Complaint prediction method and model training method and device

Abstract

The present disclosure relates to the technical field of artificial intelligence, and particularly relates to a complaint prediction method and model training method and device. The complaint prediction method comprises: obtaining a network index data set corresponding to any user identifier in a full user identifier, the network index data set comprising a first network index data subset corresponding to a network node device and / or a second network index data subset corresponding to a communication medium; preprocessing the network index data set to obtain a processed network index data set; inputting the processed network index data set into a first prediction model for prediction to obtain a prediction result, wherein the prediction result is used to indicate a complaint probability of the any user identifier within a preset number of days for network quality. The present disclosure can provide a complaint prediction mechanism and improve the accuracy of complaint prediction.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of artificial intelligence, and particularly relates to a complaint prediction method and model training method and device. BACKGROUND

[0002] With the continuous progress of society, the use of the network continuously facilitates the life of the user. When the network quality does not meet the satisfaction of the user, the user can make a complaint about the network. For example, when the user feels that the time length of sending a message is greater than 5 seconds, the user can determine that the network quality is poor, and can make a complaint about the network quality. For example, whether the user makes a complaint about the network quality can be predicted based on the fault complaint information of the target device. However, in the prediction system, only the fault state information of the target device is considered, so that the accuracy of the complaint prediction is poor. SUMMARY

[0003] The present disclosure provides a complaint prediction method, a model training method and device to provide a complaint prediction mechanism and improve the accuracy of complaint prediction. The technical solutions of the present disclosure are as follows:

[0004] According to a first aspect of an embodiment of the present disclosure, a model training method is provided, comprising:

[0005] According to the feature extraction rule, the first data set is subjected to feature extraction to obtain a second data set, wherein the first data set comprises at least one first network quality data subset corresponding to a network complaint user identifier and at least one second network quality data subset corresponding to a network non-complaint user identifier;

[0006] According to the parameter set corresponding to the obtained second prediction model, the parameters in the second prediction model are set to obtain a third prediction model;

[0007] The second data set is used to train the third prediction model to obtain a first prediction model.

[0008] According to some embodiments, the first data set is subjected to feature extraction according to the feature extraction rule to obtain a second data set, comprising:

[0009] The first data set is subjected to missing value detection, and data with missing values in the first data set is removed to obtain a third data set;

[0010] The third data set is subjected to feature extraction according to the feature extraction rule to obtain a second data set;

[0011] According to the data set ratio, a training data subset and a test data subset in the second data set are obtained.

[0012] According to some embodiments, wherein the third prediction model is an XGBoost model, training the third prediction model using the second data set to obtain the first prediction model comprises:

[0013] Using the second data set, obtaining a model error corresponding to the XGBoost model;

[0014] Using the second data set, obtaining a structure error of the XGBoost model;

[0015] According to the model error and the structure error, obtaining a target function corresponding to the XGBoost model, and obtaining the XGBoost model trained.

[0016] According to some embodiments, the method further comprises:

[0017] Obtaining an information gain change value corresponding to any XGBoost node in the XGBoost model;

[0018] When the information gain change value is greater than a first threshold value, split the any XGBoost node;

[0019] When the information gain change value is less than or equal to a second threshold value, do not split the any XGBoost node, wherein the first threshold value is greater than or equal to the second threshold value.

[0020] According to some embodiments, the training the third prediction model using the second data set to obtain the first prediction model comprises:

[0021] According to the size of the second data set, obtaining a search method corresponding to the second data set;

[0022] Using the search method and the second data set to train the third prediction model, and obtaining the first prediction model when the prediction result of the third prediction model meets the prediction requirement.

[0023] According to some embodiments, the obtaining the first prediction model when the prediction result of the third prediction model meets the prediction requirement comprises:

[0024] Obtaining a predicted complaint result corresponding to the third prediction model;

[0025] According to the complaint result corresponding to the first data set and the predicted complaint result, obtaining an accuracy, precision and recall rate corresponding to the prediction result;

[0026] According to the precision rate, the first weight corresponding to the precision rate, the recall rate, and the second weight corresponding to the recall rate, a score corresponding to the third prediction model is obtained;

[0027] In a case where the accuracy rate, the precision rate, the recall rate, and the score meet a prediction requirement, it is determined that the first prediction model is obtained.

[0028] According to a second aspect of an embodiment of the present disclosure, a complaint prediction method is provided, and the method comprises:

[0029] A network index data set corresponding to any user identifier in a full-amount user identifier is obtained, and the network index data set comprises a first network index data subset corresponding to a network node device and / or a second network index data subset corresponding to a communication medium;

[0030] The network index data set is preprocessed to obtain a processed network index data set;

[0031] The processed network index data set is input into a first prediction model for prediction to obtain a prediction result, wherein the prediction result is used to indicate a complaint probability of the any user identifier within a preset number of days with respect to network quality, and the first prediction model is obtained based on the model training method according to any one of the first aspect.

[0032] According to some embodiments, the method further comprises:

[0033] A complaint result of the any user identifier within the preset number of days is obtained;

[0034] According to the prediction result and the complaint result, the first prediction model is adjusted to obtain an adjusted first prediction model.

[0035] According to a third aspect of an embodiment of the present disclosure, a model training apparatus is provided, and the apparatus comprises:

[0036] A feature extraction unit is configured to perform feature extraction on a first data set according to a feature extraction rule to obtain a second data set, wherein the first data set comprises a first network quality data subset corresponding to at least one network complaint user identifier and a second network quality data subset corresponding to at least one network non-complaint user identifier;

[0037] A parameter setting unit is configured to set parameters in the second prediction model according to a parameter set corresponding to the obtained second prediction model to obtain a third prediction model;

[0038] A model obtaining unit is configured to train the third prediction model by using the second data set to obtain a first prediction model.

[0039] According to a fourth aspect of the embodiments of the present disclosure, a device is provided, and the device comprises:

[0040] a data acquisition unit, configured to acquire a network index data set corresponding to any user identity in the full quantity of user identities, the network index data set comprising a first network index data subset corresponding to a network node device and / or a second network index data subset corresponding to a communication medium;

[0041] a data processing unit, configured to pre-process the network index data set to obtain a processed network index data set;

[0042] a complaint prediction unit, configured to input the processed network index data set into a first prediction model for prediction to obtain a prediction result, wherein the prediction result is used to indicate a complaint probability of the any user identity within a preset number of days for network quality, and the first prediction model is obtained based on the model training method in any one of the first aspect.

[0043] According to a fifth aspect of the embodiments of the present disclosure, a network side device is provided, and the network side device comprises:

[0044] a processor;

[0045] a memory for storing instructions executable by the processor;

[0046] The processor is configured to execute the instructions to implement the complaint prediction method in any one of the preceding aspects and the model training method in any one of the preceding aspects.

