Satisfaction monitoring, alerting and attribution method and system based on long and short term historical data

Abstract

The application discloses a satisfaction monitoring, alarm and attribution method and system based on long-term and short-term historical data, and belongs to the technical field of cloud service customer satisfaction monitoring. The implementation of the method comprises the following steps: month granularity customer overall satisfaction monitoring, performing monthly overall satisfaction calculation; month granularity customer overall satisfaction analysis based on long-term and short-term satisfaction data, comprising 'fluctuation' analysis based on short-term satisfaction data and 'trend' analysis based on long-term satisfaction data; month granularity customer overall satisfaction alarm, giving customer loss alarm or satisfaction decline alarm according to the input of month granularity customer overall satisfaction analysis; month granularity customer overall satisfaction attribution, giving attribution of this month's satisfaction decline according to the signal of month granularity customer overall satisfaction alarm. The application can realize automatic monitoring, alarm and attribution of customer satisfaction, and is helpful to improving the efficiency and quality of analysis and then helping enterprises to extract beneficial information from customer feedback.

Description

Technical Field

[0001] This invention relates to the field of cloud service customer satisfaction monitoring technology, specifically to a method and system for satisfaction monitoring, alerting, and attribution based on long-term and short-term historical data. Background Technology

[0002] The rapid development of cloud-native technologies, represented by containers, microservices, DevOps, service mesh, declarative APIs, and emerging low-code development, has overcome the problems of slow deployment and low resource utilization, high costs, poor flexibility, and inability to scale elastically on servers within self-owned data centers that traditionally develop applications in local environments. This further unleashes the powerful service potential of the cloud. The increasing adoption of cloud services by enterprises and individual users has led to a continuous growth in the number of existing customers. How to provide high-quality cloud services in the era of digital intelligence, improve the refinement and intelligence of services, and win customer trust and satisfaction is a crucial strategic issue for all cloud service providers.

[0003] Customer satisfaction feedback is the most critical element for cloud service providers to iterate their products and services and maintain a competitive edge. Currently, most cloud service providers analyze customer satisfaction data by using the arithmetic mean of customer satisfaction scores. While this calculation method is simple and intuitive, it has the following drawbacks:

[0004] 1. Susceptible to outliers: For the satisfaction scores x1,…,xn of n users obtained in a certain period, the arithmetic mean (x1+…+xn) / n is susceptible to outliers. That is, a certain extreme customer score may affect the cloud service provider's assessment of the overall satisfaction, leading to one-sided or wrong conclusions, which will have a negative impact on the company's subsequent decision-making.

[0005] 2. Lack of objective methods to judge the degree of change in satisfaction: The arithmetic mean of satisfaction is an observation value given by a simple statistic. Cloud service providers can only obtain the rise or fall of current satisfaction compared to the base period through this observation value, but cannot answer whether the change is a data fluctuation caused by random sampling or truly represents a significant trend of change in satisfaction.

[0006] 3. Lack of analysis and attribution of changes in satisfaction: The arithmetic mean only serves to monitor user satisfaction without providing further alerts or attributions. This results in limited information mining and information waste, making it difficult to provide useful input for optimizing customer service for cloud service providers. Summary of the Invention

[0007] The technical objective of this invention is to address the above-mentioned shortcomings by providing a method and system for monitoring, alerting, and attributing customer satisfaction based on long-term and short-term historical data. This system is fully functional and based on scientific statistical analysis methods, enabling automated monitoring, alerting, and attribution of customer satisfaction. It helps improve the efficiency and quality of analysis, thereby helping enterprises extract useful information from customer feedback.

[0008] The technical solution adopted by this invention to solve its technical problem is:

[0009] A method for monitoring, alerting, and attributing satisfaction based on long-term and short-term historical data. The implementation of this method includes:

[0010] Monthly granular customer satisfaction monitoring and calculation;

[0011] Monthly granular customer satisfaction analysis based on short-term and long-term satisfaction data includes "fluctuation" analysis based on short-term satisfaction data and "trend" analysis based on long-term satisfaction data; the "fluctuation" analysis based on short-term satisfaction data uses a double independent t-test to compare the significance of the difference between this month's and last month's customer satisfaction; the "trend" analysis based on long-term satisfaction data uses time series analysis to predict the customer satisfaction value for the next two months to determine whether the decline this month represents a long-term downward trend;

[0012] Monthly granular customer satisfaction alerts will be issued based on the input of the monthly granular customer satisfaction analysis, corresponding to either customer churn alerts or satisfaction decline alerts.

