Customer relationship risk quantitative model research method and system

Abstract

The invention discloses a customer relationship risk quantitative model research method, which comprises the steps of obtaining a full-quantity work order, extracting a first feature value, obtaining a second feature value by constructing a decision tree and fusing a random forest calculation method, obtaining a third feature value according to artificial experience, forming a feature value set, and constructing a customer relationship risk quantitative model. Solidifying the customer relationship risk model to obtain a first customer relationship risk quantification model; the method comprises the following steps: calculating data in a full-quantity work order to obtain a first threshold value and a second threshold value, obtaining a comprehensive risk value by utilizing weight distribution and quantitative scoring of feature values in a feature value set, mapping the comprehensive risk value to correspond to different risk levels, performing reverse solution, optimizing the feature value set to obtain a first feature value set, and obtaining a second feature value set; the first comprehensive risk value is obtained according to weight distribution and quantitative scoring of the first feature value set, and the customer visiting plan is obtained according to different risk levels corresponding to secondary mapping, so that the production and operation cost of a power grid enterprise is reduced, and the visiting efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power services, and in particular to a customer relationship risk quantification model research method and system. Background Art

[0002] The third-party customer satisfaction survey covers four levels of power grid companies, namely, network, province, prefecture and county, every year, involving six provincial-level power grid companies in Guangdong, Guangxi, Yunnan, Guizhou, Hainan and Shenzhen, 65 prefecture-level power grid companies and 513 county-level power grid companies. The survey indicators are divided into three levels, including one first-level indicator, six second-level indicators and 18 third-level indicators. The traditional approach is to arrange the visit plan according to the sampling ratio of the four types of customers involved in the evaluation rules, namely, industrial, commercial, residential and other customers. The number of industrial, commercial and other customers is relatively small, and generally requires full visits. Due to the large number of residential customers in each unit, they can basically only be visited by sampling. However, in the actual visit work, due to factors such as tight time, few employees, large number of customers, and inaccurate visit purposes, it is difficult to achieve full and effective visits even if the number of customers in the first three categories is relatively small. In addition, the randomness of resident sampling makes it difficult for visits to achieve the purpose of repairing customer relationships, and it also wastes human resources.

[0003] Kaili Power Grid has nearly 2 million industrial, commercial, residential and other customers. Each customer may have a different experience in handling business with the power grid company, and their expectations are also different, and their sense of identity with the power grid company may also change. In addition, customer relationships are complex and changeable, and require comprehensive evaluation. For customers with poor experience and high complaints, door-to-door visits need to be organized to repair the relationship. In the case of large numbers and complex evaluation rules, it is necessary to establish a digital model for customer relationship risk assessment to replace the traditional "blind visits" and "ineffective visits". At the same time, the form of visits also needs to be changed from a paper questionnaire for each household to a mobile phone APP tool visit, which can not only save visit time, but also improve the efficiency of collecting, sorting and classifying customer problems after the traditional visit. The model comprehensively analyzes the eigenvalues ​​related to customer value, customer demands, and customer evaluation, takes 10 eigenvalues ​​for repeated verification, solidifies the model, and outputs a scientific and reasonable visit plan to the power supply operation command platform and mobile phone APP visit tool. It truly realizes the accurate search for customer groups with "average" or even "unsatisfied" evaluations. Through digital drive, intelligent classification, risk quantification, and generating customer visit plans based on customer demands, we can create value for both customers and enterprises. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is: the present invention solves the problems in the actual visit work of traditional power grid customers, such as tight time, few employees, large number of customers, and inaccurate visit purpose. Even if the number of customers in the first three categories is relatively small, it is difficult to achieve effective visit in full. In addition, the randomness of resident sampling makes it difficult to achieve the purpose of repairing customer relationships through visits, and also wastes human resources.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a method for researching a customer relationship risk quantification model, comprising:

[0008] Obtain a full set of work orders, extract a first eigenvalue according to a field in the full set of work orders, obtain a second eigenvalue by constructing a decision tree fused with a random forest calculation method, and obtain a third eigenvalue according to manual experience to form a eigenvalue set with the first eigenvalue and the second eigenvalue;

[0009] Constructing a customer relationship risk quantification model according to the feature value set, solidifying the customer relationship risk model, and obtaining a first customer relationship risk quantification model;

