Training method of satisfaction prediction model, satisfaction prediction method and device

Abstract

The application provides a satisfaction prediction model training method, a satisfaction prediction method and device, applied to the technical field of data processing, through obtaining a plurality of user data samples and label information corresponding to each user data sample, updating parameters of an initial classification model according to a first user data sample in the plurality of user data samples to obtain a classification model, updating parameters of an initial regression model according to a second user data sample in the plurality of user data samples except the first user data sample to obtain a regression model, and finally determining a satisfaction prediction model according to the classification model and the regression model. Through fusing the classification model and the regression model, the prediction error can be effectively reduced, the accuracy of user satisfaction prediction is improved, and the distribution of the prediction result is closer to the distribution of the real user score, so that the operator can make further improvement according to the predicted satisfaction score.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a training method, a satisfaction prediction method and apparatus for a satisfaction prediction model. Background Technology

[0002] Currently, mobile operators have fully implemented number portability services, making improving user satisfaction of significant practical importance. Operators often need to predict user satisfaction scores based on existing data and then focus on maintaining relationships with users who have lower satisfaction scores.

[0003] In existing methods, regression models are frequently used to predict user satisfaction ratings. During the training of a regression model, to reduce the average error, the model will try to make the training results as close as possible to the mean of the ratings obtained from the training data. Therefore, when using a trained regression model to predict user satisfaction ratings, the mean of the ratings obtained from the prediction data accounts for the largest proportion of the predicted results.

[0004] However, the distribution of these predictions is almost always normally distributed, which often does not match the distribution of actual user ratings. For example, in reality, user ratings tend to be concentrated around 1 and 10, while the ratings predicted by the regression model tend to be concentrated around 7 or 8. Therefore, there is a significant error between the user satisfaction ratings predicted by the regression model and the actual user ratings, which is not conducive to operators making improvements based on the prediction results. Summary of the Invention

[0005] This application provides a training method for a satisfaction prediction model, a satisfaction prediction method, and an apparatus to address the problem in the prior art where there is a large error between the predicted user satisfaction score and the actual user score, which is not conducive to operators making improvements based on the prediction results.

[0006] Firstly, this application provides a method for training a satisfaction prediction model, comprising:

[0007] Obtain multiple user data samples and the corresponding label information for each user data sample. The label information includes a score used to characterize user satisfaction.

[0008] Based on the first user data sample from multiple user data samples, update the parameters of the initial classification model to obtain the classification model. The label information of the first user data sample has a first score and a second score, with the first score being less than the second score.

[0009] Based on the second user data sample (excluding the first user data sample) from multiple user data samples, update the parameters of the initial regression model to obtain the regression model. The score in the label information of the second user data sample is the third score, which is greater than the first score and less than the second score.

[0010] Based on the classification model and the regression model, a satisfaction prediction model is determined.

[0011] Optionally, the initial classification model includes a first initial classification model and a second initial classification model.

[0012] Based on the first user data sample from multiple user data samples, update the parameters of the initial classification model to obtain the classification model, including:

[0013] A third user data sample is determined from the first user data sample, and the score in the label information of the third user data sample is the first score.

[0014] Based on the third user data sample, update the parameters of the first initial classification model to obtain the first classification model.

[0015] A fourth user data sample is determined from the first user data sample, and the score in the label information of the fourth user data sample is the second score.

[0016] Based on the fourth user data sample, update the parameters of the second initial classification model to obtain the second classification model.

[0017] Optionally, based on the first user data sample from multiple user data samples, the parameters of the initial classification model are updated to obtain the classification model, including:

[0018] Input the first user data sample into the initial classification model to obtain the first prediction result of user satisfaction.

[0019] The first loss information is determined based on the first prediction result and the label information corresponding to the first user data sample.

[0020] Based on the first loss information, the parameters of the initial classification model are updated to obtain the classification model.

[0021] Optionally, based on a second user data sample (excluding the first user data sample) from multiple user data samples, the parameters of the initial regression model are updated to obtain the regression model, including:

[0022] Input the second user data sample into the initial regression model to obtain the second prediction result of user satisfaction.

[0023] Based on the second prediction result and the label information corresponding to the second user data sample, the second loss information is determined, which includes the mean absolute error information.

[0024] Based on the second loss information, the parameters of the initial regression model are updated to obtain the regression model.

[0025] Optionally, obtain multiple user data samples, including:

[0026] Collect multiple initial user data samples.

[0027] The initial user data sample is preprocessed to obtain multiple user data samples. The preprocessing includes data cleaning and / or feature engineering.

[0028] Secondly, this application provides a method for predicting satisfaction levels, including:

[0029] Obtain user data to be predicted.

[0030] The user data to be predicted is input into the first model in the satisfaction prediction model to obtain the first prediction result. The first model includes a classification model or a regression model. The satisfaction prediction model is trained based on the training method of the satisfaction prediction model mentioned above.

[0031] Based on the first prediction result, determine the user satisfaction corresponding to the user data to be predicted.

[0032] Optionally, based on the first prediction result, determine the user satisfaction corresponding to the user data to be predicted, including:

[0033] Based on the first prediction result, determine whether to input the user data to be predicted into the second model in the satisfaction prediction model. The second model is different from the first model.

[0034] If it is determined that the user data to be predicted will be input into the second model, then user satisfaction will be determined based on the output of the second model.

[0035] If it is determined that the user data to be predicted will not be input into the second model, then user satisfaction will be determined based on the output of the first model.

[0036] Optionally, the first model may include a classification model.

[0037] Based on the first prediction result, determine whether to input the user data to be predicted into the second model in the satisfaction prediction model, including:

[0038] Determine whether the first prediction result is greater than the preset value.

[0039] If the first prediction result is not greater than the preset value, then the user data to be predicted will be input into the second model.

[0040] If the first prediction result is greater than the preset value, then the user data to be predicted will not be input into the second model.

[0041] Optionally, the classification model includes a first classification model and a second classification model.

[0042] Determining whether the first prediction result is greater than the preset value includes:

[0043] Determine whether the first prediction result corresponding to the first classification model is greater than the first preset value, and whether the first prediction result corresponding to the second classification model is greater than the second preset value.

[0044] If the first prediction result is not greater than the preset value, then the user data to be predicted will be input into the second model, including:

[0045] If the first prediction result corresponding to the first classification model is not greater than the first preset value, and the first prediction result corresponding to the second classification model is not greater than the second preset value, then it is determined that the user data to be predicted will be input into the second model.