[0013] Monthly granular customer satisfaction attribution: Based on the monthly granular customer satisfaction alert signals, this section provides the attribution for the decline in satisfaction this month.

[0014] Furthermore, the method for calculating overall customer satisfaction at the monthly granularity level is as follows:

[0015] Missing data is handled by replacing missing values ​​with the mean of that dimension.

[0016] Invalid data is handled by replacing invalid values ​​that are not between 0 and 100 with the mean of that dimension.

[0017] Duplicate data processing: Deleting duplicate records;

[0018] Outlier handling, in the case of large samples, is based on the Classical Central Limit Theorem (CCLT) and the Three-sigma Rule of Thumb to give the outlier range. In the case of small samples, the Trimmed Rule is used to remove the data closest to the two ends according to a set ratio.

[0019] The process of randomly extracting a fixed number of samples with replacement from the mean based on the bootstrap method and calculating their arithmetic mean, and then calculating the arithmetic mean.

[0020] Furthermore, the steps of the "fluctuation" analysis for calculating the significance of the monthly satisfaction difference are as follows:

[0021] First, define the null hypothesis that there is no significant difference in the overall customer satisfaction this month compared to last month. For the selected significance level alpha, construct the test statistic for the two-independent-sample t-test as:

[0022]

[0023] where,

[0024]

[0025]

[0026]

[0027] Based on this, calculate the p-value of the test; when p < alpha, the null hypothesis of the test is rejected, and the system will output a signal of significant difference, otherwise output a signal of non-significant difference.

[0028] Furthermore, the steps of the "trend" analysis for predicting the potential customer satisfaction are as follows:

[0029] Retrieve the detailed data of the customer-granularity satisfaction surveys for all historical months, and use the unit root test method (Dickey-Fuller Test) to test the stationarity of the sequence, proving that the time series data is non-stationary and the sequence has a unit root;

[0030] Conduct a non-white noise test on the sequence data to prove that the data is non-white noise and has temporal correlation;

[0031] Calculate the monthly overall satisfaction, and then train a time series model based on the overall satisfaction data for all historical monthly granularities and predict the overall satisfaction of customers in the next two months.

[0032] Furthermore, the running steps of the satisfaction abnormal decline warning are as follows:

[0033] If the signal output by the monthly-granularity customer overall satisfaction monitoring is not a decline or the "trend" analysis based on the long-term satisfaction data does not show a continuous decline prediction, the warning is not triggered;

[0034] If the signal output by the monthly-granularity customer overall satisfaction monitoring is a decline and the "fluctuation" analysis based on the short-term satisfaction data determines a significant difference, the warning triggers a "customer churn warning";

[0035] If the monthly granular customer satisfaction monitoring outputs a declining signal and the "trend" analysis based on long-term satisfaction data gives a prediction of continued decline, then an alarm will be triggered to "satisfaction decline alarm".

[0036] Furthermore, the operational steps for attributing abnormally low satisfaction levels are as follows:

[0037] If a "customer churn alert" is received, the reasons for the low score will be identified from two aspects: dissatisfied customers and dissatisfaction dimensions.

[0038] If a "satisfaction decline warning" is received, the phenomenon of declining satisfaction will be analyzed from the perspective of two long-term evolutionary processes: changes in customer structure and changes in work order structure.

[0039] This invention also claims protection for a satisfaction monitoring, alerting, and attribution system based on long-term and short-term historical data, comprising a monthly granular customer overall satisfaction monitoring module, a monthly granular customer overall satisfaction analysis module based on long-term and short-term satisfaction data, a monthly granular customer overall satisfaction alerting module, and a monthly granular customer overall satisfaction attribution module; wherein,

[0040] The monthly granular customer satisfaction monitoring module performs monthly overall satisfaction calculations.

[0041] The monthly granular customer satisfaction analysis module, based on both short-term and long-term satisfaction data, includes a "fluctuation" analysis module based on short-term satisfaction data and a "trend" analysis module based on long-term satisfaction data. The "fluctuation" analysis module based on short-term satisfaction data uses a double independent t-test to compare the significance of the difference between customer satisfaction this month and last month. The "trend" analysis module based on long-term satisfaction data uses time series analysis to predict the value of customer satisfaction in the next two months to determine whether the decline this month represents a downward trend in the long term.

[0042] The monthly granular customer satisfaction alert module provides corresponding customer churn alerts or satisfaction decline alerts based on the input from the analysis module.

[0043] The monthly granular customer satisfaction attribution module provides the attribution for the decline in satisfaction this month based on signals from the alarm module.