[0047] According to a sixth aspect of the embodiments of the present disclosure, a storage medium is provided, and when instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the complaint prediction method in any one of the preceding aspects and the model training method in any one of the preceding aspects.

[0048] According to a seventh aspect of the embodiments of the present disclosure, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the method in any one of the preceding aspects.

[0049] The technical solutions provided by the embodiments of the present disclosure at least bring the following beneficial effects:

[0050] In some or related embodiments, by acquiring a network index data set corresponding to any of the full quantity of user identifiers, the network index data set includes a first network index data subset corresponding to a network node device and / or a second network index data subset corresponding to a communication medium; preprocessing the network index data set to obtain a processed network index data set; inputting the processed network index data set into a first prediction model for prediction to obtain a prediction result, wherein the prediction result is used to indicate the complaint probability of the any user identifier within a preset number of days for network quality. Therefore, a complaint prediction mechanism can be provided, relevant network index data can be selected layer by layer according to the online service process, the situation that the complaint prediction is inaccurate due to the determination of the complaint prediction based on only the fault of the target device can be reduced, the accuracy of the prediction result within the preset number of days can be improved, and the accuracy of the complaint prediction can be improved.

[0051] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, and are not limiting to the present disclosure. BRIEF DESCRIPTION OF DRAWINGS

[0052] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments consistent with the present disclosure and serve to explain the principles of the present disclosure, and do not limit the present disclosure.

[0053] Figure 1 is a flowchart of a model training method according to an exemplary embodiment;

[0054] Figure 2 is an exemplary schematic diagram of a network topology according to an exemplary embodiment;

[0055] Figure 3 is an exemplary schematic diagram of a complaint prediction system according to an exemplary embodiment;

[0056] Figure 4 is an exemplary schematic diagram of a model training and complaint prediction method according to an exemplary embodiment;

[0057] Figure 5 is a flowchart of a complaint prediction method according to an exemplary embodiment;

[0058] Figure 6 is a block diagram of a model training device according to an exemplary embodiment;

[0059] Figure 7 is a block diagram of a complaint prediction device according to an exemplary embodiment;

[0060] Figure 8is a block diagram of a network-side device according to an exemplary embodiment. DETAILED DESCRIPTION

[0061] In order for those skilled in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the drawings.

[0062] The present disclosure embodiments propose a complaint prediction method, a model training method and device. In some embodiments, the complaint prediction method or the model training method can be mutually replaced with the information processing method, the communication method and other terms, the complaint prediction device or the model training device can be mutually replaced with the information processing device, the communication device and other terms, and the information processing system, the communication system and other terms can be mutually replaced.

[0063] The present disclosure embodiments are not exhaustive, but are only a part of the embodiments, and are not specific limitations on the protection scope of the present disclosure. In the case of no contradiction, each step in an embodiment can be implemented as an independent embodiment, and the steps can be combined arbitrarily, for example, the scheme after removing part of the steps in an embodiment can also be implemented as an independent embodiment, and the order of the steps in an embodiment can be exchanged arbitrarily, in addition, the optional implementation manners in an embodiment can be combined arbitrarily; in addition, the embodiments can be combined arbitrarily, for example, part or all of the steps of different embodiments can be combined arbitrarily, and an embodiment can be combined with the optional implementation manners of other embodiments.

[0064] In the present disclosure embodiments, the terms and / or descriptions between the embodiments are consistent if there is no special description and logical conflict, and can be mutually referred to, and the technical features in different embodiments can be combined to form new embodiments according to their inherent logical relationship.

[0065] The terms used in the present disclosure embodiments are only for the purpose of describing specific embodiments, and not as a limitation on the present disclosure.

[0066] In some embodiments, the acquisition of data, information and the like can comply with the laws and regulations of the place.

[0067] It should be noted that the terms "first", "second", and the like in the specification and claims of the present disclosure and the above-described drawings are used to distinguish similar objects, and do not necessarily have to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The implementation described in the following exemplary embodiments does not represent all implementations consistent with the present disclosure. Rather, they are only examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0068] Figure 1 is a flowchart of a model training method according to an exemplary embodiment, as Figure 1 shown, the model training method can be used in a communication scenario, including the following steps:

[0069] In step S11, feature extraction is performed on the first data set according to a feature extraction rule to obtain a second data set, wherein the first data set includes at least one first network quality data subset corresponding to a network complaint user identifier and at least one second network quality data subset corresponding to a network non-complaint user identifier;

[0070] According to some embodiments, the feature extraction rule may, for example, be a preset rule, which may, for example, be a rule for extracting data in the first data set, wherein the extraction rule for each data type is different. For example, the feature extraction rule may, for example, be determined by the properties corresponding to each data type. Different data types correspond to different feature extraction rules.

[0071] In some embodiments, the first data set may, for example, be a data set for training a prediction model. The first data set does not refer to a fixed data set. For example, when at least one network complaint user identifier or at least one network non-complaint user identifier changes, the first data set can also change accordingly. For example, when the first network quality data subset or the second network quality data subset changes, the first data set can also change accordingly.

[0072] In some embodiments, the first network quality data subset refers to a data set corresponding to at least one network complaint user identifier. The first network quality data subset does not refer to a fixed data set, for example, when at least one network complaint user identifier changes, the first network quality data subset can also change accordingly. For example, when any data in the first network quality data subset changes, the first network quality data subset can also change accordingly.

[0073] According to some embodiments, the feature extraction of the first data set according to the feature extraction rule to obtain the second data set includes: performing missing value detection on the first data set, removing data with missing values in the first data set to obtain a third data set; performing feature extraction on the third data set using the feature extraction rule to obtain the second data set; and obtaining a training data subset and a test data subset in the second data set according to a data set ratio. Therefore, the missing value processing of the first data set can improve the accuracy of the second data set, improve the accuracy of the model, and further improve the accuracy of the complaint prediction.

[0074] In some embodiments, for example, when there is a missing value in the network quality data corresponding to any user identifier in the first data set, the network quality data corresponding to the any user identifier is removed.

[0075] According to some embodiments, the data set after removing the missing values, i.e., the third data set, is divided into a training set and a test set in proportion, for example, the training set and the test set can be set as 7:3. The training data subset is used to iterate the second prediction model , and the test data subset is used to evaluate the performance of the first prediction model .