[0046] If the first prediction result is not greater than the preset value, then the user data to be predicted will be input into the second model, including:

[0047] If the first prediction result corresponding to the first classification model is greater than the first preset value, or the first prediction result corresponding to the second classification model is greater than the second preset value, then it is determined that the user data to be predicted will not be input into the second model.

[0048] Optionally, the first model may include a regression model.

[0049] Based on the first prediction result, determine whether to input the user data to be predicted into the second model in the satisfaction prediction model, including:

[0050] Determine whether the first prediction result is a third score that is greater than the first score and less than the second score, where the first score is less than the second score.

[0051] If the first prediction result is not the third score, then the user data to be predicted will be input into the second model.

[0052] If the first prediction result is the third score, then it is determined that the user data to be predicted will not be input into the second model.

[0053] Thirdly, this application provides a training device for a satisfaction prediction model, comprising:

[0054] The acquisition module is used to acquire multiple user data samples and the corresponding label information for each user data sample. The label information includes a score used to characterize user satisfaction.

[0055] The update module is used to update the parameters of the initial classification model based on the first user data sample from multiple user data samples, and obtain the classification model. The label information of the first user data sample has a first score and a second score, and the first score is less than the second score.

[0056] The update module is also used to update the parameters of the initial regression model based on the second user data sample (excluding the first user data sample) from multiple user data samples, to obtain the regression model. The score in the label information of the second user data sample is the third score, which is greater than the first score and less than the second score.

[0057] The determination module is used to determine the satisfaction prediction model based on the classification model and the regression model.

[0058] Optionally, the initial classification model includes a first initial classification model and a second initial classification model.

[0059] The update module is specifically used for:

[0060] A third user data sample is determined from the first user data sample, and the score in the label information of the third user data sample is the first score.

[0061] Based on the third user data sample, update the parameters of the first initial classification model to obtain the first classification model.

[0062] A fourth user data sample is determined from the first user data sample, and the score in the label information of the fourth user data sample is the second score.

[0063] Based on the fourth user data sample, update the parameters of the second initial classification model to obtain the second classification model.

[0064] Optional, update module, specifically used for:

[0065] Input the first user data sample into the initial classification model to obtain the first prediction result of user satisfaction.

[0066] The first loss information is determined based on the first prediction result and the label information corresponding to the first user data sample.

[0067] Based on the first loss information, the parameters of the initial classification model are updated to obtain the classification model.

[0068] Optional, update module, specifically used for:

[0069] Input the second user data sample into the initial regression model to obtain the second prediction result of user satisfaction.

[0070] Based on the second prediction result and the label information corresponding to the second user data sample, the second loss information is determined, which includes the mean absolute error information.

[0071] Based on the second loss information, the parameters of the initial regression model are updated to obtain the regression model.

[0072] Optional, the acquisition module is specifically used for:

[0073] Collect multiple initial user data samples.

[0074] The initial user data sample is preprocessed to obtain multiple user data samples. The preprocessing includes data cleaning and / or feature engineering.

[0075] Fourthly, this application provides a satisfaction prediction device, comprising:

[0076] The acquisition module is used to acquire user data to be predicted.

[0077] The prediction module is used to input the user data to be predicted into the first model in the satisfaction prediction model to obtain the first prediction result. The first model includes a classification model or a regression model. The satisfaction prediction model is trained based on the training method of the satisfaction prediction model mentioned above.

[0078] The determination module is used to determine the user satisfaction corresponding to the user data to be predicted based on the first prediction result.

[0079] Optional, determine the module, specifically used for:

[0080] Based on the first prediction result, determine whether to input the user data to be predicted into the second model in the satisfaction prediction model. The second model is different from the first model.

[0081] If it is determined that the user data to be predicted will be input into the second model, then user satisfaction will be determined based on the output of the second model.

[0082] If it is determined that the user data to be predicted will not be input into the second model, then user satisfaction will be determined based on the output of the first model.

[0083] Optionally, the first model may include a classification model.

[0084] The module is specifically used for:

[0085] Determine whether the first prediction result is greater than the preset value.

[0086] If the first prediction result is not greater than the preset value, then the user data to be predicted will be input into the second model.

[0087] If the first prediction result is greater than the preset value, then the user data to be predicted will not be input into the second model.

[0088] Optionally, the classification model includes a first classification model and a second classification model.

[0089] The module is specifically used for:

[0090] Determine whether the first prediction result corresponding to the first classification model is greater than the first preset value, and whether the first prediction result corresponding to the second classification model is greater than the second preset value.

[0091] If the first prediction result corresponding to the first classification model is not greater than the first preset value, and the first prediction result corresponding to the second classification model is not greater than the second preset value, then it is determined that the user data to be predicted will be input into the second model.

[0092] If the first prediction result corresponding to the first classification model is greater than the first preset value, or the first prediction result corresponding to the second classification model is greater than the second preset value, then it is determined that the user data to be predicted will not be input into the second model.

[0093] Optionally, the first model may include a regression model.

[0094] The module is specifically used for:

[0095] Determine whether the first prediction result is a third score that is greater than the first score and less than the second score, where the first score is less than the second score.

[0096] If the first prediction result is not the third score, then the user data to be predicted will be input into the second model.

[0097] If the first prediction result is the third score, then it is determined that the user data to be predicted will not be input into the second model.

[0098] Fifthly, this application provides an electronic device, including: a memory and a processor;

[0099] Memory is used to store computer programs.

[0100] The processor is configured to read a computer program stored in the memory and execute the training method of the satisfaction prediction model described in the first aspect based on the computer program in the memory, or to execute the satisfaction prediction method described in the second aspect.

[0101] Sixthly, this application provides a readable storage medium having a computer program stored thereon, the computer program storing computer execution instructions, which, when executed by a processor, are used to implement the training method for the satisfaction prediction model as described in the first aspect above, or to implement the satisfaction prediction method as described in the second aspect above.

[0102] In a seventh aspect, embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements the training method for the satisfaction prediction model of the first aspect described above, or is used to implement the satisfaction prediction method of the second aspect described above.

