode determination auxiliary rule and the model classification result.

6. The method for confirming the lost connection mode according to claim 5, characterized in that: The step of determining and outputting the lost connection mode information of the lost connection customer based on the preset mode determination auxiliary rule and the model classification result comprises: Determining whether the model classification result is consistent with the rule-assisted result obtained based on the preset mode determination auxiliary rule; When the model classification result is consistent with the rule-assisted result, outputting the model classification result or the rule-assisted result as the lost connection mode information of the lost connection customer; When the model classification result is inconsistent with the rule-assisted result, the model classification result is used as the model prediction result, the rule-assisted result is used as the experience prediction result, and the model prediction result and the experience prediction result are output as the loss mode information of the lost customer.

7. The method for confirming the lost connection mode according to claim 6, characterized in that: Before the step of determining whether the rule classification result is consistent with the auxiliary result obtained based on the preset mode determination auxiliary rule, the following step is further included: Compare the lost connection data of the lost customer with the threshold values ​​of various indicators in the preset mode determination rules; Record the index thresholds satisfied by the lost connection data of the lost customer, determine the lost connection mode corresponding to the lost customer based on the satisfied index thresholds, and use the corresponding lost connection mode as the rule-assisted result.

8. An electronic device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the method for confirming the lost connection mode according to any one of claims 1 to 7.

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, the steps of the lost connection mode confirmation method according to any one of claims 1 to 7 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the method for confirming the lost connection mode as claimed in any one of claims 1 to 7 are implemented.

Images