[0044] Furthermore, the implementation process of this system is as follows:

[0045] 1) After the system obtains the detailed customer satisfaction survey data for this month, it will first calculate the overall satisfaction for this month through the monthly granular customer overall satisfaction monitoring module and compare it with the overall satisfaction for last month. If the satisfaction increases compared to last month, the process ends; otherwise, the system will transfer the data to the monthly granular customer overall satisfaction analysis module.

[0046] 2) The "fluctuation" analysis module based on short-term satisfaction data performs hypothesis testing on the significance of the difference between the overall satisfaction of this month and last month. If the overall satisfaction of this month is significantly lower than that of last month, the system will issue a "customer churn" alarm; otherwise, the data will be transferred to the "trend" analysis module based on long-term satisfaction data.

[0047] 3) The "Trend" analysis module based on long-term satisfaction data uses detailed customer satisfaction survey data for all historical months to perform time series analysis and predict the changes in overall satisfaction in the next two months. If the model shows that customer satisfaction will continue to decline in the next two months, the system will issue a "continuous decline" alarm if the satisfaction in this month does not decline significantly; otherwise, the process ends.

[0048] 4) After the monthly granular customer satisfaction alarm module is triggered, the system will perform adaptive attribution based on the alarm category transmitted in.

[0049] The present invention also claims a device for monitoring, alarming and attributing satisfaction based on long-term and short-term historical data, comprising: at least one memory and at least one processor;

[0050] The at least one memory is used to store a machine-readable program;

[0051] The at least one processor is used to call the machine-readable program to implement the above method.

[0052] The present invention also claims protection for a computer-readable medium storing computer instructions that, when executed by a processor, cause the processor to perform the method described above.

[0053] Compared with existing technologies, the satisfaction monitoring, alarm, and attribution method and system based on long-term and short-term historical data of the present invention have the following advantages:

[0054] This invention automates the monitoring, alerting, and attribution of customer satisfaction. Compared to existing satisfaction analysis methods based on arithmetic mean, it utilizes more scientific statistical analysis methods (such as the law of large numbers, hypothesis testing, and time series analysis) and possesses more comprehensive functions (in addition to calculation and monitoring, this system can also provide alerts and attributions based on long-term and short-term data analysis). It effectively avoids the problems of existing methods, such as susceptibility to outliers, lack of objective methods for judging the degree of change in satisfaction, and lack of analysis and attribution of satisfaction changes. This helps enterprises extract useful information from customer feedback. Specifically, compared to satisfaction analysis methods based on arithmetic mean, this invention achieves the following beneficial effects:

[0055] More comprehensive: The method or system for monitoring, alerting and attributing customer satisfaction based on long-term and short-term historical data includes four modules, which can comprehensively and fully realize the functions of monitoring, analyzing, alerting and attributing customer satisfaction.

[0056] More automated: Through reasonable logical connections, the system can achieve automation. That is, after receiving the detailed customer satisfaction survey data for the month, the system will automatically execute all steps. The data-driven alarm module will output customer churn alarms, satisfaction decline alarms, or no alarms. The attribution module will provide multi-angle analysis of the causes corresponding to the alarm information.

[0057] More scientific: The design of each module is based on scientific statistical analysis methods. The outlier handling technology of the monitoring module is based on the classical central limit theorem and the 3 sigma principle of normal distribution. The "fluctuation" analysis module based on short-term satisfaction data uses the double independent t test to test the significance of data differences. The "trend" analysis module based on long-term satisfaction data first performs stationarity and non-white noise tests on the data to prove that the data is stationary and has temporal correlation. Then, it scientifically uses a time series model to fit the data and achieve prediction. Attached Figure Description

[0058] Figure 1 This is a system architecture diagram of satisfaction monitoring, alarming, and attribution based on long-term and short-term historical data provided in this embodiment of the invention;

[0059] Figure 2 This is a flowchart illustrating the monthly granular customer overall satisfaction monitoring process provided in an embodiment of the present invention;

[0060] Figure 3 This is a flowchart illustrating the "fluctuation" analysis process based on short-term satisfaction data provided in this embodiment of the invention.

[0061] Figure 4 This is a flowchart illustrating the "trend" analysis process based on long-term satisfaction data provided in an embodiment of the present invention.

