User churn early warning system and method based on machine learning

Through a user churn warning system based on machine learning, combining the characteristic data of telecom operators and users, a competitiveness index and satisfaction change rate are generated, and the user churn probability is predicted, which solves the problem of insufficient consideration of telecom operators and users in the existing technology. The problem of interactive relationships improves the accuracy of churn prediction.

CN118446739BActive Publication Date: 2024-11-22HEFEI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410669189.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-28
Publication Date
2024-11-22
Estimated Expiration
2044-05-28

AI Technical Summary

Technical Problem

The existing user churn prediction model fails to fully consider the interaction between telecom operators and users, resulting in a decrease in the accuracy of churn prediction.

Method used

Using a user churn warning system based on machine learning, collects characteristic data of telecom operators and users, generates coefficients of network quality, price factors, competitive environment and customer relationships, and combines multi-layer perceptron deep neural network model to analyze telecom operators' Competitiveness index and user satisfaction change rate predict the probability of user churn and issue early warning signals.

Benefits of technology

The prediction accuracy of user churn is improved. By comprehensively analyzing the characteristic data of telecom operators and users, considering the interactive relationship between users and operators, the accuracy of prediction is enhanced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118446739B_ABST
    Figure CN118446739B_ABST
Patent Text Reader

Abstract

The present invention provides a user churn warning system and method based on machine learning, which relates to the technical field of user churn warning, and includes: a data collection module, which is used to collect characteristic data of telecommunications operators and users in the Mth month and the Nth month, wherein, #imgabs0#, the characteristic data of telecommunications operators include network quality, price factors, competitive environment and customer relations; the influencing parameters of network quality include signal strength and data transmission speed; the influencing parameters of price factors include the package price and flow fee of the party; the influencing parameters of the competitive environment include market share and competitor's package price; the influencing parameters of customer relations include complaint handling time and the number of complaints per unit time. When predicting the situation of user churn, the present invention collects characteristic data of telecommunications operators and users, considers the interaction and relationship between users and operators, comprehensively analyzes the situation of user churn, and can improve the prediction accuracy of user churn.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of user churn warning technology, and in particular to a user churn warning system and method based on machine learning. Background Art

[0002] Customer churn is a pain point in the business development of most companies. It refers to customers abandoning the company's existing products or services for some reason. In the context of increasingly fierce market competition, services are constantly upgraded and diversified, and it is much more difficult to acquire new customers than to retain old customers. Therefore, it is very necessary for companies to predict customer churn intentions in advance and retain customers by improving services. Especially for telecom operators, there have been many studies on customer churn cases. There are methods that use traditional statistical theories to analyze and summarize customer status and judge user churn; there are also methods that use neural networks to train historical user data sets to predict customer churn.

[0003] In the prior art, publication number CN115619448A discloses a user churn prediction method, apparatus, computer equipment and storage medium, comprising the following steps: acquiring feature data of batch users according to a model pre-training request, wherein the pre-training request includes an acquisition address corresponding to the feature data of the batch users, and the feature data includes gender, age, job category, marital status, place of residence, family income level, credit rating, education level, purchased insurance type, payment method, price, purchase channel, and number of payments; expanding the feature data according to the SMOTE algorithm; inputting the pre-processed feature data as a training set into an initialized customer churn prediction model for pre-training, and obtaining a pre-trained customer churn prediction model; acquiring feature data of a target user according to a model prediction request, wherein the prediction request includes an acquisition address corresponding to the feature data of the target user; inputting the feature data as a prediction set into the customer churn prediction model for prediction, and obtaining a prediction result; identifying the churn status of the target user according to the prediction result, wherein the churn status includes a churned status and a non-churned status. The present application uses the SMOTE algorithm to ensure the sufficiency and scientificity of feature data during pre-training, and constructs a random forest tree prediction model before pre-training, and uses an improved particle swarm algorithm to tune the parameters of the random forest tree prediction model during pre-training, thereby ensuring the high availability and accuracy of the prediction model.

[0004] However, there are still the following deficiencies: As can be seen from the above statements, the above scheme only collects characteristic data of users, obtains customer churn prediction models, and makes churn status judgments, but does not collect characteristic data of telecom operators. For example, it does not consider data specific to telecom operators, such as network quality, price factors, competitive environment, and customer relationships, etc., which are usually closely related to user churn, resulting in the churn prediction model failing to fully consider the interaction and relationship between users and operators, thereby resulting in a decrease in the prediction accuracy of churn.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0006] The purpose of the present invention is to provide a user churn warning system and method based on machine learning to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A user churn early warning system based on machine learning, comprising:

[0009] The data collection module is used to collect the characteristic data of telecom operators and users in the Mth month and the Nth month.

[0010] Telecom operators’ characteristic data include network quality, price factors, competitive environment and customer relations;

[0011] Parameters affecting network quality include signal strength and data transmission speed;

[0012] The influencing parameters of price factors include the price of one's own package and traffic fee;

[0013] The parameters influencing the competitive environment include market share and competitor package prices;

[0014] The influencing parameters of customer relations include complaint handling time and number of complaints per unit time;

[0015] User characteristic data includes customer experience, purchase frequency and activity;

[0016] Wherein, (NM) is greater than or equal to 3;

[0017] The data processing module is used to process the signal strength and data transmission speed to generate the network quality coefficient, process the package price and traffic fee of the party to generate the price factor coefficient, process the market share and the package price of competitors to generate the competition environment coefficient, and process the complaint processing time and the number of complaints per unit time to generate the customer relationship coefficient;

[0018] A data analysis module is used to process and analyze the network quality coefficient, price factor coefficient, competition environment coefficient and customer relationship coefficient to generate a competitiveness index of telecom operators;

[0019] A model building module is used to build a machine learning network model, taking the characteristic data of users in the Mth month as a training set and the corresponding satisfaction coefficient of users in the Mth month as a label, and inputting them into the machine learning network model for training;

[0020] A coefficient acquisition module is used to input the characteristic data of the users in the Nth month as a training set into the machine learning network model to obtain the satisfaction coefficient of the users in the Nth month;

[0021] A change rate acquisition module is used to process the user satisfaction coefficient of the Mth month and the user satisfaction coefficient of the Nth month to generate a change rate of the user satisfaction coefficient from the Mth month to the Nth month;

[0022] The prediction module is used to process data and conduct correlation analysis on the change rate of the competitiveness index of the telecommunications operator and the user's satisfaction coefficient, generate a correlation coefficient, obtain the probability of user churn based on the correlation coefficient between the change rate of the competitiveness index and the user's satisfaction coefficient, predict the situation of telecommunications user churn based on the interval range of the probability of user churn, and issue an early warning signal.