[0010] The data in the full amount of work orders are calculated to obtain a first threshold and a second threshold. In the first customer relationship risk quantification model, a comprehensive risk value is obtained by using weight distribution and quantitative scoring of feature values ​​in the feature value set. Through the judgment of the first threshold and the second threshold and the comprehensive risk value, the comprehensive risk value is mapped to corresponding different risk levels, and the obtained risk level result is reversely solved;

[0011] The feature set is trained using the first customer relationship risk quantification model, and the feature value set is optimized by cross-validation of feature values ​​to obtain a first feature value set. A first comprehensive risk value is obtained based on the weight distribution and quantitative scoring of the first feature value set. The first comprehensive risk value is secondary mapped to correspond to different risk levels, and a customer visit plan is obtained based on different risk levels.

[0012] As a preferred solution of the customer relationship risk quantitative model research method of the present invention, the third eigenvalue, the first eigenvalue and the second eigenvalue constitute a eigenvalue set including:

[0013] The characteristic values ​​include electricity consumption category, business type and business subcategory in customer value;

[0014] Information sources, business tags, keywords, urgency and importance in customer requests;

[0015] Number of calls and archived reviews from customer testimonials;

[0016] A customer relationship risk model is constructed through three dimensions and ten eigenvalues.

[0017] As a preferred solution of the customer relationship risk quantification model research method of the present invention, building a decision tree includes:

[0018] Build a decision tree from the starting point of the decision tree, select the appeal type as the division basis of the current node, and divide the appeal type into complaints and other two branches;

[0019] Carry out the first processing of the complaint branch. If it is not satisfactory, select the external channel and internal channel as branches of the first processing node for the second processing;

[0020] Perform a third processing on the external channel based on the result of the second processing, select the professional type as the branch of the third processing node, perform a fourth processing based on the result of the third processing, and select the customer type as the branch of the fourth processing node to perform the corresponding sixth processing;

[0021] For internal channels, the third processing is carried out according to the result of the second processing. If the result of the second processing is other, the customer type is used as the branch of the third processing node for the corresponding sixth processing. If the result of the second processing is unsatisfactory, the professional type is selected as the branch of the third processing node, and the fourth processing is carried out according to the result of the third processing. For cross-professional types, the number of appeals is used as the branch of the fourth processing node, and the fifth processing is carried out according to the result of the fourth processing. For multiple appeals, the customer type is used as the branch of the fifth processing node for the corresponding sixth processing.

[0022] As a preferred solution of the customer relationship risk quantification model research method of the present invention, it also includes:

[0023] Perform the first processing on other branches. If they are not satisfied, perform the third processing. Select the professional type as the branch of the third processing node. Perform the fourth processing according to the third processing result. For cross-professional types, use the number of appeals as the branch of the fourth processing node. Perform the fifth processing according to the fourth processing result. Use the customer type as the branch of the fifth processing node for the corresponding sixth processing.

[0024] If the first processing is other, a third processing is performed. According to the result of the third processing, the customer type is selected as the branch of the third processing node to perform the corresponding sixth processing.

[0025] As a preferred solution of the customer relationship risk quantification model research method of the present invention, mapping the comprehensive risk value to different risk levels includes:

[0026] Determine the key factors that affect risk assessment, define characteristic values ​​for each risk factor, assign weights to each risk factor, multiply each characteristic value by the corresponding weight to obtain the score of each characteristic, and add up the scores of all characteristics to obtain the comprehensive risk value;

[0027] The comprehensive risk value is mapped to different risk levels. When the comprehensive risk value is greater than the first threshold, it is a red risk; when the comprehensive risk value is less than or equal to the first threshold and greater than the second threshold, it is a yellow risk; when the comprehensive risk value is less than or equal to the second threshold, it is a green risk.

[0028] As a preferred solution of the customer relationship risk quantification model research method of the present invention, the reverse solution of the obtained risk level result includes:

[0029] The predicted risk level result is compared with the actual risk level result. If the predicted risk level result is consistent with the actual risk level result, the first customer relationship risk quantification model predicts accurately.

[0030] As a preferred solution of the customer relationship risk quantification model research method of the present invention, wherein: optimizing the feature value set by cross-validation feature value method includes:

[0031] Divide the first eigenvalue set into K subsets, select one subset as the test set, and K-1 subsets as the training set, repeat K times, and select a different subset as the test set each time;

[0032] Each trained model is evaluated on the test set to obtain the first performance indicator of the customer portrait model;

[0033] Divide the first feature value set into time windows in chronological order, select one time window as the test set, and the remaining time windows as the training set. Repeat this process, selecting a different time window as the test set each time.