[0062] Figure 5 This is a flowchart illustrating the monthly granular customer overall satisfaction alarm process provided in an embodiment of the present invention. Detailed Implementation

[0063] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0064] This invention provides a method for monitoring, alerting, and attributing satisfaction based on long-term and short-term historical data. The implementation of this method includes:

[0065] Monthly granular customer satisfaction monitoring and calculation;

[0066] Monthly granular customer satisfaction analysis based on short-term and long-term satisfaction data includes "fluctuation" analysis based on short-term satisfaction data and "trend" analysis based on long-term satisfaction data; the "fluctuation" analysis based on short-term satisfaction data uses a double independent t-test to compare the significance of the difference between this month's and last month's customer satisfaction; the "trend" analysis based on long-term satisfaction data uses time series analysis to predict the customer satisfaction value for the next two months to determine whether the decline this month represents a long-term downward trend;

[0067] Monthly granular customer satisfaction alerts will be issued based on the input of the monthly granular customer satisfaction analysis, corresponding to either customer churn alerts or satisfaction decline alerts.

[0068] Monthly granular customer satisfaction attribution: Based on the monthly granular customer satisfaction alert signals, this section provides the attribution for the decline in satisfaction this month.

[0069] The implementation process of this method is as follows:

[0070] After the system obtains the detailed customer satisfaction survey data for this month, it will first calculate the overall satisfaction for this month through the monthly granular customer overall satisfaction monitoring and compare it with the overall satisfaction for the previous month. If the satisfaction has increased compared to the previous month, the process ends; otherwise, the system will transfer the data to the analysis module.

[0071] The "fluctuation" analysis module based on short-term satisfaction data performs a hypothesis test on the significance of the difference between the overall satisfaction of this month and last month. If the overall satisfaction of this month is significantly lower than that of last month, the system will issue a "customer churn" alarm; otherwise, the data will be transferred to the "trend" analysis module based on long-term satisfaction data.

[0072] The "Trend" analysis module based on long-term satisfaction data uses detailed customer satisfaction survey data from all historical months to perform time series analysis and predict changes in overall satisfaction over the next two months. If the model shows that customer satisfaction is on a continuous downward trend over the next two months, the system will issue a "continuous decline" warning if satisfaction does not decrease significantly this month; otherwise, the process ends.

[0073] After the alarm module is triggered, the system will adaptively attribute the cause based on the type of alarm transmitted.

[0074] refer to Figure 2 As shown, the method for calculating overall customer satisfaction at the monthly granularity level is as follows:

[0075] Missing data is handled by replacing missing values ​​with the mean of that dimension.

[0076] Invalid data is handled by replacing invalid values ​​that are not between 0 and 100 with the mean of that dimension.

[0077] Duplicate data processing: Delete duplicate records;

[0078] Outlier processing: In the case of a large sample [sample size ≥ 30], the outlier interval is given based on the classic central limit theorem (Classic Central Limit Theorem, CCLT) and the 3-sigma law (Three-sigma Rule of Thumb). In the case of a small sample [sample size < 30], the trimmed rule is used to remove the data closest to both ends according to a set ratio.

[0079] Mean based on the bootstrap method: Repeatedly and randomly draw a fixed number of samples to calculate the arithmetic mean and calculate its arithmetic mean.

[0080] Reference Figure 3 As shown, the steps of the "fluctuation" analysis for calculating the significance of the monthly satisfaction difference are as follows:

[0081] First, define the null hypothesis that there is no significant difference in the overall satisfaction of customers this month compared to last month. For the selected significance level alpha, the statistic for constructing a two-independent-sample t-test is:

[0082]

[0083] where,

[0084]

[0085]

[0086]

[0087] Based on this, calculate the p-value of the test. When p < alpha, the null hypothesis of the test is rejected, and the system will output a signal of significant difference; otherwise, it will output a signal of non-significant difference.

[0088] Reference Figure 4 As shown, the steps of the "trend" analysis for predicting the potential satisfaction of customers are as follows:

[0089] Retrieve the detailed satisfaction survey data of all historical months at the customer level, and use the unit root test method (Dickey-Fuller Test) to test the stationarity of the sequence, proving that the time series data is non-stationary and the sequence has a unit root;

[0090] Perform a non-white noise test on the sequence data to prove that the data is non-white noise and has temporal correlation;

[0091] Calculate the overall monthly satisfaction level, then train a time series model based on the overall satisfaction data of all months throughout history, and predict the overall customer satisfaction level for the next two months.

[0092] refer to Figure 5 As shown, the operation steps for the alarm of abnormally low satisfaction are as follows:

[0093] If the signal output by the monthly granular customer satisfaction monitoring is no decline or the "trend" analysis based on long-term satisfaction data does not show a prediction of a continuous decline, the alarm will not be triggered.

[0094] If the monthly granular customer satisfaction monitoring output signal is declining and the difference is significant based on the "fluctuation" analysis of short-term satisfaction data, then an alarm will be triggered to "customer churn alarm".