[0023] Furthermore, the signal strength and data transmission speed are processed to generate the network quality coefficient, based on the following formula:

[0024]

[0025] Among them, XS WZ is the network quality coefficient, S is the signal strength, S′ is the normalized value of the signal strength, V is the data transmission speed, V′ is the normalized value of the data transmission speed, α is the parameter for adjusting the strength of S′ and V′ in the exponential function, e -α(S′+V′) is an exponential function used to balance the impact of normalized signal strength and data transmission speed on the overall network quality, and log(1+SV) is a composite function used to enhance the impact of the product of signal strength and data transmission speed;

[0026] Using the above formula, we can get the network quality coefficients of the Mth month and the Nth month, that is: and

[0027] Furthermore, the package price and traffic fee are processed to generate the price factor coefficient, based on the following formula:

[0028]

[0029] Among them, XS JY is the price factor coefficient, P is the package price of the party, P′ is the normalized value of the package price of the party, F is the traffic fee, F′ is the normalized value of the traffic fee, β is the parameter for adjusting the price factor coefficient of the package price and traffic fee of the party, λ is the weight coefficient, which is used to adjust the relative importance of the package price P′ and traffic fee F′ in the calculation;

[0030] Using the above formula, we can get the price factor coefficients for the Mth month and the Nth month, that is: and

[0031] Furthermore, the market share and competitor package prices are processed to generate the competitive environment coefficient, based on the following formula:

[0032]

[0033] Among them, XS JH is the competitive environment coefficient, n is the total number of competitors in the market, J i is the package price of the ith competitor, M i is the market share of the ith competitor, δ is the parameter that adjusts the effect of the competitor’s package price on the competitive environment coefficient, and ε is the parameter that adjusts the effect of the market share on the competitive environment coefficient;

[0034] Using the above formula, we can obtain the competitive environment coefficients for the Mth month and the Nth month, namely: and

[0035] Furthermore, the complaint handling time and the number of complaints per unit time are processed to generate the customer relationship coefficient, based on the following formula:

[0036]

[0037] T(t)=η·T(t-1)+(1-η)·t cut

[0038] C(t)=log(1+c t )

[0039] Among them, XS KG is the customer relationship coefficient, c t is the number of complaints per unit time, C(t) is the logarithmic transformation of the number of complaints per unit time, t is the complaint processing time, T(t) is the average complaint processing time, is the adjustment coefficient, which is used to balance the impact of complaint processing time and number of complaints; μ is the weight coefficient, which determines the relative weight of C(t); cut is the current complaint handling time, η is the smoothing coefficient, and T(t-1) is the smoothed estimated value at the previous time point (t-1);

[0040] Using the above formula, we can get the customer relationship coefficients of the Mth month and the Nth month, that is, and

[0041] Furthermore, the network quality coefficient, price factor coefficient, competition environment coefficient and customer relationship coefficient are processed and correlated to generate the competitiveness index of telecom operators, based on the following formula:

[0042]

[0043] Among them, ZS is the competitiveness index of telecom operators, R is the number of factors, R = 4;

[0044] Using the above formula, we can obtain the competitiveness index of telecom operators in the Mth month and the Nth month, i.e., ZS M and ZS N .

[0045] Furthermore, the machine learning network model is composed of a deep neural network based on a multilayer perceptron, and the deep neural network of the multilayer perceptron includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer and an output layer. The first hidden layer, the second hidden layer and the third hidden layer each have at least two neurons, and each uses ReLU as an activation function.

[0046] Furthermore, the user satisfaction coefficients of the Mth month and the Nth month are processed to generate the change rate of the user satisfaction coefficients from the Mth month to the Nth month, according to the following formula:

[0047]

[0048] Among them, BL is the change rate of the user's satisfaction coefficient, S N is the user satisfaction coefficient of the Nth month, S M is the user satisfaction coefficient in the Mth month.

[0049] Furthermore, the change rates of the competitiveness index of telecom operators and the satisfaction coefficient of users are processed and correlated to generate a correlation coefficient, based on the following formula:

[0050]

[0051] Among them, r is the correlation coefficient between the competitiveness index and the change rate of the user satisfaction coefficient, ZS N is the competitiveness index of the telecom operator in the Nth month, BL is the change rate of the user satisfaction coefficient from the Mth month to the Nth month, is the average competitiveness index from month M to month N, is the mean of the change rate of the satisfaction coefficient from the Mth month to the Nth month;

[0052] When r<0, it means that the competitiveness index and the user satisfaction coefficient have opposite trends. The improvement of competitiveness will lead to a decrease in user satisfaction, which will increase the user churn rate.

[0053] When r = 0, it means that there is no correlation between the competitiveness index and the rate of change of the user satisfaction coefficient;

[0054] When r>0, it means that the competitiveness index and the user satisfaction coefficient have the same changing trend. The improvement of competitiveness will lead to the improvement of user satisfaction, thereby reducing the user churn rate.

