CN115062858A - User complaint behavior prediction method, device, equipment and storage medium - Google Patents

Abstract

The application provides a method, a device, equipment and a storage medium for predicting user complaint behaviors, wherein the method comprises the following steps: acquiring personal information of a target user, a real complaint amount of the target user on a target product in a recent preset time range and use behavior data; predicting the complaint quantity of the target user in a subsequent time range based on the real complaint quantity and the trained sequence prediction model, and determining the current predicted complaint quantity from the complaint quantity in the subsequent time range; and predicting the complaint probability of the target user based on the current predicted complaint amount, the personal information of the target user and the use behavior data so as to obtain predicted complaint users which are quantitatively matched with the predicted complaint amount. According to the method and the device, the accuracy of the prediction of the complaint probability of the user is improved, an accurate list of potential complaint users is provided, classification processing can be performed on different users, and the complaint rate of the user is effectively reduced. The user satisfaction is improved.

Description

User complaint behavior prediction method, device, equipment and storage medium

technical field

Embodiments of the present invention relate to the technical field of artificial intelligence, and in particular, to a method, device, device, and storage medium for predicting user complaint behavior.

Background technique

With the continuous development of mobile business, the number of complaints from members of various products, super members and other privileged users is also increasing. Complaints from users of rights and interests are a key indicator reflecting the quality of a supplier's service.

Regarding the complaints of rights and interests users, there are currently only technical solutions that can predict the probability of complaints from potential complaining users.

However, the current complaint prediction scheme only covers static user data, which cannot fully reflect the possible complaints behavior of users, and cannot predict the amount of complaints in a certain period of time in the future, which makes it impossible to dynamically screen out potential complaining users in a certain period of time. Timely handling of potential complaints from users, resulting in a high rate of user complaints.

SUMMARY OF THE INVENTION

The present application provides a method, device, equipment and storage medium for predicting user complaint behavior, so as to solve the problem that the existing technology cannot fully reflect the complaint behavior that may be generated by users, and cannot predict the amount of complaints in a certain period of time in the future, resulting in the inability to accurately screen out a certain complaint behavior. Potential complaining users in a period of time will be dealt with in a timely manner, resulting in a high rate of user complaints.

In the first aspect, the present application provides a method for predicting user complaint behavior, including:

Obtain the target user's personal information, the target user's real complaint volume and usage behavior data for the target product within the most recent preset time range.

Based on the real complaint volume and the trained sequence prediction model, the target user's complaint volume in the subsequent time range is predicted, and the current predicted complaint volume is determined from the complaint volume in the subsequent time range.

The complaint probability of the target user is predicted based on the current predicted complaint volume, the target user's personal information and the usage behavior data, so as to obtain the predicted complaining users matching the predicted complaint volume in quantity.

Optionally, the trained sequence prediction model is a seasonal time series model ISARIMA model.

The prediction of the complaint volume within the subsequent time range of the target user based on the real complaint volume and the trained sequence prediction model includes:

The true complaint volume is input into the seasonal time series model ISARIMA model.

The seasonal time series model ISARIMA model is used to predict the complaint volume of the target user within the subsequent time range according to the real complaint volume, and output the complaint volume within the subsequent time range.

Optionally, predicting the complaint probability of the target user based on the current predicted complaint volume, the personal information of the target user, and the usage behavior data, so as to obtain the predicted complaint user that matches the predicted complaint volume in quantity. ,include:

Predict the complaint probability of the target user based on the current predicted complaint volume, the target user's personal information, the usage behavior data and the trained complaint user prediction model, so as to obtain a prediction that quantitatively matches the predicted complaint volume Complain about the user.

Optionally, the target user's complaint probability is predicted based on the current predicted complaint volume, the target user's personal information, the usage behavior data and the trained complaint user prediction model, so as to obtain the same number of complaints as the predicted complaint volume. A matched number of predicted complaining users, including:

The current predicted complaint volume, the target user's personal information and the usage behavior data are input into the trained complaint user prediction model.

The trained complaint user prediction model is used to predict the complaint probability of the target user according to the target user's personal information and the usage behavior data.

The trained complaint user prediction model is used to screen out the predicted complaint users equal to the predicted complaint amount according to the current predicted complaint amount and the target user's complaint probability.

Optionally, before predicting the complaint volume of the target user in the subsequent time range based on the real complaint volume and the trained sequence prediction model, the method further includes:

Obtain the first training sample set for training the initial sequence prediction model.

Determine the range of values for each training parameter used to train the initial sequence prediction model.

The value range of each training parameter is traversed to determine all possible values of the training parameter group composed of the training parameters.

All possible values of the training parameter group are traversed, and the initial sequence prediction model is trained by using the first training sample set to obtain the sequence prediction model with the smallest fitting goodness index AIC.

The sequence prediction model with the smallest fit goodness index AIC was determined as the trained sequence prediction model.

Optionally, the trained complaint user prediction model is an error back-propagation BP model.

The target user's complaint probability is predicted based on the current predicted complaint volume, the target user's personal information, the usage behavior data and the trained complaint user prediction model, so as to obtain a quantitative match with the predicted complaint volume. Predictions of complaining users before, also include:

A sample set for training and testing the initial complaint user prediction model is obtained, and the division ratio of the second training sample set and the test sample set in the sample set is determined according to the prediction accuracy.

After determining the division ratio of the second training sample set and the test sample set in the sample set, train and test the initial complaint user prediction model according to the second training sample set and the test sample set in the divided ratio, and perform training and testing during the training and testing process. , adjust the number of neural network layers and training parameters in the initial complaint user prediction model.

The complaint user prediction model with the highest prediction accuracy is determined as the trained complaint user prediction model.

Optionally, before obtaining the sample set for training and testing the initial complaint user prediction model, the method further includes:

Get the original dataset.

A structured query SQL statement is used to extract the target user's personal information and usage behavior data from the original data set, and the target user's personal information and the usage behavior data are preprocessed and stored in a corresponding sample set table.

Optionally, the method further includes:

Compare the predicted volume of complaints in the follow-up time range with the actual volume of complaints in the follow-up time range to obtain a comparison result.