[0034] Each trained model is evaluated on the test set to obtain the second performance indicator of the customer portrait model;

[0035] By combining the first performance indicator and the second performance indicator of the customer portrait model, an optimal parameter combination is obtained, which is the second eigenvalue set.

[0036] In a second aspect, the present invention provides a system for studying a customer relationship risk quantification model, comprising:

[0037] A feature value extraction module is used to obtain a full set of work orders, extract a first feature value according to a field in the full set of work orders, obtain a second feature value by building a decision tree fusion random forest calculation method, and obtain a third feature value according to manual experience to form a feature value set with the first feature value and the second feature value;

[0038] A model building module, used to build a customer relationship risk quantification model according to the feature value set, solidify the customer relationship risk model, and obtain a first customer relationship risk quantification model;

[0039] a calculation module, configured to calculate the data in the full work order to obtain a first threshold and a second threshold, obtain a comprehensive risk value by using the weight distribution and quantitative scoring of the eigenvalues ​​in the eigenvalue set in the first customer relationship risk quantification model, map the comprehensive risk value to corresponding different risk levels through the judgment of the first threshold and the second threshold and the comprehensive risk value, and perform reverse solution on the obtained risk level result;

[0040] A mapping module is used to train the feature set using the first customer relationship risk quantification model, optimize the feature value set by cross-validating the feature value method to obtain a first feature value set, obtain a first comprehensive risk value based on the weight distribution and quantitative scoring of the first feature value set, secondary map the first comprehensive risk value to correspond to different risk levels, and obtain a customer visit plan based on different risk levels.

[0041] In a third aspect, the present invention provides a computing device, comprising:

[0042] Memory and processor;

[0043] The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the customer relationship risk quantitative model research method are implemented.

[0044] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the customer relationship risk quantification model research method.

[0045] Beneficial effects of the invention: The customer relationship risk quantification model research method proposed in the invention realizes the efficient operation of digital empowerment business, constructs a customer relationship risk quantification digital model, can accurately find customer groups with poor "sense of gain" and high "complaint" in the electricity use process, and provide a basis for business risk analysis, trend forecasting, and strategy formulation, so as to achieve business quality improvement, cost reduction, and efficiency improvement. Through digital drive, intelligent classification, and risk quantification, it can improve service quality, optimize the electricity business environment, improve customer satisfaction, reduce the production and operation costs of power grid enterprises, improve visit efficiency, and improve relationship repair effects, thereby generating direct and indirect economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0047] Figure 1 This is an overall flow chart of a customer relationship risk quantification model research method provided by the present invention. DETAILED DESCRIPTION

[0048] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0049] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0050] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0051] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0052] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0053] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0054] Example 1

[0055] Reference Figure 1 , as an embodiment of the present invention, provides a method for researching a customer relationship risk quantification model, comprising:

[0056] S100: Obtain all work orders, extract a first eigenvalue according to the fields in the full work orders, obtain a second eigenvalue by building a decision tree fusion random forest calculation method, and obtain a third eigenvalue according to manual experience to form a eigenvalue set with the first eigenvalue and the second eigenvalue;

[0057] In the embodiment of the present application, the characteristic value includes the electricity usage category, business type and business subcategory in the customer value;

[0058] Information sources, business tags, keywords, urgency and importance in customer requests;

[0059] Number of calls and archived reviews from customer testimonials;

[0060] A customer relationship risk model is constructed through three dimensions and ten eigenvalues.

[0061] Specifically, before building the model, all work order data is preprocessed, including removing duplicates and missing values, extracting some eigenvalues ​​from the fields of all work orders, and then analyzing and screening them by decision tree based on the types of customer demands in the work order data to obtain some eigenvalues. Other eigenvalues ​​are derived based on manual experience to form ten eigenvalues. Then, the decision tree of each dimension is solved and fused with the random forest to calculate ten eigenvalues, and then scores are assigned to each sub-item in each eigenvalue.

[0062] The present invention preferably selects these ten characteristic values, based on the long-term work experience of business experts on the business, and adds the ten characteristic values ​​in the three dimensions of the customer to calculate a more accurate customer risk level. If other characteristic values ​​are selected, it is impossible to accurately judge the risk level;

[0063] It should be noted that by extracting features from different angles and methods, the information in the data can be captured more comprehensively, thereby improving the prediction accuracy of the model. The diversity of feature values ​​helps the model to generalize better when facing new data and reduce the risk of overfitting.