[0055] According to the correlation coefficient between the competitiveness index and the change rate of the user's satisfaction coefficient, the probability of user churn is obtained according to the following formula:

[0056]

[0057] Where P is the probability of user churn, a<0, b>0;

[0058] According to the interval range of the probability of user churn, the process of predicting the loss of telecommunications users and issuing early warning signals is as follows:

[0059] When r<0, 0.5 <P<1;

[0060] When r = 0, P = 0.5;

[0061] When r>0, 0 <P<0.5;

[0062] When P ranges from 0 to 1, 0 means there is no possibility of user loss, and 1 means all users will be lost;

[0063] In summary, when the probability of user churn is 0 < P < 0.5, less than half of the users may churn, and a low-risk warning signal is issued; when the probability of user churn is 0.5, half of the users may churn, and a medium-risk warning signal is issued; when the probability of user churn is 0.5 < P < 1, more than half of the users may churn, and a high-risk warning signal is issued.

[0064] A user churn warning method based on machine learning, which is applied to the user churn warning system based on machine learning described in any one of the above, and the specific steps include:

[0065] S1. Collect the characteristic data of the telecommunications operator and users in the Mth month and the Nth month,

[0066] The characteristic data of the telecommunications operator includes network quality, price factors, competitive environment, and customer relationship;

[0067] The influencing parameters of network quality include signal strength and data transmission speed;

[0068] The influencing parameters of price factors include the package price and traffic fee of one's own side;

[0069] The influencing parameters of the competitive environment include the market share and the package price of competitors;

[0070] The influencing parameters of customer relationship include complaint handling time and the number of complaints per unit time;

[0071] The characteristic data of users includes customer experience, purchase frequency, and activity;

[0072] Among them, (N - M) is greater than or equal to 3;

[0073] S2. Process the signal strength and data transmission speed to generate a network quality coefficient, process the package price and traffic fee of one's own side to generate a price factor coefficient, process the market share and the package price of competitors to generate a competitive environment coefficient, and process the complaint handling time and the number of complaints per unit time to generate a customer relationship coefficient;

[0074] S3. Process and perform correlation analysis on the network quality coefficient, price factor coefficient, competitive environment coefficient, and customer relationship coefficient to generate a competitiveness index of the telecommunications operator;

[0075] S4. Build a machine learning network model, use the characteristic data of users in the Mth month as the training set, and the corresponding user satisfaction coefficient in the Mth month as the label, and input them into the machine learning network model for training;

[0076] S5. Input the characteristic data of the user in the Nth month as a training set into the machine learning network model to obtain the satisfaction coefficient of the user in the Nth month;

[0077] S6. Process the user satisfaction coefficient of the Mth month and the user satisfaction coefficient of the Nth month to generate the rate of change of the user satisfaction coefficient from the Mth month to the Nth month;

[0078] S7. Perform data processing and correlation analysis on the competitiveness index of telecom operators and the change rate of user satisfaction coefficient to generate a correlation coefficient. According to the correlation coefficient between the change rate of the competitiveness index and the user satisfaction coefficient, obtain the probability of user churn. According to the interval range of the probability of user churn, predict the situation of telecom user churn and issue an early warning signal.

[0079] Compared with the prior art, the present invention has the following beneficial effects:

[0080] The present invention collects both characteristic data of telecommunication operators and characteristic data of users, comprehensively analyzes the situation of user loss from the perspectives of telecommunication operators and users, performs data processing on signal strength and data transmission speed of telecommunication operators to generate a network quality coefficient, performs data processing on package prices and traffic charges of the operator to generate a price factor coefficient, performs data processing on market share and package prices of competitors to generate a competition environment coefficient, performs data processing on complaint processing time and the number of complaints per unit time to generate a customer relationship coefficient, performs data processing and correlation analysis on network quality coefficient, price factor coefficient, competition environment coefficient and customer relationship coefficient to generate a competitiveness index of telecommunication operators, utilizes a machine learning network model to obtain user satisfaction coefficient and the rate of change of satisfaction coefficient, performs data processing and correlation analysis on the rate of change of telecommunication operator competitiveness index and user satisfaction coefficient to generate a correlation coefficient, obtains the probability of user loss according to the correlation coefficient between the rate of change of the competitiveness index and the user satisfaction coefficient, predicts the situation of user loss according to the interval range of the probability of user loss, and issues an early warning signal. Therefore, when predicting user churn, collecting characteristic data of telecom operators and users, considering the interaction and relationship between users and operators, and comprehensively analyzing user churn can improve the prediction accuracy of user churn. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 It is a block diagram of the module composition of the present invention;

[0082] Figure 2 It is a schematic diagram of the overall method flow of the present invention. DETAILED DESCRIPTION

[0083] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments.

[0084] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0085] Example:

[0086] See also Figure 1 , the present invention provides a technical solution:

[0087] A user churn early warning system based on machine learning, comprising:

[0088] The data collection module is used to collect the characteristic data of telecom operators and users in the Mth month and the Nth month.

[0089] Among them, (NM≥3), the characteristic data of telecom operators include network quality, price factors, competitive environment and customer relations;

[0090] Parameters affecting network quality include signal strength and data transmission speed;

[0091] The influencing parameters of price factors include the price of one's own package and traffic fee;

[0092] The parameters influencing the competitive environment include market share and competitor package prices;

[0093] The influencing parameters of customer relations include complaint handling time and number of complaints per unit time;

[0094] User characteristic data includes customer experience, purchase frequency and activity;

[0095] The data processing module is used to process the signal strength and data transmission speed to generate the network quality coefficient, process the package price and traffic fee of the party to generate the price factor coefficient, process the market share and the package price of competitors to generate the competition environment coefficient, and process the complaint processing time and the number of complaints per unit time to generate the customer relationship coefficient;