If it is determined that the comparison result satisfies the parameter adjustment condition of the sequence prediction model, the parameter adjustment process is performed on the trained sequence prediction model.

In a second aspect, the present application provides a user complaint behavior prediction device, including:

The acquisition module is used to acquire the personal information of the target user, the actual amount of complaints of the target user about the target product within the most recent preset time range, and the usage behavior data.

The first prediction module is used to predict the complaint volume of the target user within the subsequent time range based on the real complaint volume and the trained sequence prediction model, and determine the current predicted complaint volume from the complaint volume within the subsequent time range .

The second prediction module is configured to predict the complaint probability of the target user based on the current predicted complaint volume, the target user's personal information and the usage behavior data, so as to obtain the predicted complaints that match the predicted complaint volume in quantity user.

In a third aspect, the application provides an electronic device, comprising:

a processor, and a memory in communication with the processor.

The memory stores computer-executable instructions.

The processor executes computer-executable instructions stored in the memory to implement the method of any one of the first aspects.

In a fourth aspect, the present application provides a computer-readable storage medium, where computer-executable instructions are stored in the computer-readable storage medium, and when the computer-executable instructions are executed by a processor, are used to implement any one of the first aspect. method described.

The user complaint behavior prediction method provided in this application obtains the personal information of the target user, the actual amount of complaints of the target user against the target product within the most recent preset time range, and the data on usage behavior. Based on the real complaint volume and the trained sequence prediction model, the target user's complaint volume in the subsequent time range is predicted, and the current predicted complaint volume is determined from the complaint volume in the subsequent time range. The complaint probability of the target user is predicted based on the current predicted complaint volume, the target user's personal information and the usage behavior data, so as to obtain the predicted complaining users matching the predicted complaint volume in quantity. Since the trained sequence prediction model can predict the volume of complaints in the subsequent time range, the current volume of complaints corresponding to a certain period in the future can be determined from the subsequent time range. And when predicting the complaint volume, the real complaint volume of the target product is considered, so the accuracy of predicting the current complaint volume is improved. Since the impact of the current predicted complaint volume, the personal information of the target user, and the use behavior data is comprehensively considered when predicting the complaining users, it is possible to accurately predict the potential complaints in a certain period of time in the future that match the predicted number of complaints. user. It can deal with potential complaint users in a timely manner, effectively reduce the user complaint rate, and improve user satisfaction and product experience.

Description of drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the application and together with the description serve to explain the principles of the application.

1 is an application scenario diagram of a method for predicting user complaint behavior provided by an embodiment of the present application;

2 is a method flow chart of a method for predicting a user complaint behavior provided in Embodiment 1 of the present application;

3 is a method flow chart of a method for predicting a user complaint behavior provided in Embodiment 2 of the present application;

4 is a method flow chart of a method for predicting a user complaint behavior provided in Embodiment 4 of the present application;

5 is a method flow chart of a method for predicting a user complaint behavior provided in Embodiment 5 of the present application;

6 is a method flow chart of a method for predicting a user complaint behavior provided in Embodiment 6 of the present application;

FIG. 7 is a method flow chart of a method for predicting a user complaint behavior provided in Embodiment 7 of the present application;

FIG. 8 is a method flow chart of a method for predicting a user complaint behavior provided in Embodiment 8 of the present application;

FIG. 9 is a schematic structural diagram of a user complaint behavior prediction device provided in Embodiment 1 of the present application;

FIG. 10 is a schematic structural diagram of an electronic device provided in Embodiment 1 of the present application.

Detailed ways

Exemplary embodiments will be described in detail herein, examples of which are illustrated in the accompanying drawings. Where the following description refers to the drawings, the same numerals in different drawings refer to the same or similar elements unless otherwise indicated. The implementations described in the illustrative examples below are not intended to represent all implementations consistent with this disclosure. Rather, they are merely examples of apparatus and methods consistent with some aspects of the present disclosure as recited in the appended claims.

In order to clearly understand the technical solutions of the present application, the solutions of the prior art are first introduced in detail.

The existing complaint prediction scheme only covers static user data, such as user personal information, and cannot fully reflect the user's possible complaint behavior, so the prediction of user complaint behavior is not accurate enough, and the existing technology cannot predict the daily complaint volume in the future period, resulting in It is impossible to accurately screen out the potential complaining users who predict the number of complaints per day, so that the product cannot be improved in a timely manner, resulting in a higher user complaint rate and a lower product experience.

Based on the inaccurate prediction of user complaint behavior in the prior art, and the problem that the predicted number of potential complaining users cannot be accurately screened, the inventor found through creative research that the time series-based model can be used to predict the amount of complaints in the future period, and in order to Improve the accuracy of the amount of complaints in the future period, and then use the time series model to predict the amount of complaints in the future period, you can use the target user's real complaints about the target product within the most recent preset time range to predict the target user's follow-up time range The current complaint volume corresponding to a certain period in the future can be determined from the complaint volume in the subsequent time range. In order to accurately predict the potential complaint users in a certain period of time in the future that matches the predicted number of complaints, it is necessary to comprehensively consider the influence of the current predicted complaint volume, the personal information of the target users, and the usage behavior data, so the above influencing factors can be comprehensively considered. , to predict the potential complaining users at a certain time in the future. The time series model can be a trained sequence prediction model. Through the embodiments of the present application, potential complaining users in a certain period in the future that match the predicted amount of complaints can be accurately predicted, the user complaint rate can be reduced, and the user satisfaction and the experience of using the product can be improved.

The following introduces the network architecture and application scenarios of the user complaint behavior prediction method provided by the present application.

FIG. 1 is an application scenario diagram of the user complaint behavior prediction method provided by the present application. The network structure corresponding to the user complaint behavior prediction method provided in the embodiment of the present application may include: a database 1 , an electronic device 2 , and a user terminal 3 . The electronic device 2 may be a server. The electronic device 2 periodically reads the target user's real complaint volume for the target product, the user's personal information and usage behavior data from the database 1, and inputs the real complaint volume within the most recent preset time range into the trained sequence prediction model to obtain Predict the number of complaints in the follow-up time range, determine the current predicted number of complaints, input the current predicted number of complaints, user personal information and user behavior data into the trained complaint user prediction model, and get the same number of potential complaint users as the current predicted complaint volume . The user terminal 3 displays, on the complaint prediction page, the predicted amount of complaints in the follow-up time range of a certain product and the list of potential complaint users.