[0076] According to some embodiments, the original user network indicator data and the complaint data can be obtained by a data acquisition unit, and the specific data acquisition unit includes at least one of an Electric Operation Maintenance System (EOMS) platform, an Operation Support Systems (OSS) network management system, deep packet inspection (DPI), and a soft probe. For example, the user identifier of the complaint can be obtained every month. Since the number of complaint user identifiers generated every month is limited, the sample size is limited, and the user data can be obtained from the data acquisition unit regularly to update the first prediction model iteratively to improve the robustness of the prediction model. In an experimental environment, the first prediction model can be updated iteratively every month, for example. For example, the user identifier of the complaint due to network quality in the month is extracted regularly every month, and the number of complaint user identifiers is set as , and a certain proportion of non-complaint user identifiers are randomly extracted, and the number of non-complaint users is set as . The complaint ratio of the number of complaint user identifiers to the number of non-complaint user identifiers can be the real complaint ratio, which can be 1:200, for example. Since the user complaint behavior has a lagging nature relative to the fluctuation of the network indicator, a "historical data window" is introduced in the feature engineering, i.e., the network indicator data of the user n days before the complaint is extracted. For example, , for example, the network indicator data of a user identifier from January 1 to January 7 can be obtained, the features of the user are extracted, and whether the user will have a complaint behavior on January 8 is predicted.

[0077] ​The user identifier can be marked, for example, by performing complaint reason field screening on the complaint work order data, wherein the user who complains due to network reasons can be marked as 0, and the user who does not complain can be marked as 1. Therefore, the complaint user identifier and the complaint user identifier can be obtained, and the first network quality data subset corresponding to the complaint user identifier is obtained.

[0078] The data (indicators) included in the first data set can be selected based on network topology, use scenario, and network quality characteristics, for example.

[0079] For example, 485 indicators including the traffic overrun proportion of the optical line terminal (OLT) uplink link, video large packet download rate, Hypertext Transfer Protocol (HTTP) large packet download rate, HTTP response success rate, whether the optical network unit ONU weak light, packet loss rate, transmission control protocol TCP handshake total delay, and the like can be selected, and specifically, for example, as shown in Table 1. Therefore, the coverage of the network situation can be improved, the accuracy of the first data set acquisition can be improved, and the accuracy of the prediction model acquisition can be improved.

[0080] First, the network topology dimension. Figure 2 is an example schematic diagram of a network topology according to an example embodiment. The diagram can also be a network structure diagram of network nodes and communication media, for example. The network topology diagram includes a network layer, an OLT, a first-level optical splitter, a distribution optical cable, a second-level optical splitter, an indoor optical cable, an ONU, an ONU-to-customer terminal device line, a customer router, and a customer terminal. The present application proposal starts from the whole process of fiber to the home, selects relevant network indicators layer by layer, and the selected indicators cover network communication processes such as gateway, OLT machine room, passive optical network (PON) port splitting, ONU to user terminals. Each process and each node can affect network quality and cause customer perception to deteriorate, which can further lead to customer complaints. Therefore, obtaining the first data set based on the network topology dimension can cover each stage of the network topology, improve the accuracy of the first data set acquisition, and improve the accuracy of the prediction result acquisition.

[0081] Second, the use scenario dimension. The specific use scenario of the user also affects his perception. For example, from the whole process of fiber to the home to providing various services, various application scenarios of user Internet access are fully considered, including browsing web pages, playing games, watching videos, downloading, and the like.

[0082] Thirdly, network quality feature dimension. Considering the demand characteristics of users using broadband services, multi-dimensional indexes including delay, rate, packet loss rate, success rate, usage rate, etc. are selected.

[0083] Among them, based on 485 network indexes, network index screening is carried out combined with feature engineering. Feature engineering is an indispensable part of model construction in machine learning. Feature engineering can extract features from raw data to the greatest extent. According to the statistical characteristics of different network indexes, feature engineering can form new features through the statistical values of a single or multiple variables such as maximum value, minimum value, mean value, etc. to extract features from raw data to the greatest extent for algorithms and models. Finally, 31 types of network quality data are selected. For example, according to the use of the network corresponding to the user, the index weight of the index dimension that the user in different characteristics and different scenes values in the actual network use can be determined. Therefore, the association between network index data and user complaint data can be established, which can improve the comprehensiveness of network index data acquisition and the accuracy of prediction result acquisition.

[0084] According to some embodiments, after obtaining the user original data, i.e. the first data set, feature extraction can be performed on the first data set to obtain the second data set. The second data set is used to input into the feature extraction unit to obtain the prediction model. The feature extraction work is performed according to the feature extraction rule. The feature extraction rule of each index is determined by the data type (the nature of the data), i.e. the influence of the model on the network quality. For example: the lower the http response success rate, the worse the network quality. The specific rules are shown in Table 1. Among them, max represents the maximum value of the index within 24 hours, min represents the minimum value of the index within 24 hours, mean represents the mean value of the index within 24 hours, and add represents the sum of all data of the index within 24 hours. ​​​​characteristic matrix, each sample data has 33 dimensions, such as (00001D64C9B4F538189D201334F30DBC, 1332.0, 0.9369, 2063.6114279827534, 4889.571428571428, 1119.0, 0.0046, 0.0134, 5164.285714285715, 0.6474, 256.0, 512.0, 17.0, 7.0, 25.0, 129.0, 4.0, 2.0, 2.1996571428571428, 1.8408571428571427, -20.42338571428571, -21.547185714285714, 1509.0161, 17464.1953, 0.0, 1.0, 249870.0, 237.0, 121.0, 60.0, 0.0, 0.0, 1), wherein, in the 33 dimensions, the first column is the user account number encrypted by MD5, and the last column is the label, indicating whether the user makes a complaint, 0 represents that the user is a non-complaint user, and 1 represents that the user is a complaint user, and the 31 columns in the middle are the feature data obtained based on table 1.

[0085] Table 1

[0086]

[0087] According to some embodiments, the second prediction model of the embodiments of the present disclosure may, for example, include algorithms such as decision tree, random forest, XGBoost, etc., wherein the first prediction model may, for example, predict the complaint probability in the future m days. The m days may, for example, be 1 day, and may also be 3 days. Wherein, the preset number of days of the embodiments of the present disclosure may, for example, be 1 day. Wherein, the preset number of days may, for example, be determined according to the prediction result of the first prediction model for the test data subset.

[0088] According to some embodiments, the third prediction model is an XGBoost model, and the first prediction model is obtained by training the third prediction model using the second data set. The model error corresponding to the XGBoost model is obtained using the second data set. The structure error of the XGBoost model is obtained using the second data set. The objective function corresponding to the XGBoost model is obtained according to the model error and the structure error, and the XGBoost model trained is obtained. Therefore, the complexity of the first prediction model can be penalized, the overfitting of the first prediction model can be reduced, the error between the prediction result and the actual result corresponding to the sample can be reduced, and the accuracy of the first prediction model can be improved.