[0095] If the monthly granular customer satisfaction monitoring outputs a declining signal and the "trend" analysis based on long-term satisfaction data gives a prediction of continued decline, then an alarm will be triggered to "satisfaction decline alarm".

[0096] The steps for attributing abnormally low satisfaction levels are as follows:

[0097] If a "customer churn alert" is received, the reasons for the low score will be identified from two aspects: dissatisfied customers and dissatisfaction dimensions.

[0098] If a "satisfaction decline warning" is received, the phenomenon of declining satisfaction will be analyzed from the perspective of two long-term evolutionary processes: changes in customer structure and changes in work order structure.

[0099] This invention also provides a satisfaction monitoring, alerting, and attribution system based on long-term and short-term historical data, such as... Figure 1 As shown, it includes a monthly granular customer satisfaction monitoring module, a monthly granular customer satisfaction analysis module based on long-term and short-term satisfaction data, a monthly granular customer satisfaction alert module, and a monthly granular customer satisfaction attribution module; among which,

[0100] The monthly granular customer satisfaction monitoring module (hereinafter referred to as the monitoring module) performs monthly overall satisfaction calculation;

[0101] The monthly granular customer satisfaction analysis module (hereinafter referred to as the analysis module) based on short-term and long-term satisfaction data includes a "fluctuation" analysis module based on short-term satisfaction data and a "trend" analysis module based on long-term satisfaction data. The "fluctuation" analysis module based on short-term satisfaction data uses a double independent t-test to compare the significance of the difference between customer satisfaction this month and last month. The "trend" analysis module based on long-term satisfaction data uses time series analysis to predict the value of customer satisfaction in the next two months to determine whether the decline this month represents a downward trend in the long term.

[0102] The monthly granular customer overall satisfaction alarm module (hereinafter referred to as the alarm module) provides customer churn alarms or satisfaction decline alarms based on the input of the analysis module.

[0103] The monthly granular customer satisfaction attribution module (hereinafter referred to as the attribution module) provides the attribution for the decline in satisfaction this month based on the signals from the alarm module.

[0104] The implementation process of this system is as follows:

[0105] 1) After the system obtains the detailed customer satisfaction survey data for this month, it will first calculate the overall satisfaction for this month through the monthly granular customer overall satisfaction monitoring module and compare it with the overall satisfaction for last month. If the satisfaction increases compared to last month, the process ends; otherwise, the system will transfer the data to the monthly granular customer overall satisfaction analysis module.

[0106] 2) The "fluctuation" analysis module based on short-term satisfaction data performs hypothesis testing on the significance of the difference between the overall satisfaction of this month and last month. If the overall satisfaction of this month is significantly lower than that of last month, the system will issue a "customer churn" alarm; otherwise, the data will be transferred to the "trend" analysis module based on long-term satisfaction data.

[0107] 3) The "Trend" analysis module based on long-term satisfaction data uses detailed customer satisfaction survey data for all historical months to perform time series analysis and predict the changes in overall satisfaction in the next two months. If the model shows that customer satisfaction will continue to decline in the next two months, the system will issue a "continuous decline" alarm if the satisfaction in this month does not decline significantly; otherwise, the process ends.

[0108] 4) After the monthly granular customer satisfaction alarm module is triggered, the system will perform adaptive attribution based on the alarm category transmitted in.

[0109] The following combination Figure 2 The monthly customer satisfaction monitoring process is as follows: The detailed customer satisfaction survey data for the current month is processed sequentially, including missing data processing, invalid data processing, duplicate data processing, and outlier handling. Based on this, the bootstrap mean is output as the overall monthly satisfaction score.

[0110] Missing data is filled using the mean of the corresponding dimension.

[0111] Invalid data is replaced by the mean of the corresponding dimension, which is not within the 0-100 range.

[0112] Deduplicated data processing: Remove duplicate data records.

[0113] Outlier handling gives the outlier interval based on the classical central limit theorem in statistics combined with the 3-sigma principle in the case of large samples [sample size >= 30]. In the case of small sample sizes [sample size < 30], the data is sorted row by row based on the trimmed principle (TrimmedRule), and a certain proportion of the data closest to both ends is removed;

[0114] Finally, the mean based on the bootstrap method is defined as the arithmetic mean of the arithmetic means calculated by randomly sampling a fixed number of samples multiple times.