The user complaint behavior prediction method provided by the present application can be specifically applied to the complaint prediction of various mobile service products. For example, the user ordered a traffic pack product. The electronic device 2 obtains the data on the user complaint volume of the data package product in the last half year from the database 1 every week, and inputs it into the time series prediction model to predict the daily user complaint volume in the next week. Determine the current predicted complaint volume, obtain the personal information and usage behavior data of Xunxinxiang product users on the date corresponding to the current predicted complaint volume, and input them into the complaint user prediction model to predict the complaint probability of data package product users. Sort the complaint probability in descending order, and filter out the potential complaint users whose number matches the current predicted complaint volume.

The technical solutions of the present application and how the technical solutions of the present application solve the above-mentioned technical problems will be described in detail below with specific examples. The following specific embodiments may be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below with reference to the accompanying drawings.

Example 1

FIG. 2 is a method flowchart of a method for predicting a user complaint behavior according to Embodiment 1 of the present application. In this embodiment, the existing technology cannot predict the amount of complaints in a certain period of time in the future, resulting in the inability to dynamically screen out potential complaining users in a certain period of time, resulting in a high user complaint rate and a low user experience of the product, and provides a user Complaint behavior prediction method. The method in this embodiment is applied to an apparatus for predicting user complaint behavior, and the apparatus for predicting user complaint behavior may be located in an electronic device. Among them, the electronic device can be a digital computer representing various forms. Such as cell phones, smart phones, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. As shown in FIG. 2 , the method for predicting user complaint behavior in the embodiment of the present application includes the following steps:

Step S101 , acquiring the personal information of the target user, the actual amount of complaints of the target user about the target product within the most recent preset time range, and the data on usage behavior.

Among them, the user's personal information refers to the user's personal basic information. Such as gender, age, number, province and other information.

Among them, the usage behavior data refers to the behavior data of the user in the process of using the product. Including user network data, user product usage data, user tracking behavior data, user complaint data, user-derived tag data, etc. Exemplarily, the user network data may include total call charges, total traffic, whether to bill or not. User product usage data may include information such as order time, order channels, and membership status. The user tracking behavior data may include a series of click events of the user on the order page, and may include information such as click time, number of clicks, and browsing time. User complaint data may include the user's complaint time, complaint reason, processing result and other data. User-derived tag data may include data such as user unsubscribe probability, user product preference, and the like. The most recent preset time range is not limited. It can be the last half year or the last year, etc.

Specifically, the target user's personal information, the target user's real complaints about the target product within the most recent preset time range, and the use behavior data can be stored in the data table of the preset database, and the data can be obtained by accessing the data table nationwide or in a certain country. The personal information of the target user of the target product within the province and the daily usage behavior data of the target product user within the most recent preset time range.

Step S102: Predict the complaint volume of the target user in the subsequent time range based on the real complaint volume and the trained sequence prediction model, and determine the current predicted complaint volume from the complaint volume in the subsequent time range.

Wherein, the trained sequence prediction model is not limited in scope, and can be various time series models, such as AR model, MA model, and the like. The follow-up time range is not limited. Can be for the next week or the next month.

Specifically, the real complaint volume in the latest preset time range of the whole country or a certain province is regularly input into the trained time series model, and the predicted complaint volume in the subsequent time range of the target user is obtained. Exemplarily, a certain day in the subsequent time range is determined as the current date, and the current predicted complaint volume is determined as the current predicted complaint volume.

Exemplarily, the daily real complaint volume for nearly half a year is input into the time series model, and the predicted complaint volume for the next week from May 24 to May 30 is obtained, and the predicted complaint volume on May 24 can be determined as the current forecast. volume of complaints.

Step S103: Predict the complaint probability of the target user based on the current predicted complaint volume, the target user's personal information and the usage behavior data, so as to obtain the predicted complaint users matching the predicted complaint volume in quantity.

Specifically, on the date corresponding to the current predicted complaint volume, the target user's complaint probability is predicted according to the current predicted complaint volume, the target user's personal information, and the target user's usage behavior data within the most recent preset time range. The prediction method is not specifically limited. Sort the predicted complaint probabilities in descending order, and then determine the predicted complaint users whose number is consistent with the current predicted complaint volume. In the embodiment of the present application, the personal information of the target user, the actual amount of complaints of the target user about the target product within the most recent preset time range, and the usage behavior data are obtained. Based on the real complaint volume and the trained sequence prediction model, the target user's complaint volume in the subsequent time range is predicted, and the current predicted complaint volume is determined from the complaint volume in the subsequent time range. Predict the complaint probability of the target user based on the current predicted complaint volume, the target user's personal information and usage behavior data, so as to obtain the predicted complaint users that match the predicted complaint volume in quantity. Due to the use of a time series-based forecasting model, it is possible to predict the volume of complaints in the subsequent time range and determine the current forecast volume of complaints. When predicting the current predicted complaint volume, since the real complaint volume is used as the data basis, the accuracy of the current predicted complaint volume is improved. Due to the comprehensive consideration of the impact of the current forecasted complaint volume, target users’ personal information, and usage behavior data, it is possible to obtain potential complaining users that match the predicted complaint volume, and then urge suppliers to deal with potential complaining users and improve Product quality, in order to reduce the complaint rate, improve user satisfaction and user product experience.

Embodiment 2

FIG. 3 is a flowchart of a method for predicting user complaint behavior provided by the second embodiment of the present application. As shown in FIG. 3 , based on the above embodiment, the embodiment of the present application involves step S102 based on the actual amount of complaints and the trained sequence prediction model. A specific implementation method for predicting the amount of complaints in the subsequent time range, wherein the trained sequence prediction model is the seasonal time series model ISARIMA model. Based on the real complaint volume and the trained sequence prediction model, the target user's complaint volume in the follow-up time range is predicted, including the following steps:

Step S201, input the real complaint volume into the seasonal time series model ISARIMA model.