[0096] A data analysis module is used to process and analyze the network quality coefficient, price factor coefficient, competition environment coefficient and customer relationship coefficient to generate a competitiveness index of telecom operators;

[0097] A model building module is used to build a machine learning network model, taking the characteristic data of users in the Mth month as a training set and the corresponding satisfaction coefficient of users in the Mth month as a label, and inputting them into the machine learning network model for training;

[0098] A coefficient acquisition module is used to input the characteristic data of the users in the Nth month as a training set into the machine learning network model to obtain the satisfaction coefficient of the users in the Nth month;

[0099] A change rate acquisition module is used to process the user satisfaction coefficient of the Mth month and the user satisfaction coefficient of the Nth month to generate a change rate of the user satisfaction coefficient from the Mth month to the Nth month;

[0100] The prediction module is used to process data and conduct correlation analysis on the change rate of the competitiveness index of the telecommunications operator and the user's satisfaction coefficient, generate a correlation coefficient, obtain the probability of user churn based on the correlation coefficient between the change rate of the competitiveness index and the user's satisfaction coefficient, predict the situation of telecommunications user churn based on the interval range of the probability of user churn, and issue an early warning signal.

[0101] When the network quality, price factors, competitive environment and customer relationships of telecom operators change, the competitiveness of telecom operators will change, which will in turn lead to changes in user churn rates. The specific reasons are as follows:

[0102] Network quality changes:

[0103] Improve network quality: If a telecom operator improves network stability, enhances call quality, and strengthens customer service support, users will be more inclined to stay with that operator.

[0104] Degraded network quality: If there are problems with network quality, such as frequent network failures and inefficient customer service responses, users may choose to churn due to dissatisfaction.

[0105] Price factor changes:

[0106] Lower prices: If telecom operators lower rates, launch promotional activities, etc., they can attract new customers and retain existing customers, thus reducing user churn.

[0107] Raising prices: If telecom charges increase or the original preferential policies are cancelled, users may consider switching to competitors with more competitive prices, resulting in an increase in user churn.

[0108] Changes in the competitive environment:

[0109] Increased competition: If competitors offer more attractive services, lower-priced packages, or a better customer experience, users may switch to competitors, leading to higher user churn rates.

[0110] Reduced competition: If competitors exit the market or service levels decline, telecom operators may gain more subscribers and their churn rate may decline.

[0111] Changes in customer relationships:

[0112] Positive customer relationships: Building good customer relationships and providing personalized service and support can enhance user loyalty and reduce user churn.

[0113] Negative customer relationships: If users feel ignored or dissatisfied, such as due to poor service attitude or untimely problem resolution, they may choose to churn.

[0114] In summary, the competitiveness of telecom operators is affected by factors such as network quality, price factors, competitive environment and customer relations. Changes in these factors will directly affect user choices and loyalty, and thus affect changes in user churn rate.

[0115] When the user's customer experience, purchase frequency, and activity change, it will lead to changes in user satisfaction, which in turn will lead to changes in user churn rate. The specific reasons are as follows:

[0116] Customer experience changes:

[0117] Positive customer experience: If users experience good customer service, a convenient purchasing process, and high-quality products, their satisfaction will increase and they are more likely to choose to continue using the telecom operator's services.

[0118] Negative customer experience: If users cannot get timely and effective solutions when they encounter problems, or feel that the service quality has declined, their satisfaction may decrease and they may consider changing operators, and the user churn rate will increase.

[0119] Changes in purchase frequency:

[0120] Frequent purchases: If users' purchase frequency increases, it indicates that their demand for telecommunications services has increased, which may mean higher satisfaction with the operator and a possible decrease in user churn.

[0121] Decreased purchase frequency: If users purchase less frequently, it may mean that their demand for telecommunications services has decreased, and they may consider changing operators, and the user churn rate may increase.

[0122] Activity changes:

[0123] Increased activity: If users become more active in telecom services, such as using functions such as calling, surfing the Internet, and sending text messages more frequently, it may mean that they are more satisfied with the service and the user churn rate may decrease.

[0124] Decreased activity: If users’ activity decreases, it may mean that their interest and demand for services have decreased, and they may consider changing operators, and the user churn rate may increase.

[0125] In summary, the user's customer experience, purchase frequency and activity are important factors affecting user satisfaction. When these factors change, they will directly affect the user's attitude and choice of telecommunications services, and thus affect the change of user churn rate.

[0126] Based on the above embodiment, the signal strength and data transmission speed are processed to generate a network quality coefficient according to the following formula:

[0127]

[0128] Among them, XS WZ is the network quality coefficient, S is the signal strength, S′ is the normalized value of the signal strength, V is the data transmission speed, V′ is the normalized value of the data transmission speed, α is the parameter for adjusting the strength of S′ and V′ in the exponential function, e -α(S′+V′) is an exponential function used to balance the impact of normalized signal strength and data transmission speed on the overall network quality, and log(1+SV) is a composite function used to enhance the impact of the product of signal strength and data transmission speed.

[0129] Using the above formula, the network quality coefficient of the Mth month and the Nth month can be obtained, that is, and

[0130] The above formula represents the functional relationship between signal strength, data transmission speed and network quality coefficient for the following reasons:

[0131] First, when the signal strength S or the data transmission speed V increases, their product SV also increases accordingly, which can be reflected in XS through the log(1+SV) function.WZ On, resulting in a higher network quality coefficient;

[0132] Second, normalization of S′ and V′ ensures that the data are compared at the same level, making the model more stable;

[0133] Third, exponential function e -α(S′+V′) The influence of S′ and V′ can be adjusted according to the value of k. When the k value is large, even small changes will cause XS WZ significant changes.