[0089] In the XGBoost model, the specific function may, for example, be as shown in formula (1):

[0090] (1)

[0091] In formula (1), y represents the true value of the sample, and is the model error, that is, the error between the true value and the predicted value of the sample, where represents the true value, represents the predicted value. is the structure error of the model, that is, the regularization term, which is used to limit the complexity of the model.

[0092] According to an embodiment, the complaint problem of the present disclosure may, for example, be a binary classification problem, and the loss function may, for example, be a cross-entropy loss function. The formula of the cross-entropy loss function may, for example, be as shown in formula (2):

[0093] (2)

[0094] In formula (2), n represents the number of training samples, represents the true value of the sample, which is usually 0 and 1 in a binary classification problem, represents the probability that the sample is predicted to be 0.

[0095] According to some embodiments, the method further comprises:

[0096] obtaining an information gain change value corresponding to any XGBoost node in the XGBoost model;

[0097] when the information gain change value is greater than a first threshold value, splitting the any XGBoost node;

[0098] when the information gain change value is less than or equal to a second threshold value, not splitting the any XGBoost node, where the first threshold value is greater than or equal to the second threshold value.​

[0099] The first threshold value can be the same as the second threshold value, and the first threshold value and the second threshold value can be, for example, both zero.

[0100] The formula of the information gain change value can be as shown in formula (3):

[0101] (3)

[0102] According to some embodiments, refers to the information gain of the left child node;

[0103] refers to the information gain of the right child node;

[0104] refers to the information entropy of the left child node;

[0105] refers to the information entropy of the right child node;

[0106] refers to the regularization term.

[0107] In step S12, according to the obtained parameter set corresponding to the second prediction model, the parameters in the second prediction model are set, and a third prediction model is obtained.

[0108] According to some embodiments, the parameter set corresponding to the second prediction model can be obtained. When the parameter set is obtained, according to the obtained parameter set corresponding to the second prediction model, the parameters in the second prediction model are set, and a third prediction model is obtained.

[0109] For example, it can be built by relying on Scikit-learn, xgboost, pandas, numpy and other packages in Python. The model parameter adjustment, training and result evaluation can be automatically implemented.

[0110] According to some embodiments, the parameter set can include, for example, the parameter booster and the booster parameter. The booster parameter can be, for example, the parameter in the XGBoost model, and can be, for example, the parameter that needs to be adjusted when the booster is ‘gbtree’.

[0111] The parameter booster has three optional values [gbtree, gblinear, dart], which are used to specify the type of weak learner. The default value is ‘gbtree’, which means that the tree-based model is used for calculation. The booster parameter can be, for example, as shown in Table 2.

[0112] Table 2

[0113]

[0114] According to some embodiments, eta / learning rate is used to prevent overfitting by using step shrinkage in the update. After each boosting step, we can directly get the weights of the new features, eta shrinks the weights of the features, making the boosting process more conservative. Value range: [0, 1]. Recommended candidate values: [0.01, 0.015, 0.025, 0.05, 0.1].

[0115] In some embodiments, gamma is used to specify the minimum loss reduction required to make a split on a leaf node, with a default value of 0. The larger the value set, the more conservative the model. Value range: [0, ∞]. Recommended candidate values: [0, 0.05 0.1, 0.3, 0.5, 0.7, 0.9, 1].

[0116] In some embodiments, alpha / reg_alpha is the L1 regularization weight term. Increasing this value will make the model more conservative. Recommended candidate values: [0, 0.01 0.1, 1].

[0117] In some embodiments, lambda / reg_lambda is the L2 regularization weight term. Increasing this value will make the model more conservative. Recommended candidate values: [0, 0.1, 0.5, 1].

[0118] In some embodiments, max_depth is used to indicate the maximum depth of the tree. Increasing this value will make the model more complex and more prone to overfitting. 0 means no limit on depth. Value range: [0, ∞]. Recommended values: [3, 5, 6, 7, 9, 12, 15, 17, 25].

[0119] In some embodiments, min_child_weight is used to specify the minimum sample weight sum in the child node, and if the sample weight sum of a leaf node is less than min_child_weight, the splitting process ends, with a default value of 1. The larger the parameter, the more conservative the algorithm. Value range: [0, ∞]. Recommended candidate values: [1, 3, 5, 7]

[0120] In some embodiments, the default value of subsample is 1, which specifies that subsample * n_samples samples are sampled for training weak learners. If set to 0.5, it means that XGBoost will randomly draw 50% of the sample set to build tree models, which can prevent overfitting. The value range is (0, 1]. The recommended candidate values are: [0.6, 0.7, 0.8, 0.9, 1]

[0121] In some embodiments, tree_method is used to indicate the tree construction algorithm used in XGBoost. The default is auto, which uses heuristics to choose the fastest method. For small datasets, the exact greedy method will be used. For larger datasets, the approximate algorithm is selected.

[0122] In some embodiments, the task parameter mainly needs to modify the parameter objective, which is used to specify the learning task and the corresponding learning goal, for example, set the value of the parameter objective to "binary:logistic", and output the probability of user identification complaint and non-complaint and .

[0123] For example, according to some embodiments, a threshold can be set for the output of the prediction model in the training stage, so as to eliminate the user identification whose probability output by the prediction model is less than the threshold, thereby improving the prediction accuracy of the prediction model. For example, 0 represents that the user identification is a non-complaint user identification, and 1 represents that the user identification is a complaint user identification. In the training model stage, it is necessary to explicitly obtain the complaint and non-complaint user identifications. In the application stage, 0 (non-complaint) and 1 (complaint) are the prediction results of the model. The probability refers to the probability of predicting that the user identification is complaint and non-complaint, and the sum is 1. By default, the prediction model will output the category with a probability greater than 0.5. For example, the threshold α can be set to 0.7, which can improve the accuracy of the model output result.

[0124] The threshold can also be determined based on a threshold setting instruction, which is not limited in the embodiments of the present disclosure.

[0125] In step S13, a third prediction model is trained using the second data set to obtain a first prediction model.

[0126] According to some embodiments, the training of the third prediction model using the second data set to obtain the first prediction model comprises:

[0127] According to the set size corresponding to the second data set, a search mode corresponding to the second data set is obtained; the third prediction model is trained by using the search mode and the second data set, and in a case where a prediction result of the third prediction model meets a prediction requirement, a first prediction model is obtained. Therefore, the efficiency of model traversal data set can be improved, the convenience time is reduced, and the consumption of resources is saved.