[0064] S200: constructing a customer relationship risk quantification model according to the characteristic value set, solidifying the customer relationship risk model, and obtaining a first customer relationship risk quantification model;

[0065] In the embodiment of the present application, a decision tree is constructed from the starting point of the decision tree, and the appeal type is selected as the basis for dividing the current node, and the appeal type is divided into two branches: complaint and other branches;

[0066] Carry out the first processing of the complaint branch. If it is not satisfactory, select the external channel and internal channel as branches of the first processing node for the second processing;

[0067] Perform a third processing on the external channel based on the result of the second processing, select the professional type as the branch of the third processing node, perform a fourth processing based on the result of the third processing, and select the customer type as the branch of the fourth processing node to perform the corresponding sixth processing;

[0068] For internal channels, the third processing is carried out according to the result of the second processing. If the result of the second processing is other, the customer type is used as the branch of the third processing node for the corresponding sixth processing. If the result of the second processing is unsatisfactory, the professional type is selected as the branch of the third processing node, and the fourth processing is carried out according to the result of the third processing. For cross-professional types, the number of appeals is used as the branch of the fourth processing node, and the fifth processing is carried out according to the result of the fourth processing. For multiple appeals, the customer type is used as the branch of the fifth processing node for the corresponding sixth processing.

[0069] Perform the first processing on other branches. If they are not satisfied, perform the third processing. Select the professional type as the branch of the third processing node. Perform the fourth processing according to the third processing result. For cross-professional types, use the number of appeals as the branch of the fourth processing node. Perform the fifth processing according to the fourth processing result. Use the customer type as the branch of the fifth processing node for the corresponding sixth processing.

[0070] If the first processing is other, a third processing is performed. According to the result of the third processing, the customer type is selected as the branch of the third processing node to perform the corresponding sixth processing.

[0071] Exemplarily, the nth processing is set to Xn, where n is a natural number, and the appeal type is selected as the division basis of the current node, and the appeal type is divided into two branches: complaint and other;

[0072] Conduct X1 on the complaint branch. If not satisfactory, select external channel 12398 and internal channel 95598 as branches of X1 to conduct X2;

[0073] Perform X2 on external channels 12398 to get "unsatisfied" and "others", perform X3 on these two branches, get "cross-professional" and "non-cross-professional" branches, perform X4 on these two branches respectively, and get "industrial", "commercial" and "others" branches, perform X6;

[0074] Conduct X2 on internal channel 95598 to get "unsatisfied" and "others", conduct X3 on "others", and get three branches of "industry", "commercial" and "others", and conduct X6; Conduct X3 on "unsatisfied", and get two branches of "cross-professional" and "non-cross-professional", and conduct X4 on these two branches; for "cross-professional", get two branches of "multiple times" and "<2 times", and conduct X5 on these two branches; for the "multiple times" branch, get three branches of "industry", "commercial" and "others", and conduct X6;

[0075] For the "other" branch, perform X1 to get "unsatisfied" and "others", perform X3 on the "unsatisfied" branch, and get two branches of "cross-professional" and "non-cross-professional", and perform X4 on these two branches. For "cross-professional", get two branches of "multiple times" and "<2 times", and perform X5 on these two branches to get three branches of "industrial", "commercial" and "others", and perform X6; and for the "other" branch, perform X3 to get three branches of "industrial", "commercial" and "others", and perform X6.

[0076] It should be noted that the decision-makers pay attention to trends, forecasts and the impact of macro indicators. Based on the results of the visits, they regularly evaluate the effectiveness of customer relationship repair, assess the changing trends of risk and satisfaction, and adjust and improve strategies in a timely manner to further optimize the business environment and improve customer satisfaction.

[0077] S300: Calculate the data in all work orders to obtain a first threshold and a second threshold. In the first customer relationship risk quantification model, use the weight distribution and quantitative scoring of the eigenvalues ​​in the eigenvalue set to obtain a comprehensive risk value. By judging the first threshold and the second threshold and the comprehensive risk value, map the comprehensive risk value to different risk levels, and perform reverse solution on the obtained risk level result.