[0134] On the basis of the above embodiment, the package price and traffic fee of the party are processed to generate the price factor coefficient, and the formula is as follows:

[0135]

[0136] Among them, XS JY is the price factor coefficient, P is the package price of the party, P′ is the normalized value of the package price of the party, F is the traffic fee, F′ is the normalized value of the traffic fee, β is the parameter for adjusting the price factor coefficient of the package price and traffic fee of the party, λ is the weight coefficient, which is used to adjust the relative importance of the package price P′ and traffic fee F′ in the calculation. When the λ value is large, it means that the normalized traffic fee F′ is more important to XS JY The impact of is greater than the normalized package price P′ of one's own party; conversely, if the λ value is small, the impact of one's own package price P′ is more significant.

[0137] Using the above formula, we can get the price factor coefficients for the Mth month and the Nth month, that is: and

[0138] Increase of P′:

[0139] If λ is small, when P′ is increased, XS JY The value will increase, which means that higher package prices are more attractive in comparison, because the impact of traffic fee F′ is relatively weakened, making higher-priced packages score higher in the evaluation; if λ is large, when P′ is increased, XS JY The increase will not be that significant because the traffic fee F′ accounts for a larger proportion in the calculation.

[0140] Increase of F′:

[0141] When F' is increased, XS JY The degree of increase depends on the value of λ. If λ is large, it means that the traffic fee has a greater influence on the user's choice of package. Therefore, when F′ increases, XS JYwill also increase significantly, which means that traffic fee becomes the dominant factor in the choice; on the contrary, if λ is small, the increase of F′ will have a significant impact on XS JY The impact of the package is smaller, and the price of one's own package plays a greater role in the choice.

[0142] Therefore, the above formula is used to characterize the functional relationship between one's own package price, traffic fee and price factor coefficient. Since user preference is affected by the weight coefficient λ, the proportion of one's own package price and traffic fee can be adjusted by adjusting the value of λ. That is, when λ is larger, users are more sensitive to traffic fees; when λ is smaller, users are more sensitive to package prices.

[0143] On the basis of the above embodiment, the market share and the package price of competitors are processed to generate the competition environment coefficient, and the formula is as follows:

[0144]

[0145] Among them, XS JH is the competitive environment coefficient, n is the total number of competitors in the market, J i is the package price of the ith competitor, M i is the market share of the ith competitor, δ is the parameter for adjusting the impact of the competitor’s package price on the competitive environment coefficient. Through the value of δ, the weight of the competitor’s package price in the formula can be adjusted to make the negative impact of high-priced competitors on the competitive environment more obvious; ε is the parameter for adjusting the impact of market share on the competitive environment coefficient. By adjusting ε, the contribution of competitors with higher market shares in the coefficient can be enhanced or weakened.

[0146] Using the above formula, we can get the competitive environment coefficients of the Mth month and the Nth month, that is, and

[0147] in, is an exponential function, which is used to deal with the impact of competitors' package prices on the competitive environment coefficient. Through this function, a higher package price will correspond to a smaller function value, which means that the contribution of competitors with higher package prices to the competitive environment is weakened; is a power function, which is used to amplify the impact of market share, allowing competitors with high market share to have a larger weight in the coefficient; is the normalization factor, ensuring the competitive environment coefficient XS JH The value of is within a reasonable range and the results are not directly affected by the number of competitors.

[0148] On the basis of the above embodiment, the complaint handling time and the number of complaints per unit time are processed to generate the customer relationship coefficient, according to the following formula:

[0149]

[0150] T(t)=η·T(t-1)+(1-η)·t cut

[0151] C(t)=log(1+c t )

[0152] Among them, XS KG is the customer relationship coefficient, c t is the number of complaints per unit time, C(t) is the logarithmic transformation of the number of complaints per unit time, t is the complaint processing time, T(t) is the average complaint processing time, is the adjustment coefficient, which is used to balance the impact of complaint processing time and number of complaints. μ is the weight coefficient, which determines the relative weight of C(t). When the μ value is large, it means that C(t) has a negative impact on XS. KG On the contrary, when μ is small, the influence of T(t) is more significant; t cut is the current complaint handling time, η is the smoothing coefficient, and T(t-1) is the smoothed estimate at the previous time point (t-1).

[0153] Using the above formula, we can get the customer relationship coefficients of the Mth month and the Nth month, that is, and

[0154] On the basis of the above embodiment, the network quality coefficient, price factor coefficient, competition environment coefficient and customer relationship coefficient are subjected to data processing and correlation analysis to generate the competitiveness index of the telecommunications operator, according to the following formula:

[0155]

[0156] Among them, ZS is the competitiveness index of telecom operators, R is the number of factors, R = 4;

[0157] Using the above formula, we can obtain the competitiveness index of telecom operators in the Mth month and the Nth month, that is, ZS M and ZS N .

[0158] Network quality factor:

[0159] Network quality is crucial to the competitiveness of telecom operators. If the network quality is high, the user experience will be better, which will increase user satisfaction and loyalty, and thus increase market share. Therefore, an increase in the network quality coefficient will have a positive impact on the competitiveness index.

[0160] Price factor coefficient:

[0161] Price is also an important factor that users consider when choosing an operator. A more reasonable price positioning can attract more users, especially in a price-sensitive market environment. Therefore, an increase in the price factor coefficient will have a positive impact on the competitiveness index.

[0162] Competitive environment coefficient:

[0163] The telecommunications market usually has a fiercely competitive environment, with many competitors and uneven market share distribution. In a fiercely competitive environment, operators need to maintain or expand their market share by continuously improving their competitiveness. Therefore, an increase in the competitive environment coefficient will have a positive impact on the competitiveness index.

[0164] Customer Relationship Coefficient:

[0165] Good customer relationships can improve user loyalty and satisfaction, reduce user churn, and thus consolidate market share. Telecom operators maintain customer relationships by providing personalized services and responding quickly to user issues. Therefore, an increase in the customer relationship coefficient will have a positive impact on the competitiveness index.