[0128] Among them, two search modes of grid search (GridSearchCV) or random search (RandomizedSearchCV) can be used to find the optimal parameters of the prediction model. When the data set is small, the GridSearchCV search mode is recommended, and when the data set is large, the RandomizedSearchCV search mode is recommended. For example, the data set can be small when the number corresponding to the data set is less than a number threshold. For example, the data set can be large when the number corresponding to the data set is greater than the number threshold.

[0129] According to some embodiments, in a case where a prediction result of the third prediction model meets a prediction requirement, a first prediction model is obtained, including:

[0130] A prediction complaint result corresponding to the third prediction model is obtained; according to the complaint result corresponding to the first data set and the prediction complaint result, an accuracy rate, a precision rate and a recall rate corresponding to the prediction result are obtained; according to the precision rate, a first weight corresponding to the precision rate, the recall rate and a second weight corresponding to the recall rate, a score corresponding to the third prediction model is obtained; in a case where the accuracy rate, the precision rate, the recall rate and the score meet a prediction requirement, it is determined that the first prediction model is obtained. Therefore, the prediction model can be evaluated, the case where the prediction result output by the prediction model is inconsistent with the actual result can be reduced, and the accuracy of the prediction model prediction can be improved.

[0131] In some embodiments, the prediction model prediction result and the actual complaint situation corresponding to the user identifier are compared to obtain a 2x2 confusion matrix, for example, as shown in Table 3.

[0132] Table 3

[0133]

[0134] Among them, the accuracy rate is the proportion of all correct predictions (positive and negative classes) in the total, and the calculation formula is formula (4):

[0135] (4)

[0136] The precision rate is the proportion of correct predictions as positive in all predictions as positive, and the calculation formula is formula (5):

[0137] (5)

[0138] The recall rate is the proportion of correct predictions of positive to all actual positives, and the calculation formula is formula (6):

[0139] (6)

[0140] The F1-Score is a calculation result considering the model precision and recall, and the larger the better. When the F1 value is small, the True Positive increases relatively, and the false decreases relatively, that is, the Precision and Recall increase relatively, that is, the F1 weights the Precision and Recall. The calculation formula is formula (7):

[0141] (7)

[0142] For example, the first data set of different time periods can be obtained, and the second prediction model can be trained and continuously adjusted, which can improve the accuracy of the first prediction model and improve the accuracy of the complaint prediction.

[0143] For example, the accuracy of the prediction result of the different preset days predicted by the prediction model can also be determined to determine the preset days corresponding to the prediction model, thereby improving the accuracy of the prediction result.

[0144] In one or related embodiments, the first data set can be subjected to feature extraction according to a feature extraction rule to obtain a second data set, wherein the first data set includes at least one first network quality data subset corresponding to a network complaint user identifier and at least one second network quality data subset corresponding to a network non-complaint user identifier; the parameters in the second prediction model are set according to the obtained parameter set corresponding to the second prediction model to obtain a third prediction model; and the third prediction model is trained using the second data set to obtain the first prediction model. Therefore, the first data set can be subjected to feature extraction, the data features can be screened, the model can be simplified, the training time of the model can be reduced, the precision of the prediction model can be improved, and the accuracy of the prediction result can be improved according to the complaint data set.

[0145] According to some embodiments, Figure 3is an example schematic diagram of a complaint prediction system according to an example embodiment. The complaint prediction system includes a data acquisition unit, a feature extraction unit, a model parameter adjustment unit, a model training unit, a threshold screening unit, and a model evaluation unit. The data acquisition unit is configured to acquire a network indicator data set to obtain a first data set. The feature extraction unit is configured to extract features from the first data set according to a feature extraction rule to obtain a second data set. The model parameter adjustment unit is configured to set parameters of a prediction model. The model training unit is configured to train a first prediction unit. The threshold screening unit is configured to determine a threshold value. The model evaluation unit is configured to determine a prediction result. The specific functions of each unit are described above and will not be repeated here.

[0146] According to some embodiments, Figure 4 is an example schematic diagram of a model training and complaint prediction method according to an example embodiment. For example, the method can include acquiring a network indicator data set, which can include 485 indicators and home broadband user complaint data. The home broadband user complaint data can be used to determine whether a user identifier is a complaint user identifier. Feature engineering can use a feature extraction rule to extract features to obtain training samples.

[0147] To improve the prediction effect of the model, the model is optimized and then evaluated, mainly including a double-window mechanism of model training and complaint prediction, an intelligent optimization mechanism of machine learning algorithm, a double optimization mechanism of prediction model based on data updating and feedback calibration, etc. The double-window mechanism of model training and complaint prediction refers to window one: for the data window before the complaint, the training set and the test set are selected to construct the model by using data of different time spans; window two: for the prediction window, the complaint situation of different time spans is predicted. The intelligent optimization mechanism of machine learning algorithm: different complaint data can be suitable for different machine learning algorithms. By evaluating the complaint prediction results of multiple machine learning algorithms, the optimal algorithm is selected according to the prediction results. The double optimization mechanism of the prediction model refers to the fact that the complaint situation of each time period is different, so the model needs to be iterated and adjusted. Based on the double improvement mechanism of data updating and feedback calibration, the model is iterated and the parameters are continuously adjusted based on the newly added complaint data every month to improve the robustness of the model.

[0148] Figure 5 is a flowchart of a complaint prediction method according to an example embodiment, as Figure 5 shown, the complaint prediction method can be used in a communication scenario and includes the following steps:

[0149] In step S21, a network index data set corresponding to any user identifier in the full-quantity user identifier is obtained, the network index data set including a first network index data subset corresponding to a network node device and / or a second network index data subset corresponding to a communication medium;

[0150] According to some embodiments, the technical solutions of the embodiments of the present disclosure can be used in a communication scenario, for example, in the personal home service subfield in the field of artificial intelligence. The technical solutions of the embodiments of the present disclosure can be used for predicting complaints of home broadband users based on network quality, so as to discover problems in network as a service in advance, and to timely warn and rectify, so as to improve the satisfaction of users using the network.

[0151] According to some embodiments, the full-quantity user identifier can indicate the identification information of all users in a certain area. The full-quantity user identifier of the embodiments of the present disclosure can refer to all users in the area used by the complaint prediction method.

[0152] In some embodiments, the network index data set refers to a collection of at least one network index data, which can be a network index data set corresponding to any user identifier in the full-quantity user identifier. The network index data set does not refer to a fixed data set. For example, when the data type set corresponding to the network index data set changes, the network index data set can also change accordingly. For example, when the specific data included in the network index data set changes, the network index data set can also change accordingly.