[0115] The following combines Figure 3 to illustrate the "fluctuation" analysis process based on short-term satisfaction data: The significance of the difference in customer-level satisfaction data between this month and last month is tested through the two independent t-test method. The null hypothesis of this test is that there is no significant difference in the overall customer satisfaction this month compared to last month. For the selected significance level alpha, the test statistic is constructed as:

[0116]

[0117] where

[0118]

[0119]

[0120]

[0121] When the null hypothesis holds, the statistic follows a t(n + m - 2) distribution, and based on this, the p-value of the test can be calculated. When p < alpha, the null hypothesis of the test is rejected, and the system will output a signal of significant difference; otherwise, a signal of non-significant difference is output.

[0122] The following combines Figure 4 to illustrate the "trend" analysis process based on long-term satisfaction data: The system retrieves the details of customer-level satisfaction survey data for all historical months and predicts the overall customer satisfaction for the next two months with the help of a time series model. Specifically, before fitting the time series model, it is necessary to perform stationarity and non-white noise tests on the data to prove that the data is stationary and has temporal correlation. The steps of the "trend" analysis for predicting customer potential satisfaction are as follows:

[0123] Retrieve the details of customer-level satisfaction survey data for all historical months, and use the unit root test method (Dickey-Fuller Test) to test the stationarity of the sequence, proving that the time series data is non-stationary and the sequence has a unit root;

[0124] Perform a non-white noise test on the sequence data to prove that the data is non-white noise and has temporal correlation;

[0125] Calculate the overall monthly satisfaction level, then train a time series model based on the overall satisfaction data of all months throughout history, and predict the overall customer satisfaction level for the next two months.

[0126] The following combination Figure 5 The monthly customer satisfaction alert process is as follows: The alert module adaptively judges the decline in customer satisfaction based on the output of the previous system monitoring and analysis modules and provides corresponding alerts. If the signal output by the monitoring module is no decline or the "trend" analysis module based on long-term satisfaction data does not predict a continuous decline, the alert module will not be triggered. If the signal output by the monitoring module is a decline and the "fluctuation" analysis module based on short-term satisfaction data determines a significant difference, the alert module will trigger a "customer churn alert". If the signal output by the monitoring module is a decline and the "trend" analysis module based on long-term satisfaction data gives a prediction of a continuous decline, the alert module will trigger a "gradual decline in satisfaction alert".

[0127] The following explains the monthly granular customer satisfaction attribution process: The attribution module returns attribution conclusions based on the input from the alarm module. If a "customer churn alarm" is received from the alarm module, the attribution module will locate the reasons for the low score from two aspects: dissatisfied customers and dissatisfaction dimensions. If a "gradual decline in satisfaction alarm" is received from the alarm module, the attribution module will analyze the decline in satisfaction from the perspective of two long-term evolutionary processes: changes in customer structure and changes in work order structure.

[0128] Regarding the issues mentioned in the background technology, specifically for issue 1, the monthly granular customer satisfaction monitoring module sequentially processes the detailed customer satisfaction survey data for the current month, handling missing data, invalid data, duplicate data, and outliers. Based on this, it outputs the bootstrap mean as the monthly overall satisfaction score. Outlier handling, in the case of a large sample, uses the Classic Central Limit Theorem (CCLT) and the Three-Sigma Rule of Thumb to determine the outlier range. In the case of a small sample, the Trimmed Rule is used to remove data from the closest two extremes at a certain ratio, aiming to effectively avoid the impact of outliers on the overall customer satisfaction monitoring.

[0129] Regarding question 2, the "fluctuation" analysis module based on short-term satisfaction data uses a double independent t-test to compare the significance of the difference in customer satisfaction between this month and last month; the "trend" analysis module based on long-term satisfaction data first performs stationarity and non-white noise tests on the data to prove that the data is stationary and has temporal correlation, and then uses time series analysis to predict the value of customer satisfaction in the next two months to determine whether the decline this month represents a downward trend in the long term. The goal of both is to objectively judge the degree of change in satisfaction based on scientific statistical theory.

[0130] Regarding question 3, the monthly granular customer overall satisfaction alarm module provides customer churn alarms or satisfaction decline alarms based on the input from the system's previous analysis module. The monthly granular customer overall satisfaction attribution module provides the attribution for the decline in satisfaction this month based on the signal from the system alarm module. The goal of the two modules is to realize the analysis link for analyzing and attributing changes in satisfaction.

[0131] This invention also provides a device for monitoring, alerting, and attributing satisfaction based on long-term and short-term historical data, comprising: at least one memory and at least one processor;

[0132] The at least one memory is used to store a machine-readable program;

[0133] The at least one processor is used to call the machine-readable program to implement the satisfaction monitoring, alarm and attribution method based on long and short-term historical data described in the above embodiments.