[0166] Taking all the above factors into consideration, the competitiveness index is calculated by multiplying and dividing by the square root of the number of comprehensive factors. This can comprehensively reflect the comprehensive competitiveness of telecom operators in terms of network quality, price, competitive environment and customer relations. It can more comprehensively consider the impact of various factors on competitiveness, and reflect the interactive relationship between these factors through multiplication.

[0167] Based on the above embodiment, the machine learning network model is composed of a deep neural network based on a multilayer perceptron, and the deep neural network of the multilayer perceptron includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer and an output layer. The first hidden layer, the second hidden layer and the third hidden layer each have at least two neurons, and each uses ReLU as an activation function.

[0168] In this embodiment, the input features of the deep neural network of the multilayer perceptron include: customer experience, purchase frequency and activity: 3 features;

[0169] The structure of a deep neural network with a multilayer perceptron is:

[0170] Input layer**: receives input of 3 features;

[0171] The first hidden layer has 32 neurons and uses ReLU as the activation function.

[0172] The second hidden layer has 64 neurons and also uses the ReLU activation function.

[0173] The third hidden layer has 32 neurons and uses the ReLU activation function.

[0174] Output layer: has 1 neuron, user satisfaction coefficient.

[0175] Based on the above embodiment, the user satisfaction coefficient of the Mth month and the user satisfaction coefficient of the Nth month are processed to generate the change rate of the user satisfaction coefficient from the Mth month to the Nth month, according to the following formula:

[0176]

[0177] Among them, BL is the change rate of the user's satisfaction coefficient, S N is the user satisfaction coefficient of the Nth month, S M is the user satisfaction coefficient in the Mth month.

[0178] When the user satisfaction coefficient S in the Nth month N Compared to S in the Mth month M When is large, the rate of change is positive, indicating an increase in user satisfaction;

[0179] When the user satisfaction coefficient S in the Nth month N Compared to S in the Mth month M hours, the change rate is negative, indicating that user satisfaction decreases;

[0180] When the user satisfaction coefficient S in the Nth month N and S in the Mth month M When they are equal, the rate of change is 0, indicating that user satisfaction has not changed.

[0181] On the basis of the above embodiment, the change rates of the competitiveness index of the telecommunications operator and the user satisfaction coefficient are subjected to data processing and correlation analysis to generate a correlation coefficient, and the formula based on this is as follows:

[0182]

[0183] Among them, r is the correlation coefficient between the competitiveness index and the change rate of the user satisfaction coefficient, ZS N is the competitiveness index of the telecom operator in the Nth month, BL is the change rate of the user satisfaction coefficient from the Mth month to the Nth month, is the average of the competitiveness index from the M-th month to the N-th month, is the average of the change rate of the satisfaction coefficient from the M-th month to the N-th month.

[0184] When r < 0, it indicates that the change trends of the competitiveness index and the change rate of the user satisfaction coefficient are opposite. An increase in competitiveness will lead to a decrease in user satisfaction, thereby increasing the user churn rate;

[0185] When r = 0, it indicates that there is no correlation between the competitiveness index and the change rate of the user satisfaction coefficient;

[0186] When r > 0, it indicates that the change trends of the competitiveness index and the change rate of the user satisfaction coefficient are the same. An increase in competitiveness will bring about an increase in user satisfaction, thus reducing the user churn rate.

[0187] According to the correlation coefficient between the competitiveness index and the change rate of the user satisfaction coefficient, obtain the probability of user churn. The basis formula is as follows:

[0188]

[0189] where P is the probability of user churn, a < 0 and |a| is large enough, b > 0 and |b| is large enough.

[0190] The process of predicting the situation of telecom user churn and sending out warning signals according to the range of the probability of user churn is as follows:

[0191] When r < 0, 0.5 < P < 1;

[0192] When r = 0, P = 0.5;

[0193] When r > 0, 0 < P < 0.5.

[0194] When the range of P is between 0 and 1, 0 means there is no possibility of user churn, and 1 means all users will churn;

[0195] To sum up, when the probability of user churn is 0 < P < 0.5, send out a low-risk warning signal. Sending out a low-risk warning signal means that the probability of user churn is low, and less than half of the users may churn; when the probability of user churn is 0.5, send out a medium-risk warning signal. Sending out a medium-risk warning signal means that the probability of user churn is medium, and half of the users may churn; when the probability of user churn is 0.5 < P < 1, send out a high-risk warning signal. Sending out a high-risk warning signal means that the probability of user churn is high, and more than half of the users may churn.

[0196] α, β, λ, δ, ε in the formula, The specific values ​​of and μ are generally determined by those skilled in the art according to actual conditions. The essence of this formula is a comprehensive analysis by weighted summation. The technicians in this field collect multiple groups of sample data, and set corresponding preset proportional coefficients for each group of sample data. The preset proportional coefficients and the collected sample data are substituted into the formula. Through repeated experiments and parameter adjustments, the accuracy of the model output and the rationality of the results are observed, and these factor coefficients are gradually adjusted. The performance and effect of the model under different parameter settings are compared to find the optimal coefficient combination. The calculated factor coefficients are screened and averaged to obtain α, β, λ, δ, ε, and the value of μ.

[0197] In addition, the size of the preset factor coefficient is to quantify each parameter to obtain a specific value. In order to facilitate subsequent comparison, the size of the coefficient depends on the amount of sample data and the preset proportional coefficient initially set by technical personnel in this field for each set of sample data. It is not unique as long as it does not affect the proportional relationship between the parameter and the quantized value.