[0153] The first network index data subset corresponding to the network node device is only used to distinguish the second network index data subset, and does not refer to a fixed data subset. The network node device includes but is not limited to network layer, optical line terminal (OLT), primary optical splitter, secondary optical splitter, optical network unit (ONU), customer router, customer terminal, etc. The communication medium includes but is not limited to wiring optical cable, indoor skin optical cable, ONU to customer terminal device line, indoor optical fiber, etc.

[0154] According to some embodiments, when the complaint prediction method is executed, a network index data set corresponding to any user identifier in the full-quantity user identifier can be obtained, the network index data set including a first network index data subset corresponding to a network node device and / or a second network index data subset corresponding to a communication medium.

[0155] In step S22, the network index data set is preprocessed to obtain a processed network index data set.

[0156] According to some embodiments, the pre-processing of the network indicator data set may, for example, be obtaining a network indicator data subset corresponding to each data type in the network indicator data set, and obtaining network indicator data corresponding to each data type according to a data extraction manner corresponding to the data type, and adding the network indicator data to the processed network indicator data set.

[0157] The processed network indicator data set is not specific to a fixed set. For example, when the pre-processing manner changes, the processed network indicator data set can also change accordingly. For example, when the network indicator data changes, the processed network indicator data set can also change accordingly.

[0158] According to some embodiments, when the network indicator data set is obtained, the network indicator data set can be pre-processed to obtain a processed network indicator data set.

[0159] In step S23, the processed network indicator data set is input into the first prediction model for prediction to obtain a prediction result, wherein the prediction result is used to indicate the probability of complaint against network quality within a preset number of days for any user identifier.

[0160] According to some embodiments, the first prediction model may, for example, be a model that has been trained and can be used for complaint prediction. The first prediction model is not specific to a fixed model. For example, when the model parameters corresponding to the first prediction model change, the first prediction model can also change accordingly. For example, when the specific algorithm corresponding to the first prediction model changes, the first prediction model can also change accordingly. The first prediction model is Figures 1 to 4 obtained by the model training method.

[0161] In some embodiments, the prediction result is the result output by the first prediction model when the processed network indicator data set is input into the first prediction model. The prediction result may, for example, be used to indicate the probability of complaint against network quality within a preset number of days for any user identifier. The prediction result is not specific to a fixed result, for example, when the first prediction model

[0162] The preset number of days may, for example, be a future preset number of days, and the preset number of days is not specific to a fixed number of days. For example, when a modification instruction for the preset number of days is received, the preset number of days can also change accordingly. The preset number of days may, for example, be 1 day, and the preset number of days may, for example, also be 3 days.

[0163] Optionally, the complaint probability may be, for example, a probability indicating that any user identifier complains about network quality. The complaint probability does not refer to a fixed probability. For example, when the specific probability value of the complaint probability changes, the complaint probability may also change accordingly. For example, when the network indicator data set or the first prediction model changes, the complaint probability may also change accordingly.

[0164] According to some embodiments, when the network indicator data set is obtained, the network indicator data set may be preprocessed to obtain a processed network indicator data set, and the processed network indicator data set may be input into the first prediction model for prediction to obtain a prediction result, wherein the prediction result is used to indicate a complaint probability of the any user identifier within a preset number of days for network quality.

[0165] According to some embodiments, the complaint result of the any user identifier within a preset number of days is obtained.

[0166] According to the prediction result and the complaint result, the first prediction model is adjusted to obtain an adjusted first prediction model, so that the first prediction model can be adjusted, the first prediction model can be optimized, and the accuracy of the first prediction model can be improved to improve the probability of obtaining the prediction result.

[0167] According to some embodiments, after the prediction result is obtained, the corresponding measures can be performed on the user identifier in advance to reduce the complaint probability of the user identifier, and the user identifier can be pre-positioned.

[0168] In some or related embodiments, by obtaining a network indicator data set corresponding to any user identifier in a full quantity of user identifiers, the network indicator data set includes a first network indicator data subset corresponding to a network node device and / or a second network indicator data subset corresponding to a communication medium; the network indicator data set is preprocessed to obtain a processed network indicator data set; the processed network indicator data set is input into a first prediction model for prediction to obtain a prediction result, wherein the prediction result is used to indicate a complaint probability of the any user identifier within a preset number of days for network quality. Therefore, a complaint prediction mechanism can be provided, relevant network indicator data can be selected layer by layer according to the online service process, the situation that the complaint prediction is inaccurate due to the determination of the complaint prediction based only on the fault of the target device can be reduced, the accuracy of the prediction result within the preset number of days can be improved, and the accuracy of the complaint prediction can be improved.

[0169] Figure 6 is a model training device block diagram according to an exemplary embodiment. Referring to Figure 6 , the model training device 600 includes:

[0170] The feature extraction unit 601 is configured to perform feature extraction on the first data set according to a feature extraction rule to obtain a second data set, wherein the first data set includes at least one first network quality data subset corresponding to a network complaint user identifier and at least one second network quality data subset corresponding to a network non-complaint user identifier.

[0171] The parameter setting unit 602 is configured to set parameters in the second prediction model according to a parameter set corresponding to the obtained second prediction model to obtain a third prediction model.

[0172] The model obtaining unit 603 is configured to train the third prediction model by using the second data set to obtain a first prediction model.

[0173] According to some embodiments, the feature extraction unit 601 is configured to, when performing feature extraction on the first data set according to the feature extraction rule to obtain the second data set, specifically configured to:

[0174] perform missing value detection on the first data set, and remove data with missing values in the first data set to obtain a third data set;

[0175] perform feature extraction on the third data set by using the feature extraction rule to obtain the second data set;

[0176] obtain a training data subset and a test data subset in the second data set according to a data set ratio.

[0177] According to some embodiments, the third prediction model is an XGBoost model, and the model obtaining unit 603 is configured to, when training the third prediction model by using the second data set to obtain the first prediction model, specifically configured to:

[0178] obtain a model error corresponding to the XGBoost model by using the second data set;

[0179] obtain a structure error of the XGBoost model by using the second data set;

[0180] obtain a target function corresponding to the XGBoost model according to the model error and the structure error to obtain the trained XGBoost model.

[0181] According to some embodiments, the model obtaining unit 603 is further configured to:

[0182] obtain an information gain change value corresponding to any XGBoost node in the XGBoost model;

[0183] when the information gain change value is greater than a first threshold value, splitting the any XGBoost node;

[0184] when the information gain change value is less than or equal to a second threshold value, not splitting the any XGBoost node, wherein the first threshold value is greater than or equal to the second threshold value.

[0185] According to some embodiments, the model obtaining unit 603 is configured to train a third prediction model by using the second data set, and obtain the first prediction model, specifically configured to:

[0186] obtain a search manner corresponding to the second data set according to a set size corresponding to the second data set;

[0187] train the third prediction model by using the search manner and the second data set, and obtain the first prediction model when a prediction result of the third prediction model meets a prediction requirement.