[0078] In the embodiment of the present application, the key factors affecting the risk assessment are determined, a characteristic value is defined for each risk factor, and a weight is assigned to each risk factor. Each characteristic value is multiplied by the corresponding weight to obtain a score for each characteristic, and the scores of all characteristics are added together to obtain a comprehensive risk value.

[0079] The comprehensive risk value is mapped to different risk levels. When the comprehensive risk value is greater than the first threshold, it is a red risk; when the comprehensive risk value is less than or equal to the first threshold and greater than the second threshold, it is a yellow risk; when the comprehensive risk value is less than or equal to the second threshold, it is a green risk.

[0080] Exemplarily, the data in all work orders are calculated to obtain a first threshold of 30 points and a second threshold of 20 points. If the comprehensive risk value is greater than 30 points, it is determined as a "red risk"; 20<risk value≤30 points is determined as a "yellow risk"; and risk value≤20 points is determined as a "green risk".

[0081] In the embodiment of the present application, the predicted risk level result is compared with the actual risk level result. If the predicted risk level result is consistent with the actual risk level result, the first customer relationship risk quantification model predicts accurately.

[0082] Specifically, through reverse solving, key fields in the information source, business tag fields, key fields in the call content, and work orders in the appeal content that require cross-disciplinary communication and coordination are verified;

[0083] For example, the verification of key fields in the information source includes: the information source is divided into three types: 12398, public opinion express, and remote service channels. Based on business experience, the more sensitive 12398 should be classified as "red risk". The calculation output result of the digital analysis model: All 18 12398 work orders in 2023 are within the "red risk" level, and the calculation result is consistent with the business experience prediction;

[0084] Verification of the business tag field includes that the business tag is a special mark for the agent to make a service warning based on the urgency and importance of the caller's problem, so that the responsible unit can contact and deal with it in time. The calculation output result of the digital analysis model is: in 2023, there are 523 work orders marked with the service warning field, including 109 "red risks", 393 "yellow risks", and 21 "green risks". From the perspective of risk level classification, medium and high risks account for 96%, which is a large proportion and conforms to the logic of this field.

[0085] Verification of key fields in the call content includes the preset score of 10 points for the key field with the highest risk in the call content. All 56 work orders mentioned "12398", of which 38 were "red risks", 18 were "yellow risks", and 0 were "green risks". The proportion of medium and high risks was 100%, which is in line with the logic of giving this field the highest score.

[0086] Verification of the comprehensive calculation results of all 77,411 work orders: Through the comprehensive calculation scores and output risk levels of all work orders, there are 189 "red risks", 4,853 "yellow risks", and 72,369 "green risks". From the perspective of the content summary and data analysis, the overall results are comprehensive, scientific, and reasonable.

[0087] Verification of work orders involving cross-disciplinary communication and coordination in the content of the appeal includes extracting 1,954 work orders that may involve cross-disciplinary communication and coordination in the work order business subcategories, of which 47 are "red risk", 1,744 are "yellow risk", and 163 are "green risk". The proportion of medium and high risk in this business type accounts for 36% of the total medium and high risk work orders. This shows that: the demand for cross-disciplinary communication and cooperation involved in customer appeals is relatively large, highlighting that the resources and capabilities for solving problems need to match the corresponding level of personnel, which is in line with the definition of the customer relationship risk model.

[0088] Specifically, in the early stage, a single decision tree was used to analyze 77,411 archived work orders for the whole year of 2023. It was found that the missing or unclear feature values ​​of sensitive data led to poor classification and even affected the classification results. The calculation method was not scientific enough. Therefore, the decision trees were reversed one by one again, and the random forest calculation method was integrated to obtain the probability data of each branch.

[0089] It should be noted that the executive layer pays attention to the detailed content. The customer visit plan generated by the model is simple and clear, and only includes the "red, yellow, and green" levels of customer relationship risks, as well as the "demand summary". The customer managers of the district and county power supply bureaus can clearly see the goals and purposes of the visits, which can improve the efficiency of visits and the effect of relationship repair. By paying attention to the visit completion rate, customer demand problem resolution rate, return visit satisfaction rate, and changes in customer complaints through the Yunjing visual dashboard, the convenience and effectiveness of management personnel's supervision can be improved. Based on the results of the visits, the effect of customer relationship repair is regularly evaluated, the changing trends of risk and satisfaction are evaluated, and the improvement strategy is adjusted in a timely manner to further optimize the business environment and improve customer satisfaction.