[0134] This invention also provides a computer-readable medium storing computer instructions. When executed by a processor, the computer instructions cause the processor to perform the satisfaction monitoring, alarm, and attribution method based on long-term and short-term historical data described in the above embodiments. Specifically, a system or apparatus equipped with a storage medium storing software program code that implements the functions of any of the above embodiments can be provided, and the computer (or CPU or MPU) of the system or apparatus can read and execute the program code stored in the storage medium.

[0135] In this case, the program code read from the storage medium can itself implement the function of any of the above embodiments, and therefore the program code and the storage medium storing the program code constitute part of the present invention.

[0136] Examples of storage media used to provide program code include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, program code can be downloaded from a server computer via a communication network.

[0137] Furthermore, it should be clear that not only can the program code read by the computer be executed, but also the operating system or other components operating on the computer can be instructed based on the program code to perform some or all of the actual operations, thereby realizing the function of any of the embodiments described above.

[0138] Furthermore, it is understood that the program code read from the storage medium is written to the memory set in the expansion board inserted into the computer or to the memory set in the expansion unit connected to the computer. Then, based on the instructions of the program code, the CPU or other components installed on the expansion board or expansion unit execute some and all of the actual operations, thereby realizing the function of any of the embodiments described above.

[0139] The present invention has been shown and described in detail above with reference to the accompanying drawings and preferred embodiments. However, the present invention is not limited to these disclosed embodiments. Based on the above embodiments, those skilled in the art will know that more embodiments of the present invention can be obtained by combining the code review methods in the different embodiments. These embodiments are also within the protection scope of the present invention.

Claims

1. A satisfaction monitoring, alerting and attribution system based on long short term historical data, characterized in that, The system comprises a monthly customer overall satisfaction monitoring module, a monthly customer overall satisfaction analysis module based on long-term and short-term satisfaction data, a monthly customer overall satisfaction alarm module, and a monthly customer overall satisfaction attribution module. The monthly customer overall satisfaction monitoring module calculates the monthly overall satisfaction. The monthly customer overall satisfaction analysis module comprises a "fluctuation" analysis module based on short-term satisfaction data and a "trend" analysis module based on long-term satisfaction data. The "fluctuation" analysis module compares the difference in customer satisfaction between the current month and the previous month by means of double independent t-test. The "trend" analysis module predicts the value of customer satisfaction in the next two months by time series analysis to determine whether the decline in the current month represents a long-term downward trend. The monthly customer overall satisfaction alarm module gives customer loss alarm or satisfaction downward trend alarm according to the input of the analysis module. The monthly customer overall satisfaction attribution module gives the attribution of the decline in the current month's satisfaction according to the signal of the alarm module. The calculation of monthly customer overall satisfaction is as follows: Missing data is replaced by the mean value in the monthly granularity dimension. Invalid data is replaced by the mean value in the monthly granularity dimension. Repeated data is deleted. In the case of a large sample, the abnormal value interval is determined based on the classical central limit theorem and the 3sigma law. In the case of a small sample, the most extreme data is removed according to the trimming principle. The "trend" analysis step for predicting customer potential satisfaction is as follows: The historical monthly customer granularity satisfaction survey data details are retrieved, and the unit root test method is used to test the stationarity of the sequence.

2. The system for satisfaction monitoring, alerting and attribution based on long short term historical data as claimed in claim 1, wherein, The non-white noise test is performed on the sequence data to prove that the data is non-white noise and has time correlation. The monthly overall satisfaction is calculated, and the time series model is trained based on all monthly granularity overall satisfaction data since history to predict the overall satisfaction of customers in the next two months. The implementation process of the system is as follows: 1) After the system obtains the monthly customer granularity satisfaction survey data details, it calculates the overall satisfaction of the current month by the monthly customer overall satisfaction monitoring module and compares it with the overall satisfaction of the previous month. If the satisfaction is higher than that of the previous month, the process ends, otherwise the data is transmitted to the monthly customer overall satisfaction analysis module. 2) The "fluctuation" analysis module based on short-term satisfaction data performs hypothesis test on the significance of the difference between the overall satisfaction of the current month and the previous month. If the overall satisfaction of the current month is significantly lower than that of the previous month, the system issues a "customer loss" alarm, otherwise the data is transmitted to the "trend" analysis module based on long-term satisfaction data. 3) Trend analysis module based on long-term satisfaction data uses historical customer granularity satisfaction survey data details of all months to conduct time series analysis to predict the overall satisfaction change in the next two months. If the model shows that customer satisfaction is continuously declining in the next two months, the system will issue a "continuous decline" warning if the overall satisfaction does not decrease significantly this month, otherwise the process ends. 4) After triggering the monthly customer overall satisfaction warning module, the system will adaptively attribute according to the warning category transmitted.