[0103] The satisfaction prediction model training method, satisfaction prediction method, and apparatus provided in this application acquire multiple user data samples and corresponding label information for each user data sample. The label information includes scores used to characterize user satisfaction. Based on the first user data sample from the multiple user data samples, the parameters of the initial classification model are updated to obtain a classification model. The label information of the first user data sample contains a first score and a second score, with the first score being less than the second score. Based on the second user data sample from the multiple user data samples (excluding the first user data sample), the parameters of the initial regression model are updated to obtain a regression model. The label information of the second user data sample contains a third score, which is greater than the first score and less than the second score. Finally, the satisfaction prediction model is determined based on the classification model and the regression model. The classification model only needs to predict the scores at the boundary values, which avoids the problem that the classification model might ignore the progressive relationship of the scores. The regression model only needs to predict the scores in the middle interval, which avoids the error that occurs when the actual scores are mostly boundary values, while the predicted results are mostly the mean. By integrating classification and regression models, prediction errors can be effectively reduced, the accuracy of user satisfaction prediction can be improved, and the distribution of prediction results can be made closer to the distribution of real user ratings, so that operators can make further improvements based on the predicted satisfaction ratings. Attached Figure Description

[0104] Figure 1 A flowchart illustrating a training method for a satisfaction prediction model provided in an embodiment of this application;

[0105] Figure 2 A flowchart illustrating another training method for a satisfaction prediction model provided in this application embodiment;

[0106] Figure 3 A flowchart illustrating a satisfaction prediction method provided in an embodiment of this application;

[0107] Figure 4 A flowchart illustrating another satisfaction prediction method provided in an embodiment of this application;

[0108] Figure 5 A schematic diagram of the structure of a training device for a satisfaction prediction model provided in an embodiment of this application;

[0109] Figure 6 This is a schematic diagram of the structure of a satisfaction prediction device provided in an embodiment of this application;

[0110] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0111] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0112] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0113] The technical solution provided in this application can be used in scenarios where user satisfaction scores are predicted for users who have ported their numbers to other networks. Operators can use the prediction results to determine user satisfaction and follow up with users who have low satisfaction levels to improve their business capabilities, service attitude, and other aspects based on user feedback.

[0114] Current methods rely on multi-class or regression models for prediction. After data collection and cleaning, a multi-class or regression model is selected for training, allowing frontline staff to predict user satisfaction ratings based on the trained model. However, when using a multi-class model, the model treats different ratings as different categories, thus ignoring the hierarchical relationships between ratings. For example, training data shows that a user believes the operator's package pricing is unreasonable, the staff's attitude is poor, and the network speed is slow, predicting the user's satisfaction rating to be 5 points. Training data shows another user believes the operator's package pricing is unreasonable and the staff's attitude is poor, predicting the user's satisfaction rating to be 7 points. Training data shows a third user believes the operator's package pricing is unreasonable, predicting the user's satisfaction rating to be 9 points. User ratings of the operator increase with user satisfaction and experience. However, multi-classification models treat 5, 7, and 9 points as three independent categories, ignoring the increasing relationship between ratings and satisfaction. This is not conducive to operators using the increasing relationship of ratings to improve service levels and other aspects. When using regression models, the most significant portion of the predicted results is the mean of the ratings obtained from the predicted data. The distribution of such predicted results is almost always a normal distribution, which often does not match the actual distribution of user ratings, also hindering operators from using ratings to improve service levels and other aspects.

[0115] To address the above issues, this application proposes a method combining classification and regression models to predict user satisfaction scores. The classification model predicts scores at the upper and lower boundary values, such as 1 and 10; the regression model predicts scores in the middle range, such as the 2-9 range. In this method, the classification model only needs to predict the boundary values, avoiding the problem of ignoring the progressive relationship of scores; the regression model only needs to predict the middle range, avoiding the error that occurs when the actual scores are mostly boundary values ​​while the predicted results are mostly the mean. Therefore, this method can effectively reduce prediction errors, allowing operators to make further improvements based on the predicted satisfaction scores.

[0116] The technical solution of this application and how it solves the above-mentioned technical problems will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0117] Figure 1 This is a flowchart illustrating a training method for a satisfaction prediction model provided in an embodiment of this application. The training method can be executed by software and / or hardware devices, such as electronic devices like terminals or servers. For example, please refer to [link to example]. Figure 1 As shown, the training method for this satisfaction prediction model may include:

[0118] S101. Obtain multiple user data samples and the corresponding label information for each user data sample. The label information includes a score used to characterize user satisfaction.

[0119] In this step, the user data samples mainly include signaling data from the O domain, user information data from the B domain, perception data, scoring data, and complaint data.

[0120] The four types of data that can be used for model training include signaling data from the O domain, user information data from the B domain, perception data, and complaint data. The scoring data includes the satisfaction scores corresponding to the label information of each user data sample. The specific meanings of the five types of data include:

[0121] The signaling data in the O domain mainly includes information such as user login, calls, SMS, location movement, and terminal. Its main function is to reflect users' login habits, user stickiness, roaming characteristics, and terminal preferences.

[0122] B-domain user information data mainly includes basic user information, which includes the user's gender, age, package, account opening information, contracted services, etc. It is used to reflect the user's physical attributes and is generally static, not changing with the user's location.

[0123] Perception data mainly includes the user's internet uplink and downlink speeds, video buffering rate, network latency, and call drop rate under different networks (such as 2G, 3G, 4G, and 5G), which are used to intuitively reflect the user's perception of internet access or calls.

[0124] The complaint data mainly consists of complaints made by users in the past six months through customer service calls or mobile app. These complaints primarily cover aspects such as network perception, network coverage, package pricing, and customer service attitude. The main purpose is to indirectly reflect users' subjective evaluation and loyalty to the operator's network.

[0125] The rating data is the label information corresponding to each user data sample during model training. That is, the rating given by users to the questionnaires issued by the customer service department. The score is an integer from 1 to 10. The higher the score, the higher the satisfaction.

[0126] For example, when acquiring multiple user data samples, multiple initial user data samples can be collected and preprocessed to obtain multiple user data samples. The preprocessing includes data cleaning and / or feature engineering.

[0127] Specifically, data cleaning can be performed only on the initial user data sample, feature engineering can be performed only on the initial user data sample, or data cleaning can be performed on the initial user data sample first, followed by feature engineering.

[0128] The following example illustrates how initial user data samples are first cleaned and then feature-engineered.