[0090] S400: training the feature set using the first customer relationship risk quantification model, optimizing the feature value set by cross-validating the feature value method to obtain a first feature value set, obtaining a first comprehensive risk value according to the weight distribution and quantitative scoring of the first feature value set, mapping the first comprehensive risk value to correspond to different risk levels, and obtaining a customer visit plan according to the different risk levels;

[0091] In the embodiment of the present application, the first feature value set is divided into K subsets, one subset is selected as a test set, K-1 subsets are selected as training sets, and the process is repeated K times, with a different subset selected as a test set each time;

[0092] Each trained model is evaluated on the test set to obtain the first performance indicator of the customer portrait model;

[0093] Divide the first feature value set into time windows in chronological order, select one time window as the test set, and the remaining time windows as the training set. Repeat this process, selecting a different time window as the test set each time.

[0094] Each trained model is evaluated on the test set to obtain the second performance indicator of the customer portrait model;

[0095] By combining the first performance indicator and the second performance indicator of the customer portrait model, an optimal parameter combination is obtained, which is the second eigenvalue set.

[0096] Specifically, the present invention preferably uses a method of horizontal and vertical cross-validation of characteristic values, and the method of validating characteristic values ​​may also be cluster analysis and principal component analysis;

[0097] It should be noted that through horizontal and vertical cross-validation, the performance of different eigenvalues ​​on different data subsets and time points can be evaluated, thereby verifying the stability and importance of eigenvalues, which is crucial for building a reliable model and continuously optimizing the rationality of parameters so that the model can better fit the given data set. This method helps to verify the scientificity and accuracy of the model and ensure the performance of the model in practical applications. However, due to certain limitations of cluster analysis and PCA, the accuracy of the model in predicting customer relationship risks is reduced and the complexity of model construction is increased.

[0098] The above is a schematic scheme of a method for studying a customer relationship risk quantification model in this embodiment. It should be noted that the technical scheme of the device for studying a customer relationship risk quantification model is the same as the technical scheme of the method for studying a customer relationship risk quantification model. For details not described in detail in the technical scheme of the device for studying a customer relationship risk quantification model in this embodiment, please refer to the description of the technical scheme of the method for studying a customer relationship risk quantification model.

[0099] The device for studying the customer relationship risk quantification model in this embodiment includes:

[0100] A feature value extraction module is used to obtain a full set of work orders, extract a first feature value according to a field in the full set of work orders, obtain a second feature value by building a decision tree fusion random forest calculation method, and obtain a third feature value according to manual experience to form a feature value set with the first feature value and the second feature value;

[0101] A model building module, used to build a customer relationship risk quantification model according to the feature value set, solidify the customer relationship risk model, and obtain a first customer relationship risk quantification model;

[0102] a calculation module, configured to calculate the data in the full work order to obtain a first threshold and a second threshold, obtain a comprehensive risk value by using the weight distribution and quantitative scoring of the eigenvalues ​​in the eigenvalue set in the first customer relationship risk quantification model, map the comprehensive risk value to corresponding different risk levels through the judgment of the first threshold and the second threshold and the comprehensive risk value, and perform reverse solution on the obtained risk level result;

[0103] A mapping module is used to train the feature set using the first customer relationship risk quantification model, optimize the feature value set by cross-validating the feature value method to obtain a first feature value set, obtain a first comprehensive risk value based on the weight distribution and quantitative scoring of the first feature value set, secondary map the first comprehensive risk value to correspond to different risk levels, and obtain a customer visit plan based on different risk levels.

[0104] This embodiment also provides a computing device suitable for studying the customer relationship risk quantification model, including:

[0105] Memory and processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the customer relationship risk quantification model research method proposed in the above embodiment.

[0106] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for studying a customer relationship risk quantification model proposed in the above embodiment is implemented.

[0107] The storage medium proposed in this embodiment and the method for implementing a customer relationship risk quantification model research proposed in the above embodiment belong to the same inventive concept. The technical details not described in detail in this embodiment can be found in the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0108] Through the above description of the implementation mode, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course can also be implemented by hardware, but in many cases the former is a better implementation mode. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ReadOnly, Memory, ROM), random access memory (RandomAccess Memory, RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform the methods of various embodiments of the present invention.

[0109] Example 2

[0110] This embodiment is different from the first embodiment in that it provides a verification test of a customer relationship risk quantification model research method to verify and illustrate the technical effects used in this method.