[0188] According to some embodiments, the model obtaining unit 603 is configured to obtain the first prediction model when the prediction result of the third prediction model meets the prediction requirement, specifically configured to:

[0189] obtain a predicted complaint result corresponding to the third prediction model;

[0190] obtain an accuracy rate, a precision rate and a recall rate corresponding to the prediction result according to a complaint result corresponding to the first data set and the predicted complaint result;

[0191] obtain a score corresponding to the third prediction model according to the precision rate, a first weight corresponding to the precision rate, the recall rate and a second weight corresponding to the recall rate;

[0192] determine that the first prediction model is obtained when the accuracy rate, the precision rate, the recall rate and the score meet a prediction requirement.

[0193] In one or related embodiments, a feature extraction unit is used to extract features from a first dataset according to feature extraction rules to obtain a second dataset, wherein the first dataset includes at least one first subset of network quality data corresponding to at least one network complaint user identifier and at least one second subset of network quality data corresponding to at least one network non-complaint user identifier; a parameter setting unit is used to set the parameters in the third prediction model according to the parameter set corresponding to the obtained second prediction model to obtain a third prediction model; a model acquisition unit is used to train the third prediction model using the second dataset to obtain a first prediction model. Therefore, feature extraction can be performed on the first dataset, data features can be filtered, the model can be simplified, the training time of the model can be reduced, and the accuracy of the prediction model can be improved. At the same time, based on the complaint dataset, the accuracy of the prediction model acquisition and the accuracy of the prediction results acquisition can be improved.

[0194] Figure 7 This is a block diagram illustrating a complaint prediction device according to an exemplary embodiment. (Refer to...) Figure 7 The complaint prediction device 700 includes:

[0195] The data acquisition unit 701 is used to acquire a set of network indicator data corresponding to any user identifier in the full set of user identifiers. The set of network indicator data includes a first subset of network indicator data corresponding to network node devices and / or a second subset of network indicator data corresponding to communication media.

[0196] Data processing unit 702 is used to preprocess the network indicator data set to obtain a processed network indicator data set;

[0197] The complaint prediction unit 703 is used to input the processed network indicator data set into a first prediction model for prediction and obtain a prediction result. The prediction result indicates the probability of a complaint regarding network quality related to any user identifier within a preset number of days. The first prediction model is based on... Figures 1 to 4 The model was obtained using the aforementioned model training method.

[0198] According to some embodiments, the complaint prediction unit 703 is also used for

[0199] Obtain the complaint results of any of the user identifiers within a preset number of days;

[0200] Based on the prediction results and the complaint results, the first prediction model is adjusted to obtain the adjusted first prediction model.

[0201] In some or related embodiments, the data acquisition unit is configured to acquire a network index data set corresponding to any of the full quantity of user identities, the network index data set comprising a first network index data subset corresponding to a network node device and / or a second network index data subset corresponding to a communication medium; the data processing unit is configured to pre-process the network index data set to obtain a processed network index data set; and the complaint prediction unit is configured to input the processed network index data set into a first prediction model for prediction to obtain a prediction result, wherein the prediction result is used to indicate a complaint probability of the any of the user identities within a preset number of days for network quality. Thus, a complaint prediction mechanism can be provided, relevant network index data can be selected layer by layer according to the online service process, the situation that the complaint prediction is inaccurate due to the determination of the complaint prediction based on only the fault of the target device can be reduced, the accuracy of the prediction result obtained within the preset number of days can be improved, and the accuracy of the complaint prediction can be improved.

[0202] Figure 8 is a block diagram of a terminal device UE 800 according to an embodiment of the present disclosure. For example, the UE 800 can be a mobile phone, a computer, a digital broadcast terminal device, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, or the like.

[0203] Referring to Figure 8 , the UE 800 can include at least one of a processing component 802, a memory component 804, a power supply component 806, a multimedia component 808, an audio component 810, an input / output (I / O) interface 812, a sensor component 814, and a communication component 816.

[0204] The processing component 802 usually controls overall operations of the UE 800, such as operations associated with displaying, making phone calls, data communications, camera operations and recording operations. The processing component 802 can include at least one processor 820 to execute instructions to complete all or part of steps of the methods described above. Further, the processing component 802 can include at least one module to facilitate interaction between the processing component 802 and other components. For example, the processing component 802 can include a multimedia module to facilitate the interaction between the multimedia component 808 and the processing component 802.

[0205] The memory 804 is configured to store various types of data to support the operation of the UE 800. Examples of such data include instructions for any application or method operating on the UE 800, contact data, phonebook data, messages, pictures, videos, and so on. The memory 804 can be implemented by any type of volatile or nonvolatile storage devices 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 memory, flash memory, magnetic disk, or optical disk.

[0206] The power supply component 806 supplies electrical power for the various components of the UE 800. The power supply component 806 can include a power supply management system, at least one power supply, and other components associated with generating, managing and distributing electrical power for the UE 800.

[0207] The multimedia component 808 includes a screen providing an output interface between the UE 800 and a user. In some embodiments, the screen can include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen can be implemented as a touch screen to receive input signals from a user. The touch panel includes at least one touch sensor to sense touch, swiping and gesture on the touch panel. The touch sensor can not only sense a boundary of a touching or swiping action, but also detect a pressure associated with the touching or swiping action. In some embodiments, the multimedia component 808 includes a front camera and / or a rear camera. The front and / or rear camera can receive external multimedia data when the UE 800 is in an operation mode such as a shooting mode or a video mode. Each of the front and rear camera can be a fixed optical lens system or have a focal length and optical zoom capability.

[0208] The audio component 810 is configured to output and / or input audio signals. For example, the audio component 810 includes a microphone (MIC) that is configured to receive external audio signals when the UE 800 is in an operation mode such as a calling mode, a recording mode and a voice recognition mode. The received audio signals can be further stored in the memory 804 or transmitted via the communication component 816. In some embodiments, the audio component 810 also includes a speaker for outputting audio signals.

[0209] The I / O interface 812 provides an interface between the processing component 802 and peripheral interface modules such as a keyboard, a click wheel, buttons, and so on. The buttons can include, but are not limited to, a home button, a volume button, a start button and a lock button.

[0210] The sensor component 814 includes at least one sensor to provide various state assessments for the UE 800. For example, the sensor component 814 can detect an open / closed position of the UE 800, relative positioning of components, such as a display and a keypad of the UE 800, changes in position of the UE 800 or a component of the UE 800, presence or absence of user contact with the UE 800, orientation or acceleration / deceleration / g-force and temperature changes of the UE 800. The sensor component 814 can include a proximity sensor configured to detect presence of a nearby object without any physical touch. The sensor component 814 can also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor component 814 can also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor or a temperature sensor.