[0129] Specifically, the collected data can be cleaned, which mainly addresses the following types of issues:

[0130] (1) Handling null values: For columns with a null value rate exceeding 60%, delete the data in that column directly. For ordinary null data, fill with zero values ​​or the mean value;

[0131] (2) Outlier handling, such as gender data, where data other than male and female are uniformly treated as unknown. Age data, such as removing records of ages less than 10 years or greater than 100 years;

[0132] (3) Category data processing: The category data is processed into numerical indicators or sparse matrices by using one-hot encoding or label encoding methods.

[0133] After data cleaning, feature engineering can be performed. Commonly used feature engineering methods include the following:

[0134] (1) Based on the feature correlation coefficient matrix, directly delete the features with high correlation coefficients and keep only one feature;

[0135] (2) Feature dimensionality reduction is performed based on feature dimensionality reduction methods such as Principal Component Analysis (PCA) or Latent Dirichlet Allocation (LDA). In simple terms, it is to generate new features by linear combination of multiple features through principal component analysis or independent component analysis.

[0136] (3) Features can also be constructed based on expert experience and used for model training.

[0137] In this embodiment, the purpose of data cleaning is to remove duplicate information, correct existing errors, and provide data consistency. Feature engineering aims to extract the maximum number of features from the raw data for use by the algorithm and model. By performing data cleaning and / or feature engineering on the collected data, the accuracy of the data can be improved, thereby improving the accuracy of model training.

[0138] S102. Based on the first user data sample among multiple user data samples, update the parameters of the initial classification model to obtain the classification model.

[0139] In the first user data sample, the tag information has a first score and a second score, with the first score being less than the second score.

[0140] In this step, the first score and the second score can be set according to the actual situation or experience. For example, the first score can be set to 1 point and the second score can be set to 10 points. The specific values ​​of the first score and the second score are not limited in this embodiment of the application.

[0141] For example, the above-mentioned initial classification model includes a first initial classification model and a second initial classification model. When updating the parameters of the initial classification model based on the first user data sample among multiple user data samples to obtain a classification model, a third user data sample may be determined from the first user data sample, wherein the score in the label information of the third user data sample is a first score, and the parameters of the first initial classification model are updated based on the third user data sample to obtain the first classification model; alternatively, a fourth user data sample may be determined from the first user data sample, wherein the score in the label information of the fourth user data sample is a second score, and the parameters of the second initial classification model are updated based on the fourth user data sample to obtain the second classification model.

[0142] Specifically, the score in the label information corresponding to the third user data sample is the first score. Based on the third user data sample, the parameters of the first initial classification model are updated to obtain the first classification model. The score in the label information corresponding to the fourth user data sample is the second score. Based on the fourth user data sample, the parameters of the second initial classification model are updated to obtain the second classification model.

[0143] For example, the trained first and second classification models can be named Classification Model A and Classification Model B, respectively. Classification Model A's task is to select users with a satisfaction rating of 1. To train Classification Model A, the ratings from 1 to 10 are first mapped to 0 and 1. Specifically, user data samples with a rating of 1 are labeled as 1, and other user data samples are labeled as 0. By continuously training the initial classification model A with these user data samples, we can obtain Classification Model A used to extract users with a satisfaction rating of 1. Similarly, Classification Model B's task is to select users with a satisfaction rating of 10. User data samples with a rating of 10 are labeled as 1, and other user data samples are labeled as 0. By continuously training the initial classification model B with these user data samples, we can obtain Classification Model B capable of extracting users with a satisfaction rating of 10.

[0144] Furthermore, when training the initial classification model, the first user data sample can be input into the initial classification model to obtain the first prediction result of user satisfaction. Based on the first prediction result and the label information corresponding to the first user data sample, the first loss information is determined. Then, based on the first loss information, the parameters of the initial classification model are updated to obtain the classification model.

[0145] Specifically, based on the first prediction result representing user satisfaction and the label information corresponding to the first user data sample (i.e., the pre-assigned score), the first loss information is determined. The parameters of the initial classification model are then updated based on this first loss information. This process is repeated until the classification model reaches convergence. The classification model that has reached convergence is then determined as the final classification model. This model training method can improve the accuracy of the satisfaction prediction model training.

[0146] For example, the cross-entropy function can be used as the loss function: This is used to determine the first loss information. The closer the predicted result is to the actual score, the smaller the cross-entropy will be.

[0147] S103. Based on the second user data sample (excluding the first user data sample) from multiple user data samples, update the parameters of the initial regression model to obtain the regression model.

[0148] Among them, the third score in the tag information of the second user data sample is greater than the first score but less than the second score.

[0149] In this step, the parameters of the initial regression model are updated based on the second user data sample with the third score, thus obtaining the regression model. The third score can be an interval score, and can be set according to actual conditions or experience, for example, it can be set to 2-9 points. This embodiment of the application does not limit the specific value of the third score.

[0150] For example, to train the initial regression model R, users with ratings of 1 and 10 are removed from the training dataset, leaving only user data samples with ratings of 2 to 9 as training samples. After undergoing the same training process as the previous classification model, a regression model R suitable for ratings of 2-9 can be obtained.

[0151] For example, when training the initial regression model, the second user data sample can be input into the initial regression model to obtain the second prediction result of user satisfaction. Based on the second prediction result and the label information corresponding to the second user data sample, the second loss information is determined. The second loss information includes the mean absolute error information. Then, based on the second loss information, the parameters of the initial regression model are updated to obtain the regression model.

[0152] Specifically, based on the second prediction result and the label information corresponding to the second user data sample, the second loss information is determined. The second loss information includes the mean absolute error information. Then, based on the second loss information, the parameters of the initial regression model are updated to obtain the regression model.

[0153] For example, compared to classification models, regression models use mean absolute error as their evaluation function: Since only the absolute error of the prediction result needs to be reduced when predicting satisfaction, mean absolute error is more suitable for the use case of satisfaction prediction than the commonly used mean square error.

[0154] S104. Determine the satisfaction prediction model based on the classification model and the regression model.

[0155] In this step, the trained classification model and regression model are fused to obtain the satisfaction prediction model.