3. A method for satisfaction monitoring, alerting and attribution based on long short term historical data, the method is implemented based on the system of claim 1 or 2, characterized in that, The implementation of the method includes: Monthly customer overall satisfaction monitoring, performing monthly overall satisfaction calculation; Monthly customer overall satisfaction analysis based on long-term and short-term satisfaction data, including "fluctuation" analysis based on short-term satisfaction data and "trend" analysis based on long-term satisfaction data; "fluctuation" analysis based on short-term satisfaction data, using double independent t-test to compare the significance of the difference between this month and last month customer satisfaction; "trend" analysis based on long-term satisfaction data; predict the value of customer satisfaction in the next two months through time series analysis to determine whether the decline this month represents a long-term downward trend; Monthly customer overall satisfaction warning, giving customer churn warning or satisfaction decline warning according to the input of monthly customer overall satisfaction analysis; Monthly customer overall satisfaction attribution, giving the attribution of this month's satisfaction decline according to the signal of monthly customer overall satisfaction warning; The way to calculate monthly customer overall satisfaction is as follows: Missing data processing uses the mean value under the monthly granularity dimension to replace the missing values; Invalid data processing uses the mean value under the monthly granularity dimension to replace the invalid values that are not between 0 and 100; Repeated data processing deletes repeated records; Outlier processing: under the condition of large sample, based on the classical central limit theorem and 3sigma law to give the outlier interval, under the condition of small sample, use the truncation principle to remove the data closest to both ends according to the set proportion; Based on the bootstrap method, the process of repeatedly randomly sampling a fixed number of samples to calculate the arithmetic mean value and calculating its arithmetic mean value; The "trend" analysis steps to predict customer potential satisfaction are as follows: Retrieve historical customer granularity satisfaction survey data details of all months, use unit root test method to test the stationarity of the sequence, prove that the time series data is unstable and the sequence has unit root; Non-white noise test on sequence data, prove that the data is non-white noise and has time correlation; Calculate the monthly overall satisfaction, then train the time series model based on all monthly granularity overall satisfaction data since history and predict the overall satisfaction of customers in the next two months.

4. The method of claim 3, wherein, The "fluctuation" analysis steps to calculate the significance of the difference between monthly satisfaction are as follows: First, define the null hypothesis that there is no significant difference in this month's customer overall satisfaction compared to last month, construct the statistic of double independent t-test for the selected significance level alpha: ， Based on this, calculate the test p-value; when p<alpha, the null hypothesis of the test is rejected, the system will output the signal of significant difference, otherwise output the signal of insignificant difference. ， ， ， The running steps of satisfaction abnormal decline warning are as follows:

5. The method of claim 3, wherein the method further comprises: ​ If the signal output by the monthly granularity customer overall satisfaction monitoring is not decreased or the "trend" analysis based on long-term satisfaction data does not give a prediction of continuous decrease, the alarm is not triggered; If the signal output by the monthly granularity customer overall satisfaction monitoring is decreased and the "fluctuation" analysis based on short-term satisfaction data determines that the difference is significant, the alarm triggers a "customer churn alarm"; If the signal output by the monthly granularity customer overall satisfaction monitoring is decreased and the "trend" analysis based on long-term satisfaction data gives a prediction of continuous decrease, the alarm triggers a "satisfaction dark fall alarm".

6. The method of claim 5, wherein the method is based on long short-term history data for monitoring, alerting, and attribution of satisfaction. The operation steps for attributing the abnormal decrease of satisfaction are as follows: If the "customer churn alarm" is received, the low-score reasons are located from the aspects of dissatisfied customers and dissatisfied dimensions; If the "satisfaction dark fall alarm" is received, the decrease of satisfaction is analyzed from the aspects of customer structure change and ticket structure change, which are two long-term evolution processes.

7. A satisfaction monitoring, alerting and attribution apparatus based on long short term historical data, characterized by, The system comprises: at least one memory and at least one processor; the at least one memory is used to store machine readable programs; the at least one processor is used to call the machine readable programs to realize the method in any one of claims 3 to 6.

8. A computer readable medium characterized by The computer readable medium stores computer instructions, and the computer instructions make the processor execute the method in any one of claims 3 to 6 when executed by the processor. The computer readable medium stores computer instructions, and the computer instructions make the processor execute the method in any one of claims 3 to 6 when executed by the processor.

Images