Among them, SARIMA model is a seasonal time series model used to describe time series with seasonal variation. ISARIMA is an improved seasonal time series model.

Specifically, the daily real complaint volume of the most recent preset time range is input into the seasonal time series model ISARIMA model to predict the complaint volume of the subsequent time range.

Step S202 , using the seasonal time series model ISARIMA model to predict the amount of complaints of the target user within the subsequent time range according to the actual amount of complaints, and output the amount of complaints within the subsequent time range.

The follow-up time range is not limited, and can be a week or a month in the future.

Specifically, the seasonal time series model ISARIMA model receives the daily real complaint volume in the latest preset time range, predicts the daily complaint volume in the subsequent time range based on the daily real complaint volume in the latest preset time range, and outputs the subsequent Daily forecast complaint volume for the time frame.

In the embodiment of the present application, the trained sequence prediction model is the seasonal time series model ISARIMA model. Input the true complaint volume into the seasonal time series model ISARIMA model. The seasonal time series model ISARIMA model is used to predict the complaint volume of the target user in the subsequent time range according to the real complaint volume, and output the complaint volume in the subsequent time range. Since the amount of user complaints has periodic changes, and the seasonal time series model SARIMA is used to describe the time series with seasonal or periodic changes, therefore, the use of the improved seasonal time series model ISARIMA can improve the forecast in the subsequent time range. The accuracy of the daily complaint volume, and the ISARIMA is trained, so further improves the accuracy of the daily complaint volume prediction.

Embodiment 3

Based on the content of Embodiment 1 or Embodiment 2, the embodiment of the present application involves step S103 to predict the complaint probability of the target user based on the current predicted complaint volume, the target user's personal information and usage behavior data, so as to obtain and predict the number of complaints. A specific achievable way of matching predicted complaining users, including:

Based on the current predicted complaint volume, the target user's personal information, usage behavior data and the trained complaint user prediction model, the target user's complaint probability is predicted to obtain the predicted complaint users that match the predicted complaint volume in quantity.

Among them, the trained complaint user prediction model used in this application is a BP model of error back propagation, and the trained complaint user prediction model is not specifically limited, and can be exemplarily an LSTM model, or an iterative decision tree model, logic Either regression model or random forest model.

Specifically, on the date corresponding to the current predicted complaint volume, input the current predicted complaint volume, the target user's personal information and the target user's usage behavior data in the most recent preset time range into the trained complaint user prediction model, and output the target user's predicted complaint probability . Sort the predicted complaint probabilities in descending order, and then determine the predicted complaint users whose number is consistent with the current predicted complaint volume.

In the embodiment of the present application, the complaint probability of the target user is predicted based on the current predicted complaint volume, the target user's personal information, usage behavior data and the trained complaint user prediction model, so as to obtain a prediction that matches the predicted complaint volume in quantity. Complain about the user. When predicting the current predicted complaint volume, since the real complaint volume is used as the data basis, the accuracy of the current predicted complaint volume is improved. Since the trained complaint user prediction model is used, the user complaint probability can be accurately predicted. Due to the comprehensive consideration of the impact of the current forecasted complaint volume, target users’ personal information, and usage behavior data, it is possible to obtain potential complaining users that match the predicted complaint volume, and then urge suppliers to deal with potential complaining users and improve Product quality, in order to reduce the complaint rate, improve user satisfaction and user product experience.

Embodiment 4

FIG. 4 is a flowchart of a method for predicting user complaint behavior provided by Embodiment 4 of the present application. As shown in FIG. 4 , based on the content of Embodiment 3, this embodiment of the present application relates to Embodiment 3 based on the current predicted complaint volume, target user personal information, A concrete and achievable way to predict the complaint probability of target users by using behavioral data and the trained complaint user prediction model to obtain the predicted complaint users that match the predicted complaint volume in quantity. Predict the complaint probability of the target user based on the current predicted complaint volume, the target user's personal information, usage behavior data and the trained complaint user prediction model to obtain the predicted complaint users that match the predicted complaint volume, including the following steps :

Step S301 , input the current predicted amount of complaints, the target user's personal information and usage behavior data into the trained complaint user prediction model.

Specifically, the current predicted complaint volume, the target user's personal information and the target user's usage behavior data within the most recent preset time range are input into the trained complaint user prediction model to predict the target user's complaint probability.

Step S302, using the trained complaint user prediction model to predict the complaint probability of the target user according to the target user's personal information and usage behavior data.

Specifically, the trained complaint user prediction model receives the personal information of the target user, the current predicted volume of complaints, and the usage behavior data of the target user within a preset time range, and uses the corresponding network layer to analyze the personal information of the target user, the current predicted volume of complaints, and the predicted volume of complaints. Set the use behavior data of the target user within the time range for feature extraction, prediction and other processing, and output the complaint probability of the target user.

Step S303 , using the trained complaint user prediction model to screen out the predicted complaint users equal to the predicted complaint volume according to the current predicted complaint volume and the target user's complaint probability.

Specifically, the complaint probabilities of the target users are sorted in descending order, and the target users corresponding to the top N complaint probabilities are determined as the predicted complaining users.

Among them, N is the predicted number of complaints, and it is also the predicted number of complaining users.

In the embodiment of the present application, the current predicted complaint volume, the target user's personal information and the usage behavior data are input into the trained complaint user prediction model. The trained complaint user prediction model is used to predict the target user's complaint probability based on the target user's personal information and usage behavior data. The trained complaint user prediction model is used to screen out the predicted complaint users equal to the predicted complaint volume according to the current predicted complaint volume and the target user's complaint probability. Since the personal information of the target users and the user behavior data are considered, the probability of user complaints can be more accurately predicted. Since the current predicted number of complaints is used, the potential complaining users with the current predicted number of complaints can be accurately screened.