[0156] The training method for the satisfaction prediction model provided in this application involves acquiring multiple user data samples and corresponding label information for each user data sample. The label information includes scores representing user satisfaction. Based on a first user data sample from the multiple user data samples, the parameters of an initial classification model are updated to obtain a classification model. The label information of the first user data sample contains a first score and a second score, where the first score is less than the second score. Based on a second user data sample (excluding the first user data sample) from the multiple user data samples, the parameters of an initial regression model are updated to obtain a regression model. The label information of the second user data sample contains a third score, which is greater than the first score and less than the second score. Based on the classification model and the regression model, a satisfaction prediction model is determined. The classification model only needs to predict the scores at boundary values, avoiding the problem of the classification model ignoring the progressive relationship of scores. The regression model only needs to predict the scores in the middle interval, avoiding the error that occurs when the actual scores are mostly boundary values ​​while the predicted results are mostly mean values. By integrating classification and regression models, prediction errors can be effectively reduced, the accuracy of user satisfaction prediction can be improved, and the distribution of prediction results can be made closer to the distribution of real user ratings, so that operators can make further improvements based on the predicted satisfaction ratings.

[0157] Figure 2 This is a flowchart illustrating another training method for a satisfaction prediction model provided in an embodiment of this application. Below, in conjunction with... Figure 2 The training process of the satisfaction prediction model will be explained in detail with specific examples:

[0158] First, multiple initial user data samples are collected as training datasets. Next, these initial user data samples are cleaned to obtain cleaned user data samples. Feature engineering is then performed on these cleaned user data samples to obtain further user data samples. Finally, users with a rating of 1 are labeled as 1, and all other users are labeled as 0. The initial classification model A is then continuously trained using the training dataset to obtain a classification model A used to extract satisfaction ratings of 1. Similarly, users with a rating of 10 are labeled as 1, and all other users are labeled as 0. The initial classification model B is then continuously trained using the training dataset to obtain a classification model B used to extract satisfaction ratings of 10. The user data samples with ratings of 1 and 10 are removed from the training dataset, leaving only user data samples with ratings of 2 to 9 as training samples. After undergoing the same training process as the previous classification models, a regression model R suitable for ratings of 2-9 is obtained. Through these steps, trained classification model A, classification model B, and regression model R are obtained.

[0159] The training method for the satisfaction prediction model provided in this application involves collecting initial user data samples, cleaning and feature-engineering these samples to obtain user data samples for model training, and then training a classification model and a regression model based on these user data samples. By fusing the classification and regression models, the accuracy of user satisfaction prediction can be improved, and the distribution of the prediction results can be made closer to the distribution of actual user ratings.

[0160] Figure 3 This is a flowchart illustrating a satisfaction prediction method provided in an embodiment of this application. The satisfaction prediction method can be executed by software and / or hardware devices, such as electronic devices like terminals or servers. For example, please refer to [link to example]. Figure 3 As shown, the satisfaction prediction method may include:

[0161] S301. Obtain user data to be predicted.

[0162] In this step, the user data to be predicted includes signaling data from the O domain, user information data from the B domain, perception data, and complaint data.

[0163] S302. Input the user data to be predicted into the first model in the satisfaction prediction model to obtain the first prediction result. The first model includes a classification model or a regression model.

[0164] Among them, the satisfaction prediction model is based on Figure 1 The method shown is used for training.

[0165] In this step, the user data to be predicted is input into the classification or regression model in the satisfaction prediction model to obtain the first prediction result.

[0166] S303. Based on the first prediction result, determine the user satisfaction corresponding to the user data to be predicted.

[0167] In this step, user satisfaction can be understood as user satisfaction rating.

[0168] For example, when determining the user satisfaction corresponding to the user data to be predicted based on the first prediction result, it is determined whether to input the user data to be predicted into the second model in the satisfaction prediction model based on the first prediction result. The second model is different from the first model. If it is determined that the user data to be predicted should be input into the second model, the user satisfaction is determined based on the output result of the second model. If it is determined that the user data to be predicted should not be input into the second model, the user satisfaction is determined based on the output result of the first model.

[0169] In one possible implementation, when the first model includes a classification model, when determining whether to input the user data to be predicted into the second model in the satisfaction prediction model, it can be determined whether the first prediction result is greater than a preset value. If the first prediction result is not greater than the preset value, then it is determined to input the user data to be predicted into the second model; if the first prediction result is greater than the preset value, then it is determined not to input the user data to be predicted into the second model.

[0170] The first prediction result can be a probability value. The preset value can be set according to the actual situation or experience, for example, it can be set to 0.86, etc. The specific value of the preset value is not limited in this embodiment of the application.

[0171] For example, it is determined whether the first prediction result corresponding to the first classification model is greater than the first preset value, and whether the first prediction result corresponding to the second classification model is greater than the second preset value. If the first prediction result corresponding to the first classification model is not greater than the first preset value, and the first prediction result corresponding to the second classification model is not greater than the second preset value, then it is determined that the user data to be predicted will be input into the second model; if the first prediction result corresponding to the first classification model is greater than the first preset value, or the first prediction result corresponding to the second classification model is greater than the second preset value, then it is determined that the user data to be predicted will not be input into the second model.

[0172] Specifically, when the first model is a classification model, the second model is a regression model. The decision to input the user data to be predicted into the second model of the satisfaction prediction model is made by determining whether the first prediction result corresponding to the first classification model is greater than a first preset value, and whether the first prediction result corresponding to the second classification model is greater than a second preset value. If the first prediction result corresponding to both the first and second classification models is not greater than the first preset value, then the user data to be predicted is input into the regression model, and the predicted user satisfaction score is determined based on the output of the regression model. If the first prediction result corresponding to either the first or second classification model is greater than the first preset value, then the user data to be predicted is not input into the regression model. In this case, the score in the label information of the first user data sample in the classification model corresponding to the first preset result is the predicted satisfaction score.

[0173] In this step, users with the first or second satisfaction prediction score can be excluded first. If the predicted satisfaction score is determined to be either the first or second score, the user data sample does not need to be input into the regression model, simplifying the satisfaction prediction process. If the predicted satisfaction score is not the first or second score, then the user data sample can be input into the regression model, improving the accuracy of the satisfaction prediction.

[0174] In another possible implementation, when the first model includes a regression model, in determining whether to input the user data to be predicted into the second model in the satisfaction prediction model, it can be determined whether the first prediction result is a third score that is greater than the first score and less than the second score, where the first score is less than the second score; if the first prediction result is not the third score, then it is determined that the user data to be predicted will be input into the second model; if the first prediction result is the third score, then it is determined that the user data to be predicted will not be input into the second model.

[0175] The first prediction result can be a value that is greater than the first score and less than the second score.