[0198] See also Figure 2 , the present invention also provides a technical solution:

[0199] A user churn warning method based on machine learning, the method is applied to any of the above-mentioned user churn warning systems based on machine learning, and the specific steps include:

[0200] S1. Collect characteristic data of telecom operators and users in the Mth and Nth months,

[0201] Among them, (NM≥3), the characteristic data of telecom operators include network quality, price factors, competitive environment and customer relations;

[0202] Parameters affecting network quality include signal strength and data transmission speed;

[0203] The influencing parameters of price factors include the price of one's own package and traffic fee;

[0204] The parameters influencing the competitive environment include market share and competitor package prices;

[0205] The influencing parameters of customer relations include complaint handling time and number of complaints per unit time;

[0206] User characteristic data includes customer experience, purchase frequency and activity;

[0207] S2. Process the signal strength and data transmission speed to generate the network quality coefficient, process the package price and traffic fee of the party to generate the price factor coefficient, process the market share and the package price of competitors to generate the competition environment coefficient, process the complaint handling time and the number of complaints per unit time to generate the customer relationship coefficient;

[0208] S3. Perform data processing and correlation analysis on the network quality coefficient, price factor coefficient, competition environment coefficient and customer relationship coefficient to generate a competitiveness index of telecom operators;

[0209] S4. Build a machine learning network model, use the characteristic data of users in the Mth month as a training set, and the corresponding satisfaction coefficient of users in the Mth month as a label, and input it into the machine learning network model for training;

[0210] S5. Input the characteristic data of the user in the Nth month as a training set into the machine learning network model to obtain the satisfaction coefficient of the user in the Nth month;

[0211] S6. Process the user satisfaction coefficient of the Mth month and the user satisfaction coefficient of the Nth month to generate the rate of change of the user satisfaction coefficient from the Mth month to the Nth month;

[0212] S7. Perform data processing and correlation analysis on the competitiveness index of telecom operators and the change rate of user satisfaction coefficient to generate a correlation coefficient. According to the correlation coefficient between the change rate of the competitiveness index and the user satisfaction coefficient, obtain the probability of user churn. According to the interval range of the probability of user churn, predict the situation of telecom user churn and issue an early warning signal.

[0213] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0214] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. Those skilled in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed by hardware or software methods depends on the specific application and design constraints of the technical solution.

[0215] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0216] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A user churn warning system based on machine learning, characterized in that: include: The data collection module is used to collect the characteristic data of telecom operators and users in the Mth month and the Nth month. Telecom operators’ characteristic data include network quality, price factors, competitive environment and customer relations; Parameters affecting network quality include signal strength and data transmission speed; The influencing parameters of price factors include the price of one's own package and traffic fee; The parameters influencing the competitive environment include market share and competitor package prices; The influencing parameters of customer relations include complaint handling time and number of complaints per unit time; User characteristic data includes customer experience, purchase frequency and activity; Wherein, (NM) is greater than or equal to 3; The data processing module is used to process the signal strength and data transmission speed to generate the network quality coefficient, process the package price and traffic fee of the party to generate the price factor coefficient, process the market share and the package price of competitors to generate the competition environment coefficient, and process the complaint processing time and the number of complaints per unit time to generate the customer relationship coefficient; A data analysis module is used to process and analyze the network quality coefficient, price factor coefficient, competition environment coefficient and customer relationship coefficient to generate a competitiveness index of telecom operators; A model building module is used to build a machine learning network model, taking the characteristic data of users in the Mth month as a training set and the corresponding satisfaction coefficient of users in the Mth month as a label, and inputting them into the machine learning network model for training; A coefficient acquisition module is used to input the characteristic data of the users in the Nth month as a training set into the machine learning network model to obtain the satisfaction coefficient of the users in the Nth month; A change rate acquisition module is used to process the user satisfaction coefficient of the Mth month and the user satisfaction coefficient of the Nth month to generate a change rate of the user satisfaction coefficient from the Mth month to the Nth month; The prediction module is used to process and analyze the data of the competitiveness index of the telecommunication operator and the change rate of the user's satisfaction coefficient, generate a correlation coefficient, obtain the probability of user loss according to the correlation coefficient between the competitiveness index and the change rate of the user's satisfaction coefficient, predict the loss of telecommunication users according to the interval range of the probability of user loss, and issue an early warning signal; The network quality coefficient, price factor coefficient, competition environment coefficient and customer relationship coefficient are processed and correlated to generate the competitiveness index of telecom operators. The formula is as follows: Among them, ZS is the competitiveness index of telecom operators, R is the number of factors, R = 4, XS WZ is the network quality coefficient, XS JY is the price factor coefficient, XS JH is the competitive environment coefficient, XS KG is the customer relationship coefficient; Using the above formula, we can obtain the competitiveness index of telecom operators in the Mth month and the Nth month, i.e., ZS M and ZS N ; The machine learning network model is composed of a deep neural network based on a multilayer perceptron, and the deep neural network of the multilayer perceptron includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer and an output layer, and the first hidden layer, the second hidden layer and the third hidden layer each have at least two neurons, and each uses ReLU as an activation function; The change rates of the competitiveness index of telecom operators and the user satisfaction coefficient are processed and correlated to generate the correlation coefficient, based on the following formula: Among them, r is the correlation coefficient between the competitiveness index and the change rate of the user satisfaction coefficient, ZS N is the competitiveness index of the telecom operator in the Nth month, BL is the change rate of the user satisfaction coefficient from the Mth month to the Nth month, is the average competitiveness index from month M to month N, is the mean of the change rate of the satisfaction coefficient from the Mth month to the Nth month; When r < 0, it indicates that the change trends of the competitiveness index and the change rate of the user satisfaction coefficient are opposite. The improvement of competitiveness will lead to the decline of user satisfaction, and further increase the user churn rate. When r = 0, it indicates that there is no correlation between the competitiveness index and the change rate of the user satisfaction coefficient. When r > 0, it indicates that the change trends of the competitiveness index and the change rate of the user satisfaction coefficient are consistent. The improvement of competitiveness will bring about the improvement of user satisfaction, thus reducing the user churn rate. According to the correlation coefficient between the competitiveness index and the change rate of the user satisfaction coefficient, obtain the probability of user churn. The formula is as follows: Where P is the probability of user churn, a < 0, b > 0. The process of predicting the situation of telecom user churn and sending out warning signals according to the range of the probability of user churn is as follows: When r<0, When r = 0, P = 0.