Embodiment 5

FIG. 5 is a flowchart of a method for predicting user complaint behavior provided in Embodiment 5 of the present application. As shown in FIG. 5 , based on the content of any one of Embodiments 1 to 4, this embodiment of the present application is based on the actual amount of complaints in step S102 and the trained sequence prediction model to predict the volume of complaints of the target user in the subsequent time range, it also includes the step of training the sequence prediction model, then the process of training the sequence prediction model includes the following steps:

Step S401: Obtain a first training sample set for training the initial sequence prediction model.

The first training sample set refers to the training sample set of the initial sequence prediction model.

Specifically, the daily user complaint volume within the most recent preset time range of the target product is used as the first training sample set, and the daily user complaint volume within the most recent preset time range is obtained from the database.

The most recent preset time range is not limited to a specific range, and may be the most recent half year or the most recent year.

Step S402: Determine the value range of each training parameter for training the initial sequence prediction model.

Among them, the initial series prediction model has two types of parameters, namely trend parameters and seasonal parameters. The trend parameters include the maximum lag order ps of the trend autoregression, the number of differences in the trend d, and the maximum lag order qs of the moving average of the trend. The seasonal parameters include the maximum lag order Ps of the seasonal autoregression, the seasonal The number of differences D, the maximum lag order Qs of the seasonal moving average, and the number of time steps of a single seasonal cycle s.

Specifically, the value range of each training parameter of the initial sequence prediction model is set by a random function.

Step S403: Traverse the value range of each training parameter to determine all possible values of the training parameter group composed of the training parameters.

Specifically, the training parameter types in each training parameter group are consistent, but the value of at least one training parameter in different training parameter groups is different. Then, when each training parameter group is determined, the value range of each parameter is traversed, and the values of different training parameters are combined, so that all possible values of each training parameter can be reflected in the form of a training parameter group.

Step S404 , traverse all possible values of the training parameter group, and use the first training sample set to train the initial sequence prediction model to obtain the sequence prediction model with the smallest fitting goodness index AIC.

Specifically, the first training sample set is input into the initial sequence prediction model, and with each training parameter group as a dimension, the value of each training parameter in the corresponding training parameter group is brought into the initial sequence prediction model to perform the initial sequence prediction model. train.

Step S405: Determine the sequence prediction model with the smallest fitting goodness index AIC as the trained sequence prediction model.

Among them, the goodness of fit index AIC is a standard to measure the goodness of fit of statistical models. AIC encourages goodness of fit to the data but tries to avoid overfitting. So the preferred model should be the one with the smallest AIC value.

Specifically, the model evaluation value AIC is determined, and the sequence prediction model corresponding to the smallest model evaluation value is determined as the optimal trained model.

In the embodiment of the present application, the first training sample set for training the initial sequence prediction model is obtained. Determine the range of values for each training parameter used to train the initial sequence prediction model. The value range of each training parameter is traversed to determine all possible values of the training parameter group composed of the training parameters. Traverse all possible values of the training parameter group, and use the first training sample set to train the initial sequence prediction model to obtain the sequence prediction model with the smallest fitting goodness index AIC. The sequence prediction model with the smallest fit goodness index AIC was determined as the trained sequence prediction model. Since the automatic parameter adjustment of the sequence prediction model is realized based on the cycle, the parameter adjustment efficiency of the initial sequence prediction model is effectively improved. Since all the possible values of the parameters are traversed, the sequence prediction model is evaluated by the fit goodness index. After the evaluation, the optimal model is determined, so the accuracy of model parameter selection is improved.

Embodiment 6

FIG. 6 is a flowchart of a method for predicting user complaint behavior provided by Embodiment 6 of the present application. As shown in FIG. 6 , based on the content of any one of Embodiments 3 to 5, in this embodiment of the present application, the trained complaint user The prediction model is the BP model of error back propagation. In the third embodiment of the present application, the target user's complaint probability is predicted based on the current predicted complaint volume, the target user's personal information, usage behavior data and the trained complaint user prediction model, so as to obtain a prediction that matches the predicted complaint volume in quantity. Before complaining about the user, it also includes the steps of training the prediction model of the initial complaining user, and the process of training the prediction model of the initial complaining user is as follows:

Step S501: Obtain a sample set for training and testing the initial complaint user prediction model, and the division ratio of the second training sample set and the test sample set in the sample set is determined according to the prediction accuracy.

The second training sample set refers to the training sample set of the initial complaint user prediction model.

Specifically, users who have complained within a preset time range are used as positive samples, and a complaint user label is set. Take the uncomplained users within the preset time range as negative samples, and set the uncomplained user label.

The second training sample set and the test sample set are allocated according to the division ratios of 9:1, 8:2, 7:3, and 6:4, respectively. Based on the loop method, traverse each division ratio of the second training sample set and the test set, input the second training sample set into the initial complaint user prediction model for training, and use the test sample set for testing to determine the division ratio with the highest prediction accuracy is the ratio of the second training sample set and the test sample set.

Step S502: After determining the division ratio of the second training sample set and the test sample set in the sample set, train and test the initial complaint user prediction model according to the divided ratio of the second training sample set and the test sample set, and perform training and testing. During the process, adjust the number of neural network layers and training parameters in the initial complaint user prediction model.

Specifically, after obtaining the second training sample set and the test sample set after each division ratio in step S501, traverse different levels of the neural network in a circular manner, train the initial complaint user prediction model, and obtain accurate predictions The number of neural network layers at the highest rate.

Based on the grid search method, the values of the training parameters of the initial complaint user prediction model are traversed, and the value of the training parameters when the prediction accuracy is the highest is determined as the trained parameters.

Step S503: Determine the complaint user prediction model with the highest prediction accuracy as the trained complaint user prediction model.

Specifically, after the initial complaint user prediction model is trained through the above steps, the complaint user prediction model with the highest prediction accuracy is obtained as the trained complaint user prediction model.

In the embodiment of the present application, a sample set for training and testing the initial complaint user prediction model is obtained, and the division ratio of the second training sample set and the test sample set in the sample set is determined according to the prediction accuracy. After determining the division ratio of the second training sample set and the test sample set in the sample set, train and test the initial complaint user prediction model according to the divided ratio of the second training sample set and the test sample set, and during the training and testing process, Adjust the number of neural network layers and training parameters in the initial complaint user prediction model. Determine the complaint user prediction model with the highest prediction accuracy as the trained complaint user prediction model. Due to the use of a loop-based method to realize automatic parameter adjustment, the efficiency of parameter adjustment of the initial complaint user prediction model is improved. By repeatedly adjusting the parameters, the overfitting problem of the initial complaint user prediction model is solved, and the accuracy of identifying potential complaining users is improved. .