[0111] Taking the traditional visit of a power supply bureau as an example, the total number of customers of the bureau by 2024 is 72,000, including 57,000 residential households, 7,000 commercial households, 4,000 industrial households, and 4,000 other households. If the traditional visit method is used, the staff needs to print out the relevant information of the visited users in the office and bring it to the site. There is no targeted and differentiated visit to customer relations, and then the problems found in the visit are brought back to the office for re-summarization and sorting. It will take at least 6 months to complete the full visit work.

[0112] Taking the calculation of the customer relationship risk quantification model of a power supply bureau as an example, 1224 full work orders of the bureau in the first half of 2024 were extracted, and the customer relationship risk quantification model was used for calculation to generate a "red, yellow, and green" customer relationship risk visit plan for 253 households, among which: 3 households with "red risk" risk level distribution have poor electricity experience and high complaints. The problems encountered have not been effectively solved, or the handling is not very satisfactory. The bureau leaders need to organize relevant professionals to conduct door-to-door visits to understand and solve problems in a timely manner to repair customer relationships; 36 households with "yellow risk" have poor electricity experience and certain complaints. The marketing service team leader or the director of the power supply station needs to conduct door-to-door visits to understand and solve problems in a timely manner to repair customer relationships; 214 households with "green risk" have relatively harmonious customer relationships and are provided to customer managers of various units to visit and repair customer relationships. In the past two weeks, the bureau has carried out on-site visits in a hierarchical manner, and read data and uploaded visit content in real time through the on-site visit tool APP to accurately, intuitively and quickly solve user problems and improve users' electricity experience. A total of 214 on-site visits to “red, yellow, and green” risk customers;

[0113] The Power Supply Bureau believes that risk quantification is more in line with the actual situation, improves visit efficiency, and has obvious results in repairing customer relationships;

[0114] The "demand summary" is just right. You can understand the problems reported by customers at a glance through the mobile phone visit APP tool, conduct accurate visits, and quickly organize and collect user feedback issues and solve them in a timely manner;

[0115] The division of risk levels into “red, yellow, and green” makes the division of responsibilities among bureau leaders, department managers, and customer managers more clear, and the difficulty of repairing the relationship is clear at a glance. It is recommended that it be promoted and applied throughout the entire network.

[0116] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A customer relationship risk quantification model research method, characterized in that: include: Obtain a full set of work orders, extract a first eigenvalue according to a field in the full set of work orders, obtain a second eigenvalue by constructing a decision tree fused with a random forest calculation method, and obtain a third eigenvalue according to manual experience to form a eigenvalue set with the first eigenvalue and the second eigenvalue; Constructing a customer relationship risk quantification model according to the feature value set, solidifying the customer relationship risk model, and obtaining a first customer relationship risk quantification model; The data in the full amount of work orders are calculated to obtain a first threshold and a second threshold. In the first customer relationship risk quantification model, a comprehensive risk value is obtained by using weight distribution and quantitative scoring of feature values ​​in the feature value set. Through the judgment of the first threshold and the second threshold and the comprehensive risk value, the comprehensive risk value is mapped to corresponding different risk levels, and the obtained risk level result is reversely solved; The feature set is trained using the first customer relationship risk quantification model, and the feature value set is optimized by cross-validation of feature values ​​to obtain a first feature value set. A first comprehensive risk value is obtained based on the weight distribution and quantitative scoring of the first feature value set. The first comprehensive risk value is secondary mapped to correspond to different risk levels, and a customer visit plan is obtained based on different risk levels.

2. The customer relationship risk quantification model research method according to claim 1, characterized in that: The third eigenvalue, the first eigenvalue, and the second eigenvalue constitute a eigenvalue set including: The characteristic values ​​include electricity consumption category, business type and business subcategory in customer value; Information sources, business tags, keywords, urgency and importance in customer requests; Number of calls and archived reviews from customer testimonials; A customer relationship risk model is constructed through three dimensions and ten eigenvalues.