[0176] Specifically, when the first model includes a regression model, the second model is a classification model. It is determined whether the first prediction result is the third score. If the first prediction result is not the third score, it is determined that the user data to be predicted will be input into the classification model. If the first prediction result is the third score, it is determined that the user data to be predicted will not be input into the classification model. At this time, the value corresponding to the first preset result is rounded to obtain the predicted value of the satisfaction score.

[0177] In this step, users whose predicted satisfaction rating is the third score can be excluded first. If the predicted satisfaction rating is determined to be the third score, the user's data sample does not need to be input into the classification model, simplifying the satisfaction prediction process. If the predicted satisfaction rating is determined not to be the third score, then the user's data sample can be input into the classification model, improving the accuracy of the satisfaction prediction.

[0178] The satisfaction prediction method provided in this application obtains user data to be predicted, inputs the user data into a first model in a satisfaction prediction model, and obtains a first prediction result. The first model includes a classification model or a regression model. Based on the first prediction result, the user satisfaction corresponding to the user data to be predicted is determined. By fusing classification and regression models, the accuracy of user satisfaction prediction can be improved, and the distribution of the prediction results can be made closer to the distribution of actual user ratings. Furthermore, this solution details the specific process of model fusion, enabling frontline personnel to conveniently use this method in their work, allowing operators to make further improvements based on the predicted satisfaction ratings.

[0179] Figure 4 This is a flowchart illustrating another satisfaction prediction method provided in an embodiment of this application. Below, in conjunction with... Figure 4 The process of the satisfaction prediction method will be explained in detail with specific examples:

[0180] First, input multiple user data samples from the prediction dataset into classification model A to obtain the probability p of a user rating of 1. a When p a If the score exceeds a certain threshold n, the user's predicted satisfaction rating is set to 1. Then, multiple user data samples from the prediction dataset are input into the classification model B to obtain the probability p of the user's rating being 10. b When p b If the score exceeds a certain threshold m, the user's predicted satisfaction score is set to 10. Finally, multiple user data samples from the remaining unpredicted dataset are input into the regression model R to obtain the prediction result, which is then rounded to an integer and used as the user's predicted satisfaction score S.

[0181] Furthermore, to find the optimal threshold, a gridded search method is used, aiming to minimize the overall mean absolute error. After searching, the optimal result for the two threshold settings can be obtained. For example, the optimal result can be m = 0.55 and n = 0.3.

[0182] The satisfaction prediction method provided in this application involves inputting multiple user data samples from the prediction dataset into classification model A, classification model B, and regression model R to obtain predicted user satisfaction scores. By fusing the classification and regression models, the accuracy of user satisfaction prediction can be improved, and the distribution of the prediction results can be made closer to the distribution of actual user scores. Furthermore, this solution allows operators to easily improve services based on the predicted satisfaction scores.

[0183] Figure 5 This application provides a schematic diagram of the structure of a training device 50 for a satisfaction prediction model, as illustrated in the embodiments of this application. For an example, please refer to [link to example diagram]. Figure 5 As shown, the training device 50 for the satisfaction prediction model includes:

[0184] The acquisition module 501 is used to acquire multiple user data samples and the corresponding label information for each user data sample. The label information includes a score used to characterize user satisfaction.

[0185] The update module 502 is used to update the parameters of the initial classification model based on the first user data sample among multiple user data samples, so as to obtain the classification model. The label information of the first user data sample has a first score and a second score, and the first score is less than the second score.

[0186] The update module 502 is also used to update the parameters of the initial regression model based on the second user data sample other than the first user data sample from multiple user data samples, to obtain the regression model. The score in the label information of the second user data sample is the third score, which is greater than the first score and less than the second score.

[0187] Module 503 is used to determine the satisfaction prediction model based on the classification model and the regression model.

[0188] Optionally, the initial classification model includes a first initial classification model and a second initial classification model.

[0189] Update module 502, specifically used for:

[0190] A third user data sample is determined from the first user data sample, and the score in the label information of the third user data sample is the first score.

[0191] Based on the third user data sample, update the parameters of the first initial classification model to obtain the first classification model.

[0192] A fourth user data sample is determined from the first user data sample, and the score in the label information of the fourth user data sample is the second score.

[0193] Based on the fourth user data sample, update the parameters of the second initial classification model to obtain the second classification model.

[0194] Optional, update module 502, specifically for:

[0195] Input the first user data sample into the initial classification model to obtain the first prediction result of user satisfaction.

[0196] The first loss information is determined based on the first prediction result and the label information corresponding to the first user data sample.

[0197] Based on the first loss information, the parameters of the initial classification model are updated to obtain the classification model.

[0198] Optional, update module 502, specifically for:

[0199] Input the second user data sample into the initial regression model to obtain the second prediction result of user satisfaction.

[0200] Based on the second prediction result and the label information corresponding to the second user data sample, the second loss information is determined, which includes the mean absolute error information.

[0201] Based on the second loss information, the parameters of the initial regression model are updated to obtain the regression model.

[0202] Optionally, module 501 is used for:

[0203] Collect multiple initial user data samples.

[0204] The initial user data sample is preprocessed to obtain multiple user data samples. The preprocessing includes data cleaning and / or feature engineering.

[0205] The training device 50 for the satisfaction prediction model shown in this application embodiment can execute the technical solution of the training method for the satisfaction prediction model in the above embodiment. Its implementation principle and beneficial effects are similar to those of the training method for the satisfaction prediction model. Please refer to the implementation principle and beneficial effects of the training method for the satisfaction prediction model. It will not be repeated here.

[0206] Figure 6 This is a schematic diagram of the structure of a satisfaction prediction device 60 provided in an embodiment of this application. For example, please refer to [link to example diagram]. Figure 6 As shown, the satisfaction prediction device 60 includes:

[0207] The acquisition module 601 is used to acquire user data to be predicted.

[0208] Prediction module 602 is used to input the user data to be predicted into the first model of the satisfaction prediction model to obtain the first prediction result. The first model includes a classification model or a regression model. The satisfaction prediction model is based on... Figure 1 The method shown is used for training.

[0209] The determination module 603 is used to determine the user satisfaction corresponding to the user data to be predicted based on the first prediction result.

[0210] Optionally, module 603 is determined, specifically for:

[0211] Based on the first prediction result, determine whether to input the user data to be predicted into the second model in the satisfaction prediction model. The second model is different from the first model.

[0212] If it is determined that the user data to be predicted will be input into the second model, then user satisfaction will be determined based on the output of the second model.

[0213] If it is determined that the user data to be predicted will not be input into the second model, then user satisfaction will be determined based on the output of the first model.