[0211] The communication component 816 is configured to facilitate wired or wireless communication between the UE 800 and another device. The UE 800 can access a wireless network based on a communication standard, such as WiFi, 2G, or 3G, or a combination thereof. In an exemplary embodiment, the communication component 816 receives a broadcast signal or broadcast related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 816 further includes a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on Radio Frequency Identification (RFID) techniques, Infrared Data Association (IrDA) techniques, Ultra-Wide Band (UWB) techniques, Bluetooth (BT) techniques, and other techniques.

[0212] In an exemplary embodiment, the UE 800 can be implemented by at least one of application-specific integrated circuit (ASIC), digital signal processor (DSP), digital signal processing device (DSPD), programmable logic device (PLD), field programmable gate array (FPGA), controller, microcontroller, microprocessor, or other electronic elements, for performing the above-described methods.

[0213] The various embodiments of the systems and techniques described above can be implemented in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a load programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device. The program code for implementing the methods of the disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces the functions / operations specified in the flowcharts and / or the block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine as a standalone software package, and partially on a remote machine or entirely on a remote machine or server.

[0214] In the context of the present disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium will include one or more lines of electrical conductors, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or Flash memory), optical fibers, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0215] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0216] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), the Internet, and a blockchain network.

[0217] The computer system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. A server can be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system, to solve the defects of large management difficulty and weak business scalability in traditional physical host and VPS service ("Virtual Private Server", or simply "VPS"). The server can also be a server of a distributed system, or a server combined with a blockchain.

[0218] It should be understood that various forms of flow shown above can be used, with steps reordered, added, or removed. For example, the steps recited in the present disclosure can be performed in parallel, in series, or in a different order, without limitation herein, as long as the desired results of the technical solutions of the present disclosure can be achieved.

[0219] The above detailed description does not limit the scope of the disclosure. Various modifications, combinations, sub-combinations and alternatives can be made to the detailed description. Any modification, equivalent replacement and improvement etc. made within the spirit and principle of the disclosure shall be included in the scope of the disclosure.

Claims

1. A model training method, characterized in that, The method comprises: performing feature extraction on the first data set according to a feature extraction rule to obtain a second data set, wherein the first data set comprises at least one first network quality data subset corresponding to a network complaint user identifier and at least one second network quality data subset corresponding to a network non-complaint user identifier; setting parameters in the second prediction model according to a parameter set corresponding to the second prediction model to obtain a third prediction model; training the third prediction model using the second data set to obtain a first prediction model.

2. The method of claim 1, wherein, The method further comprises: performing missing value detection on the first data set to remove data with missing values in the first data set to obtain a third data set; performing feature extraction on the third data set using a feature extraction rule to obtain a second data set; obtaining a training data subset and a test data subset in the second data set according to a data set ratio.

3. The method of claim 1, wherein, The third prediction model is an XGBoost model, and the method further comprises: obtaining a model error corresponding to the XGBoost model using the second data set; obtaining a structure error of the XGBoost model using the second data set; obtaining a target function corresponding to the XGBoost model according to the model error and the structure error to obtain a trained XGBoost model. The method further comprises:

4. The method of claim 3, wherein, obtaining an information gain change value corresponding to any XGBoost node in the XGBoost model; splitting the any XGBoost node when the information gain change value is greater than a first threshold value; not splitting the any XGBoost node when the information gain change value is less than or equal to a second threshold value, wherein the first threshold value is greater than or equal to the second threshold value. The method further comprises:

5. The method of claim 1, wherein, obtaining a search method corresponding to the second data set according to a set size corresponding to the second data set; training the third prediction model using the search method and the second data set, and obtaining a first prediction model when a prediction result of the third prediction model meets a prediction requirement. The method further comprises:

6. The method of claim 5, wherein, obtaining a predicted complaint result corresponding to the third prediction model; obtaining an accuracy, a precision, and a recall rate corresponding to the prediction result according to a complaint result corresponding to the first data set and the predicted complaint result; obtaining a score corresponding to the third prediction model according to the precision, a first weight corresponding to the precision, the recall rate, and a second weight corresponding to the recall rate; determining that the first prediction model is obtained when the accuracy, the precision, the recall rate, and the score meet the prediction requirement. ​ 7. A complaint prediction method characterized by, The method comprises: obtaining a network indicator data set corresponding to any user identifier in the full user identifier, the network indicator data set comprising a first network indicator data subset corresponding to a network node device and / or a second network indicator data subset corresponding to a communication medium; preprocessing the network indicator data set to obtain a processed network indicator data set; inputting the processed network indicator data set into a first prediction model for prediction to obtain a prediction result, wherein the prediction result is used to indicate a complaint probability of the any user identifier within a preset number of days for network quality, and the first prediction model is obtained based on the model training method of any one of claims 1 to 6.

8. The method of claim 7, wherein, The method further comprises: obtaining a complaint result of the any user identifier within a preset number of days; adjusting the first prediction model according to the prediction result and the complaint result to obtain an adjusted first prediction model.

9. A model training apparatus, comprising: comprises: a feature extraction unit configured to extract features from a first data set according to a feature extraction rule to obtain a second data set, wherein the first data set comprises a first network quality data subset corresponding to at least one network complaint user identifier and a second network quality data subset corresponding to at least one network non-complaint user identifier; a parameter setting unit configured to set parameters in the second prediction model according to a parameter set corresponding to the second prediction model to obtain a third prediction model; a model obtaining unit configured to train the third prediction model using the second data set to obtain a first prediction model.

10. A complaint prediction apparatus characterized by comprising: The device comprises: a data obtaining unit configured to obtain a network indicator data set corresponding to any user identifier in the full user identifier, the network indicator data set comprising a first network indicator data subset corresponding to a network node device and / or a second network indicator data subset corresponding to a communication medium; a data processing unit configured to preprocess the network indicator data set to obtain a processed network indicator data set; a complaint prediction unit configured to input the processed network indicator data set into a first prediction model for prediction to obtain a prediction result, wherein the prediction result is used to indicate a complaint probability of the any user identifier within a preset number of days for network quality, and the first prediction model is obtained based on the model training method of any one of claims 1 to 6.

11. A network-side device, characterized by comprising: comprises: a processor; a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement the model training method of any one of claims 1 to 6 and the complaint prediction method of any one of claims 7 to 8.

12. A storage medium, when instructions in the storage medium are executed by a processor of an electronic device, enabling the electronic device to perform the model training method of any one of claims 1 to 6 and the complaint prediction method of any one of claims 7 to 8.

Images