3. The customer relationship risk quantification model research method according to claim 1 or 2, characterized in that: Building a decision tree involves: Build a decision tree from the starting point of the decision tree, select the appeal type as the division basis of the current node, and divide the appeal type into complaints and other two branches; Carry out the first processing of the complaint branch. If it is not satisfactory, select the external channel and internal channel as branches of the first processing node for the second processing; Perform a third processing on the external channel based on the result of the second processing, select the professional type as the branch of the third processing node, perform a fourth processing based on the result of the third processing, and select the customer type as the branch of the fourth processing node to perform the corresponding sixth processing; For internal channels, the third processing is carried out according to the result of the second processing. If the result of the second processing is other, the customer type is used as the branch of the third processing node for the corresponding sixth processing. If the result of the second processing is unsatisfactory, the professional type is selected as the branch of the third processing node, and the fourth processing is carried out according to the result of the third processing. For cross-professional types, the number of appeals is used as the branch of the fourth processing node, and the fifth processing is carried out according to the result of the fourth processing. For multiple appeals, the customer type is used as the branch of the fifth processing node for the corresponding sixth processing.

4. The customer relationship risk quantification model research method according to claim 3, characterized in that: Also includes: Perform the first processing on other branches. If they are not satisfied, perform the third processing. Select the professional type as the branch of the third processing node. Perform the fourth processing according to the third processing result. For cross-professional types, use the number of appeals as the branch of the fourth processing node. Perform the fifth processing according to the fourth processing result. Use the customer type as the branch of the fifth processing node for the corresponding sixth processing. If the first processing is other, a third processing is performed. According to the result of the third processing, the customer type is selected as the branch of the third processing node to perform the corresponding sixth processing.

5. The customer relationship risk quantification model research method according to claim 4, characterized in that: Mapping the comprehensive risk value to different risk levels includes: Determine the key factors that affect risk assessment, define characteristic values ​​for each risk factor, assign weights to each risk factor, multiply each characteristic value by the corresponding weight to obtain the score of each characteristic, and add up the scores of all characteristics to obtain the comprehensive risk value; The comprehensive risk value is mapped to different risk levels. When the comprehensive risk value is greater than the first threshold, it is a red risk; when the comprehensive risk value is less than or equal to the first threshold and greater than the second threshold, it is a yellow risk; when the comprehensive risk value is less than or equal to the second threshold, it is a green risk.

6. The customer relationship risk quantification model research method according to claim 5, characterized in that: The reverse solution of the obtained risk level results includes: The predicted risk level result is compared with the actual risk level result. If the predicted risk level result is consistent with the actual risk level result, the first customer relationship risk quantification model predicts accurately.

7. The customer relationship risk quantification model research method according to claim 6, characterized in that: Methods for optimizing feature value sets by cross-validating feature values ​​include: Divide the first eigenvalue set into K subsets, select one subset as the test set, and K-1 subsets as the training set, repeat K times, and select a different subset as the test set each time; Each trained model is evaluated on the test set to obtain the first performance indicator of the customer portrait model; Divide the first feature value set into time windows in chronological order, select one time window as the test set, and the remaining time windows as the training set, repeat this process, and select a different time window as the test set each time; Each trained model is evaluated on the test set to obtain the second performance indicator of the customer portrait model; By combining the first performance indicator and the second performance indicator of the customer portrait model, an optimal parameter combination is obtained, which is the second eigenvalue set.

8. A system for researching a customer relationship risk quantitative model, characterized in that: include: A feature value extraction module is used to obtain a full set of work orders, extract a first feature value according to a field in the full set of work orders, obtain a second feature value by building a decision tree fusion random forest calculation method, and obtain a third feature value according to manual experience to form a feature value set with the first feature value and the second feature value; A model building module, used to build a customer relationship risk quantification model according to the feature value set, solidify the customer relationship risk model, and obtain a first customer relationship risk quantification model; a calculation module, configured to calculate the data in the full work order to obtain a first threshold and a second threshold, obtain a comprehensive risk value by using the weight distribution and quantitative scoring of the eigenvalues ​​in the eigenvalue set in the first customer relationship risk quantification model, map the comprehensive risk value to corresponding different risk levels through the judgment of the first threshold and the second threshold and the comprehensive risk value, and perform reverse solution on the obtained risk level result; A mapping module is used to train the feature set using the first customer relationship risk quantification model, optimize the feature value set by cross-validating the feature value method to obtain a first feature value set, obtain a first comprehensive risk value based on the weight distribution and quantitative scoring of the first feature value set, secondary map the first comprehensive risk value to correspond to different risk levels, and obtain a customer visit plan based on different risk levels.

9. An electronic device, characterized in that: The device comprises: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images