Embodiment 7

FIG. 7 is a flow chart of the method for predicting user complaint behavior provided by Embodiment 7 of the present application. As shown in FIG. 7 , based on the content of Embodiment 6, in the method for predicting user complaint behavior provided by the embodiment of the present application, the prediction of the initial complaint user is obtained. Before the model is trained and tested on the sample set, it also includes the following steps:

Step S601, acquiring an original data set.

Among them, the original data set refers to the data set composed of the original source data of the user complaint prediction model. In the original data set, it includes user personal information, user traffic information, user ordering and using product information, user tracking information on product ordering and using pages, and user complaint information. Get the raw dataset from the database.

Step S602, using structured query SQL statement to extract the target user's personal information and usage behavior data from the original data set, and store the target user's personal information and usage behavior data in the corresponding sample set table after preprocessing.

Among them, the personal information of the target user, including gender, age, number, province, etc. Usage behavior data, including user network data, user product usage data, user tracking behavior data, user complaint data, and user-derived tag data. User network data includes total call charges, total traffic, whether billing, total video traffic, video applications watched, mobile phone model, and mobile phone brand. User product usage data includes free flow, order time, order channel, membership status, and application usage. User tracking behavior data includes click events on the order page, click time, number of clicks, browsing time, product categories, and product names. User complaint data includes complaint time, complaint type, complaint product, complaint reason, processing result, whether to refund, and refund amount. User-derived tag data includes user unsubscribe probability, user product preference, and user VIP level.

Specifically, the first step is to create a sample data set table. Use structured query SQL statements to create tables in the database. The second step is data extraction. According to the data index system, the required data is extracted from the database using structured query SQL statements. The third step is data integration. Integrate user data through database connection operations to form a complete data set. The fourth step is data cleaning. For missing values, substitute the mean or median of the preset time range. For outliers, identify them by drawing box plots or distribution plots, and exclude outliers. For numeric fields, normalization is performed to reduce the risk of model accuracy being reduced due to large data magnitude gaps. For categorical variables, the ONE-HOT coding method is used to map categorical variables and digital identifiers one by one. The fifth step is to check the data quality. Check each field feature one by one to make the sample data meet the standard of the model input data. Finally, the sample data that meets the model input data standard after data processing are stored in the established sample data set table. Complete data modeling work.

In this embodiment of the present application, the original data set is obtained. The target user's personal information and usage behavior data are extracted from the database using structured query SQL statements, and the target user's personal information and usage behavior data are preprocessed and stored in the corresponding sample set table. The structured query SQL statement is used to integrate and clean the data when the table is built, which simplifies the data preprocessing process and effectively improves the data processing efficiency.

Embodiment 8

FIG. 8 is a flowchart of the method for predicting user complaint behavior provided by Embodiment 8 of the present application. As shown in FIG. 8 , based on the contents of any one of Embodiments 2 to 7, the method for predicting user complaint behavior provided by the embodiment of the present application is shown in FIG. 8 . , also includes the following steps:

Step S701: Compare the predicted volume of complaints within the subsequent time range with the actual volume of complaints within the subsequent time range to obtain a comparison result.

Specifically, after the predicted amount of complaints in the user's subsequent time range is determined, the predicted amount of complaints is compared with the actual amount of complaints on the corresponding date, and the difference between the two is compared with the actual amount of complaints to obtain a proportional value.

Step S702: If it is determined that the comparison result satisfies the parameter adjustment condition of the sequence prediction model, perform parameter adjustment processing on the trained sequence prediction model.

Specifically, when the ratio value obtained in the above steps is greater than the preset ratio threshold value, the parameter adjustment condition of the sequence prediction model is satisfied. Then the parameters of the sequence prediction model are automatically adjusted so that the ratio value is smaller than the preset ratio threshold value.

In the embodiment of the present application, the predicted amount of complaints in the subsequent time range is compared with the actual amount of complaints in the subsequent time range to obtain a comparison result. If it is determined that the comparison result satisfies the parameter adjustment conditions of the sequence prediction model, the parameter adjustment process is performed on the trained sequence prediction model. By comparing the predicted amount of complaints with the actual amount of complaints, the feedback information of the sequence prediction model is obtained, the relationship between the feedback information and the parameter adjustment conditions is judged, and it is decided whether to adjust the parameters of the sequence prediction model. The automatic feedback parameter adjustment of the sequence prediction model is realized, and the accuracy of the parameters of the sequence prediction model is improved.

Embodiment 9

FIG. 9 is a structural diagram of an apparatus for predicting user complaint behavior according to Embodiment 9 of the present application. As shown in FIG. 9 , the apparatus 80 for predicting user complaint behavior includes: an acquisition module 81 , a first prediction module 82 , and a second prediction module 83 .

The obtaining module 81 is used to obtain the personal information of the target user, the actual amount of complaints of the target user about the target product within the most recent preset time range, and the data on usage behavior.

The first prediction module 82 is configured to predict the complaint volume of the target user in the subsequent time range based on the real complaint volume and the trained sequence prediction model, and determine the current predicted complaint volume from the complaint volume in the subsequent time range.

The second prediction module 83 is configured to predict the complaint probability of the target user based on the current predicted complaint volume, the target user's personal information and usage behavior data, so as to obtain the predicted complaining users matching the predicted complaint volume in quantity.

Optionally, in the first prediction module 82, the trained sequence prediction model is a seasonal time series model ISARIMA model. In the prediction of the target user's complaint volume within the subsequent time range based on the real complaint volume and the trained sequence prediction model, it is specifically used for:

Input the true complaint volume into the seasonal time series model ISARIMA model. The seasonal time series model ISARIMA model is used to predict the complaint volume of the target user in the subsequent time range according to the real complaint volume, and output the complaint volume in the subsequent time range.