[0214] Optionally, the first model may include a classification model.

[0215] Module 603 is specifically used for:

[0216] Determine whether the first prediction result is greater than the preset value.

[0217] If the first prediction result is not greater than the preset value, then the user data to be predicted will be input into the second model.

[0218] If the first prediction result is greater than the preset value, then the user data to be predicted will not be input into the second model.

[0219] Optionally, the classification model includes a first classification model and a second classification model.

[0220] Module 603 is specifically used for:

[0221] Determine whether the first prediction result corresponding to the first classification model is greater than the first preset value, and whether the first prediction result corresponding to the second classification model is greater than the second preset value.

[0222] If the first prediction result corresponding to the first classification model is not greater than the first preset value, and the first prediction result corresponding to the second classification model is not greater than the second preset value, then it is determined that the user data to be predicted will be input into the second model.

[0223] If the first prediction result corresponding to the first classification model is greater than the first preset value, or the first prediction result corresponding to the second classification model is greater than the second preset value, then it is determined that the user data to be predicted will not be input into the second model.

[0224] Optionally, the first model may include a regression model.

[0225] Module 603 is specifically used for:

[0226] Determine whether the first prediction result is a third score that is greater than the first score and less than the second score, where the first score is less than the second score.

[0227] If the first prediction result is not the third score, then the user data to be predicted will be input into the second model.

[0228] If the first prediction result is the third score, then it is determined that the user data to be predicted will not be input into the second model.

[0229] The satisfaction prediction device 60 shown in this application embodiment can execute the technical solution of the satisfaction prediction method in the above embodiment. Its implementation principle and beneficial effects are similar to those of the satisfaction prediction method. Please refer to the implementation principle and beneficial effects of the satisfaction prediction method. It will not be repeated here.

[0230] Figure 7 This is a schematic diagram of the structure of an electronic device 70 provided in an embodiment of this application. For example, please refer to [link / reference]. Figure 7 As shown, the electronic device 70 may include a processor 701 and a memory 702; wherein,

[0231] Memory 702 is used to store computer programs.

[0232] The processor 701 is configured to read the computer program stored in the memory 702 and execute the training method of the satisfaction prediction model in the above embodiments according to the computer program in the memory 702, or execute the satisfaction prediction method in the above embodiments.

[0233] Optionally, the memory 702 can be either standalone or integrated with the processor 701. When the memory 702 is a device independent of the processor 701, the electronic device 70 may further include a bus for connecting the memory 702 and the processor 701.

[0234] Optionally, this embodiment further includes a communication interface, which can be connected to the processor 701 via a bus. The processor 701 can control the communication interface to realize the acquisition and transmission functions of the aforementioned electronic device 70.

[0235] For example, in this embodiment of the application, the electronic device 70 can be a terminal or a server, which can be set according to actual needs.

[0236] The electronic device 70 shown in this application embodiment can execute the technical solution of the training method of the satisfaction prediction model in the above embodiment. Its implementation principle and beneficial effects are similar to those of the training method of the satisfaction prediction model. Please refer to the implementation principle and beneficial effects of the training method of the satisfaction prediction model. Alternatively, it can execute the technical solution of the satisfaction prediction method in the above embodiment. Its implementation principle and beneficial effects are similar to those of the satisfaction prediction method. Please refer to the implementation principle and beneficial effects of the satisfaction prediction method. It will not be repeated here.

[0237] This application also provides a computer-readable storage medium storing computer-executable instructions. When a processor executes the computer-executable instructions, it implements the technical solution of the training method for the satisfaction prediction model in the above embodiments. The implementation principle and beneficial effects are similar to those of the training method for the satisfaction prediction model, and can be found in the implementation principle and beneficial effects of the training method for the satisfaction prediction model. Alternatively, the technical solution for implementing the satisfaction prediction method in the above embodiments has a similar implementation principle and beneficial effects to those of the satisfaction prediction method, and can be found in the implementation principle and beneficial effects of the satisfaction prediction method, which will not be repeated here.

[0238] This application also provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the technical solution of the training method for the satisfaction prediction model in the above embodiments. Its implementation principle and beneficial effects are similar to those of the training method for the satisfaction prediction model. Please refer to the implementation principle and beneficial effects of the training method for the satisfaction prediction model. Alternatively, the technical solution for implementing the satisfaction prediction method in the above embodiments has a similar implementation principle and beneficial effects to those of the satisfaction prediction method. Please refer to the implementation principle and beneficial effects of the satisfaction prediction method. It will not be repeated here.

[0239] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0240] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or in a combination of hardware and software functional units.

[0241] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods of the various embodiments of this application.

[0242] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules within the processor.

[0243] The memory may include high-speed random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device, and may also be a USB flash drive, external hard drive, read-only memory, disk or optical disc, etc.

[0244] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0245] The aforementioned computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The storage medium can be any available medium accessible to general-purpose or special-purpose computers.

[0246] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A training method for a satisfaction prediction model, characterized in that, The method comprises: obtaining a plurality of user data samples and label information corresponding to each of the user data samples, the label information comprising a score representing user satisfaction; updating parameters of an initial classification model according to a first user data sample in the plurality of user data samples to obtain a classification model, the label information of the first user data sample comprising a first score and a second score, the first score being less than the second score; updating parameters of an initial regression model according to a second user data sample in the plurality of user data samples other than the first user data sample to obtain a regression model, the label information of the second user data sample comprising a third score, the third score being greater than the first score and less than the second score; determining a satisfaction prediction model according to the classification model and the regression model; wherein the initial classification model comprises a first initial classification model and a second initial classification model; the updating of the parameters of the initial classification model according to the first user data sample in the plurality of user data samples to obtain a classification model comprises: determining a third user data sample from the first user data sample, the label information of the third user data sample comprising the first score; the first score being 1; mapping scores from 1 to 10 to 0 and 1; labeling the third user data sample with a score of 1 as 1, and labeling the remaining data samples in the first user data sample other than the third user data sample as 0; updating the parameters of the first initial classification model according to the labeled third user data sample to obtain a first classification model; and determining a fourth user data sample from the first user data sample, the label information of the fourth user data sample comprising the second score; the second score being 10; mapping scores from 1 to 10 to 0 and 1; labeling the fourth user data sample with a score of 10 as 1, and labeling the remaining data samples in the first user data sample other than the fourth user data sample as 0; updating the parameters of the second initial classification model according to the labeled fourth user data sample to obtain a second classification model.