5. When r>0, When the range of P is between 0 and 1, 0 indicates no possibility of user churn, and 1 indicates that all users will churn. To sum up, when the probability of user churn is 0 < P < 0.5, less than half of the users may churn, and a low-risk warning signal is sent; when the probability of user churn is 0.5, half of the users may churn, and a medium-risk warning signal is sent; when the probability of user churn is 0.5 < P < 1, more than half of the users may churn, and a high-risk warning signal is sent.

2. The user churn early warning system based on machine learning according to claim 1 is characterized by: Perform data processing on the signal strength and data transmission speed to generate a network quality coefficient. The formula is as follows: Where S is the signal strength, S′ is the normalized value of the signal strength, V is the data transmission speed, V′ is the normalized value of the data transmission speed, α is the parameter for adjusting the strength of the action of S′ and V′ in the exponential function, and e -α(S′+V′) is an exponential function used to balance the impact of normalized signal strength and data transmission speed on the overall network quality, and log(1+SV) is a composite function used to enhance the impact of the product of signal strength and data transmission speed; Using the above formula, we can get the network quality coefficients of the Mth month and the Nth month, that is: and 3. The user churn early warning system based on machine learning according to claim 1 is characterized by: Perform data processing on the package price and traffic fee of one's own side to generate a price factor coefficient. The formula is as follows: Where P is the package price of one's own side, P′ is the normalized value of the package price of one's own side, F is the traffic fee, F′ is the normalized value of the traffic fee, β is a parameter for adjusting the package price and traffic fee of one's own side on the price factor coefficient, and λ is a weight coefficient used to adjust the relative importance of the package price P′ and traffic fee F′ of one's own side in the calculation. Using the above formula, we can get the price factor coefficients for the Mth month and the Nth month, that is: and 4. The user churn warning system based on machine learning according to claim 1 is characterized by: Perform data processing on the market share and the package price of competitors to generate a competitive environment coefficient. The formula is as follows: Where n is the total number of competitors in the market, J i is the package price of the ith competitor, M i is the market share of the ith competitor, δ is the parameter that adjusts the effect of the competitor’s package price on the competitive environment coefficient, and ε is the parameter that adjusts the effect of the market share on the competitive environment coefficient; Using the above formula, we can obtain the competitive environment coefficients for the Mth month and the Nth month, namely: and 5. The user churn warning system based on machine learning according to claim 1 is characterized by: Perform data processing on the complaint handling time and the number of complaints per unit time to generate a customer relationship coefficient. The formula is as follows: T(t)=η·T(t-1)+(1-η)·t cut C(t)=log(1+c t ) Among them, c t is the number of complaints per unit time, C(t) is the logarithmic transformation of the number of complaints per unit time, t is the complaint processing time, T(t) is the average complaint processing time, is the adjustment coefficient, which is used to balance the impact of complaint processing time and number of complaints; μ is the weight coefficient, which determines the relative weight of C(t); cut is the current complaint handling time, η is the smoothing coefficient, and T(t-1) is the smoothed estimated value at the previous time point (t-1); Using the above formula, we can get the customer relationship coefficients of the Mth month and the Nth month, that is, and 6. The user churn warning system based on machine learning according to claim 1, characterized in that: Perform data processing on the user satisfaction coefficient in the Mth month and the user satisfaction coefficient in the Nth month to generate the change rate of the user satisfaction coefficient from the Mth month to the Nth month. The formula is as follows: Among them, S N is the user satisfaction coefficient of the Nth month, S M is the user satisfaction coefficient in the Mth month.

7. A user churn warning method based on machine learning, the method being applied to the user churn warning system based on machine learning according to any one of claims 1 to 6, characterized in that: The specific steps include: S1. Collect the characteristic data of telecom operators and users in the Mth month and the Nth month. The characteristic data of telecom operators include network quality, price factor, competitive environment and customer relationship. The influencing parameters of network quality include signal strength and data transmission speed. The influencing parameters of price factor include the package price and traffic fee of one's own side. The influencing parameters of competitive environment include market share and the package price of competitors. The influencing parameters of customer relationship include complaint handling time and the number of complaints per unit time. The characteristic data of users include customer experience, purchase frequency and activity. Where (N - M) is greater than or equal to 3. S2. Process the signal strength and data transmission speed to generate the network quality coefficient, process the package price and traffic fee of the party to generate the price factor coefficient, process the market share and the package price of competitors to generate the competition environment coefficient, process the complaint handling time and the number of complaints per unit time to generate the customer relationship coefficient; S3. Perform data processing and correlation analysis on the network quality coefficient, price factor coefficient, competition environment coefficient and customer relationship coefficient to generate a competitiveness index of telecom operators; S4. Build a machine learning network model, use the characteristic data of users in the Mth month as a training set, and the corresponding satisfaction coefficient of users in the Mth month as a label, and input it into the machine learning network model for training; S5. Input the characteristic data of the user in the Nth month as a training set into the machine learning network model to obtain the satisfaction coefficient of the user in the Nth month; S6. Process the user satisfaction coefficient of the Mth month and the user satisfaction coefficient of the Nth month to generate the rate of change of the user satisfaction coefficient from the Mth month to the Nth month; S7. Perform data processing and correlation analysis on the competitiveness index of telecom operators and the change rate of user satisfaction coefficient to generate a correlation coefficient. According to the correlation coefficient between the change rate of the competitiveness index and the user satisfaction coefficient, obtain the probability of user churn. According to the interval range of the probability of user churn, predict the situation of telecom user churn and issue an early warning signal.