Optionally, the second prediction module 83 predicts the complaint probability of the target user based on the current predicted complaint volume, the personal information of the target user and the usage behavior data, so as to obtain the number of predicted complaint users matching the predicted complaint volume. , specifically for:

Based on the current predicted complaint volume, the target user's personal information, usage behavior data and the trained complaint user prediction model, the target user's complaint probability is predicted to obtain the predicted complaint users that match the predicted complaint volume in quantity.

Optionally, the second prediction module 83 predicts the complaint probability of the target user based on the current predicted complaint volume, the personal information of the target user, the usage behavior data and the trained complaint user prediction model, so as to obtain the same number of complaints as the predicted complaint volume. When complaining users on matching predictions, it is specifically used to:

Input the current predicted complaint volume, target user's personal information and usage behavior data into the trained complaint user prediction model. The trained complaint user prediction model is used to predict the target user's complaint probability based on the target user's personal information and usage behavior data. The trained complaint user prediction model is used to screen out the predicted complaint users equal to the predicted complaint volume according to the current predicted complaint volume and the target user's complaint probability.

Optionally, the apparatus 80 for predicting user complaint behavior provided in this embodiment further includes a first training module. Before the first prediction module 82 predicts the complaint volume of the target user within the subsequent time range based on the real complaint volume and the trained sequence prediction model, the first training module is specifically used for:

Obtain the first training sample set for training the initial sequence prediction model. Determine the range of values for each training parameter used to train the initial sequence prediction model. The value range of each training parameter is traversed to determine all possible values of the training parameter group composed of the training parameters. All possible values of the training parameter group are traversed, and the first training sample set is used to train the initial sequence prediction model to obtain the sequence prediction model with the smallest fitting goodness index AIC. The sequence prediction model with the smallest fit goodness index AIC was determined as the trained sequence prediction model.

Optionally, the apparatus 80 for predicting user complaint behavior provided in this embodiment further includes a second training module. In the second prediction module 83, the target user's complaint probability is predicted based on the current predicted complaint volume, the target user's personal information, the usage behavior data and the trained complaint user prediction model, so as to obtain a prediction that matches the predicted complaint volume in quantity. Before complaining about users, the second training module is specifically used to:

A sample set for training and testing the initial complaint user prediction model is obtained, and the division ratio of the second training sample set and the test sample set in the sample set is determined according to the prediction accuracy. After determining the division ratio of the second training sample set and the test sample set in the sample set, train and test the initial complaint user prediction model according to the divided ratio of the second training sample set and the test sample set, and during the training and testing process, Adjust the number of neural network layers and training parameters in the initial complaint user prediction model. Determine the complaint user prediction model with the highest prediction accuracy as the trained complaint user prediction model. Among them, the trained complaint user prediction model is the BP model of error back propagation.

Optionally, the apparatus 80 for predicting user complaint behavior provided in this embodiment further includes a data processing module. Before the acquisition module 81 acquires the sample set for training and testing the initial complaint user prediction model, the data processing module is specifically used for:

Get the original dataset. The target user's personal information and usage behavior data are extracted from the database using structured query SQL statements, and the target user's personal information and usage behavior data are preprocessed and stored in the corresponding sample set table.

Optionally, the apparatus 80 for predicting user complaint behavior provided in this embodiment further includes a parameter adjustment module. Specifically for:

Compare the predicted volume of complaints in the follow-up time range with the actual volume of complaints in the follow-up time range to obtain a comparison result. If it is determined that the comparison result satisfies the parameter adjustment conditions of the sequence prediction model, the parameter adjustment process is performed on the trained sequence prediction model.

Embodiment ten

FIG. 10 is a structural diagram of an electronic device provided in Embodiment 1 of the present application. As shown in FIG. 10 , the electronic device 90 includes: a processor 91 and a memory 92 .

Memory 92 stores computer executable instructions.

The processor 91 executes the computer-executed instructions stored in the memory 92 to implement the method of any one of the first embodiment to the seventh embodiment.

In this embodiment, the memory 92 is connected to the processor 91 through a bus 93 .

Embodiments of the present application further provide a computer-readable storage medium, where computer-executable instructions are stored in the computer-readable storage medium, and when the computer-executable instructions are executed by a processor, the method provided by any of the foregoing method embodiments is implemented.

An embodiment of the present application further provides a computer program product, including a computer program, and when the computer program is executed by a processor, the method provided by any one of the embodiments of the present application is implemented.

In the several embodiments provided in this application, it should be understood that the disclosed apparatus and method may be implemented in other manners. For example, the device embodiments described above are only illustrative. For example, the division of modules is only a logical function division. In actual implementation, there may be other division methods, for example, multiple modules or components may be combined or integrated. to another system, or some features can be ignored, or not implemented. On the other hand, the shown or discussed mutual coupling or direct coupling or communication connection may be through some interfaces, indirect coupling or communication connection of devices or modules, and may be in electrical, mechanical or other forms.

Modules described as separate components may or may not be physically separated, and components shown as modules may or may not be physical modules, that is, they may be located in one place, or may be distributed to multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution in this embodiment.

In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically alone, or two or more modules may be integrated into one module. The above-mentioned integrated modules can be implemented in the form of hardware, or can be implemented in the form of hardware plus software function modules.

Program code for implementing the methods of the present application may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable compliance detection device, such that the program codes, when executed by the processor or controller, perform the functions specified in the flowcharts and/or block diagrams /Operation is implemented. The program code may execute entirely on the machine, partly on the machine, partly on the machine and partly on a remote machine as a stand-alone software package or entirely on the remote machine or server.

In the context of this application, a machine-readable medium may be a tangible medium that may contain or store the program for use by or in connection with the instruction execution system, apparatus or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media may include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media would include one or more wire-based electrical connections, portable computer disks, hard disks, random access memory (RAM), read only memory (ROM), erasable programmable read only memory (EPROM or flash memory), fiber optics, compact disk read only memory (CD-ROM), optical storage, magnetic storage, or any suitable combination of the foregoing.