2. The method of claim 1, wherein, the updating of the parameters of the initial classification model according to the first user data sample in the plurality of user data samples to obtain a classification model comprises: inputting the first user data sample into the initial classification model to obtain a first prediction result of user satisfaction; determining first loss information according to the first prediction result and the label information corresponding to the first user data sample; updating the parameters of the initial classification model according to the first loss information to obtain a classification model.

3. The method according to claim 1 or 2, characterized in that, the updating of the parameters of the initial regression model according to the second user data sample in the plurality of user data samples other than the first user data sample to obtain a regression model comprises: inputting the second user data sample into the initial regression model to obtain a second prediction result of user satisfaction; determine second loss information according to the second prediction result and label information corresponding to the second user data sample, the second loss information comprising mean absolute error information; update parameters of the initial regression model according to the second loss information to obtain a regression model.

4. The method according to claim 1 or 2, characterized in that, The obtaining of the plurality of user data samples comprises: collecting a plurality of initial user data samples; performing preprocessing on the initial user data samples to obtain the plurality of user data samples, the preprocessing comprising data cleaning processing and / or feature engineering processing.

5. A satisfaction prediction method characterized by, comprises: obtaining to-be-predicted user data; inputting the to-be-predicted user data into a first model in a satisfaction prediction model to obtain a first prediction result, the first model comprising a classification model or a regression model, the satisfaction prediction model being trained based on the method in any one of claims 1-4; determining user satisfaction corresponding to the to-be-predicted user data according to the first prediction result.

6. The method of claim 5, wherein, The determining of the user satisfaction corresponding to the to-be-predicted user data according to the first prediction result comprises: determining whether to input the to-be-predicted user data into a second model in the satisfaction prediction model according to the first prediction result, the second model being different from the first model; if it is determined to input the to-be-predicted user data into the second model, determining the user satisfaction according to an output result of the second model; if it is determined not to input the to-be-predicted user data into the second model, determining the user satisfaction according to an output result of the first model.

7. The method of claim 6, wherein, The first model comprises a classification model. The determining of whether to input the to-be-predicted user data into the second model in the satisfaction prediction model according to the first prediction result comprises: determining whether the first prediction result is greater than a preset value; if the first prediction result is not greater than the preset value, determining to input the to-be-predicted user data into the second model; if the first prediction result is greater than the preset value, determining not to input the to-be-predicted user data into the second model.

8. The method of claim 7, wherein, The classification model comprises a first classification model and a second classification model. The determining of whether the first prediction result is greater than a preset value comprises: respectively determining whether a first prediction result corresponding to the first classification model is greater than a first preset value and whether a first prediction result corresponding to the second classification model is greater than a second preset value; if the first prediction result corresponding to the first classification model is not greater than the first preset value and the first prediction result corresponding to the second classification model is not greater than the second preset value, determining to input the to-be-predicted user data into the second model; if the first prediction result corresponding to the first classification model is greater than the first preset value or the first prediction result corresponding to the second classification model is greater than the second preset value, determining not to input the to-be-predicted user data into the second model. ​ ​ 9. The method of claim 6, wherein, The first model includes a regression model; The step of determining whether to input the user data to be predicted into the second model of the satisfaction prediction model based on the first prediction result includes: Determine whether the first prediction result is a third score that is greater than the first score and less than the second score, wherein the first score is less than the second score; If the first prediction result is not the third score, then the user data to be predicted will be input into the second model. If the first prediction result is the third score, then it is determined that the user data to be predicted will not be input into the second model. 10.A satisfaction prediction model training apparatus, characterized by comprising: include: The acquisition module is used to acquire multiple user data samples and label information corresponding to each user data sample, wherein the label information includes a score used to characterize user satisfaction. The update module is used to update the parameters of the initial classification model based on the first user data sample among the plurality of user data samples to obtain the classification model, wherein the label information of the first user data sample has a first score and a second score, and the first score is less than the second score; The update module is also used to update the parameters of the initial regression model based on the second user data sample other than the first user data sample among the plurality of user data samples, to obtain a regression model, wherein the score in the label information of the second user data sample is a third score, the third score is greater than the first score and less than the second score; The determination module is used to determine the satisfaction prediction model based on the classification model and the regression model; The initial classification model includes a first initial classification model and a second initial classification model; The update module is specifically configured to determine a third user data sample from the first user data sample, wherein the label information of the third user data sample contains a score equal to the first score; the first score is 1 point; map scores from 1 to 10 to 0 and 1; mark the third user data sample with a score of 1 as 1, and mark the remaining data samples in the first user data sample other than the third user data sample as 0; update the parameters of the first initial classification model based on the marked third user data sample to obtain a first classification model; and, A fourth user data sample is determined from the first user data sample, and the score in the label information of the fourth user data sample is the second score; the second score is 10 points; the scores from 1 to 10 points are mapped to 0 and 1; the fourth user data sample with a score of 10 points is marked as 1, and the remaining data samples in the first user data sample other than the fourth user data sample are marked as 0; based on the marked fourth user data sample, the parameters of the second initial classification model are updated to obtain the second classification model.

11. A satisfaction prediction device characterized by comprising: include: The acquisition module is used to acquire user data to be predicted. a prediction module, configured to input the to-be-predicted user data into a first model in a satisfaction prediction model to obtain a first prediction result, the first model comprising a classification model or a regression model, and the satisfaction prediction model being trained based on the method in any one of claims 1-4; a determination module, configured to determine a user satisfaction corresponding to the to-be-predicted user data according to the first prediction result.

12. An electronic device, comprising: comprising: a memory and a processor; the memory, configured to store a computer program; the processor, configured to read the computer program stored in the memory, and execute the training method of the satisfaction prediction model in any one of claims 1-4 according to the computer program in the memory, or execute the satisfaction prediction method in any one of claims 5-9.

13. A readable storage medium, having stored thereon a computer program, characterized in that, the computer program, configured to store computer execution instructions, and the computer execution instructions are configured to be executed by the processor to implement the training method of the satisfaction prediction model in any one of claims 1-4, or execute the satisfaction prediction method in any one of claims 5-9.

14. A computer program product, comprising a computer program, and the computer program is configured to be executed by a processor to implement the training method of the satisfaction prediction model in any one of claims 1-4, or execute the satisfaction prediction method in any one of claims 5-9.

Images