Additionally, although operations are depicted in a particular order, this should be understood to require that such operations be performed in the particular order shown or in a sequential order, or that all illustrated operations should be performed to achieve desirable results. Under certain circumstances, multitasking and parallel processing may be advantageous. Likewise, while the above discussion contains several implementation-specific details, these should not be construed as limitations on the scope of the application. Certain features that are described in the context of separate embodiments can also be implemented in combination in a single implementation. Conversely, various features that are described in the context of a single implementation can also be implemented in multiple implementations separately or in any suitable subcombination.

Other embodiments of the present application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses or adaptations of this application that follow the general principles of this application and include common knowledge or conventional techniques in the technical field not disclosed in this application . The specification and examples are to be regarded as exemplary only, with the true scope and spirit of the application being indicated by the following claims.

It is to be understood that the present application is not limited to the precise structures described above and illustrated in the accompanying drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the application is limited only by the appended claims.

Claims (11)

1. A method for predicting complaint behaviors of a user is characterized by comprising the following steps:

acquiring personal information of a target user, and real complaint amount and use behavior data of the target user on a target product within a latest preset time range;

predicting the complaint quantity of the target user in a subsequent time range based on the real complaint quantity and the trained sequence prediction model, and determining the current predicted complaint quantity from the complaint quantity in the subsequent time range;

and predicting the complaint probability of the target user based on the current predicted complaint amount, the personal information of the target user and the use behavior data so as to obtain predicted complaint users which are quantitatively matched with the predicted complaint amount.

2. The method of claim 1, wherein the trained sequence prediction model is a seasonal time series model (ISARIMA);

predicting the complaint amount of the target user in the subsequent time range based on the real complaint amount and the trained sequence prediction model, wherein the predicting comprises the following steps:

inputting the real complaint quantity into an ISARIMA model of a seasonal time series model;

and predicting the complaint amount of the target user in the subsequent time range according to the real complaint amount by adopting the seasonal time series model ISARIMA model, and outputting the complaint amount in the subsequent time range.

3. The method of claim 1, wherein predicting the probability of complaint for the target customer based on the current predicted complaint volume, target customer personal information, and the usage behavior data to obtain a predicted complaint customer quantitatively matching the predicted complaint volume comprises:

predicting the complaint probability of the target user based on the current predicted complaint amount, the personal information of the target user, the using behavior data and the trained complaint user prediction model so as to obtain predicted complaint users which are quantitatively matched with the predicted complaint amount.

4. The method of claim 3, wherein predicting the probability of complaint for the target user based on the current predicted complaint volume, target user personal information, the usage behavior data, and the trained complaint user predictive model to obtain predicted complaint users quantitatively matching the predicted complaint volume comprises:

inputting the current predicted complaint amount, the target user personal information and the use behavior data into a trained complaint user prediction model;

predicting the complaint probability of the target user by adopting the trained complaint user prediction model according to the personal information of the target user and the using behavior data;

and screening out the predicted complaint users equal to the predicted complaint amount by adopting the trained complaint user prediction model according to the current predicted complaint amount and the complaint probability of the target user.

5. The method of any one of claims 1-4, wherein prior to predicting the complaint volume at a subsequent time frame of the target user based on the real complaint volume and the trained sequence prediction model, further comprising:

acquiring a first training sample set for training an initial sequence prediction model;

determining the value range of each training parameter for training the initial sequence prediction model;

traversing the value range of each training parameter to determine all possible values of the training parameter group consisting of the training parameters;

traversing all possible values of a training parameter set, and training the initial sequence prediction model by adopting the first training sample set to obtain a sequence prediction model with the minimum fitting goodness index (AIC);

and determining the sequence prediction model with the minimum fit goodness index AIC as the trained sequence prediction model.

6. The method of claim 3 or 4, wherein the trained complaint user prediction model is a back propagation of error BP model;

before predicting the complaint probability of the target user based on the current predicted complaint quantity, the personal information of the target user, the usage behavior data and the trained complaint user prediction model to obtain predicted complaint users quantitatively matching the predicted complaint quantity, the method further comprises:

acquiring a sample set for training and testing an initial complaint user prediction model, wherein the division ratio of a second training sample set and a test sample set in the sample set is determined according to the prediction accuracy;

after a second training sample set and a test sample set in the sample set are determined to be divided into proportions, training and testing an initial complaint user prediction model according to the second training sample set and the test sample set which are divided into proportions, and adjusting the neural network layer number and training parameters in the initial complaint user prediction model in the training and testing process;

and determining the complaint user prediction model with the highest prediction accuracy as the trained complaint user prediction model.

7. The method of claim 6, wherein prior to obtaining the sample set for training and testing the initial complaint user predictive model, further comprising:

acquiring an original data set;

and extracting target user personal information and use behavior data from the original data set by adopting a structured query SQL statement, preprocessing the target user personal information and the use behavior data, and storing the preprocessed target user personal information and the preprocessed use behavior data into a corresponding sample set table.

8. The method according to any one of claims 2-4, further comprising:

comparing the predicted complaining amount in the subsequent time range with the real complaining amount in the subsequent time range to obtain a comparison result;

and if the comparison result meets the parameter adjusting condition of the sequence prediction model, performing parameter adjusting processing on the trained sequence prediction model.

9. A user complaint behavior prediction apparatus, comprising:

the acquisition module is used for acquiring personal information of the target user, the real complaint amount of the target user to the target product in the latest preset time range and the use behavior data;

the first prediction module is used for predicting the complaint amount of the target user in a subsequent time range based on the real complaint amount and the trained sequence prediction model and determining the current predicted complaint amount from the complaint amount in the subsequent time range;

and the second prediction module is used for predicting the complaint probability of the target user based on the current predicted complaint amount, the personal information of the target user and the using behavior data so as to obtain the predicted complaint users which are matched with the predicted complaint amount in quantity.

10. An electronic device, comprising: a processor, and a memory communicatively coupled to the processor;

the memory stores computer-executable instructions;

the processor executes computer-executable instructions stored by the memory to implement the method of any of claims 1 to 8.

11. A computer-readable storage medium having computer-executable instructions stored thereon, which when executed by a processor, are configured to implement the method of any one of claims 1 to 8.

Images