Complaint data processing method, electronic equipment and storage medium

Abstract

The invention provides a complaint data processing method, a complaint data construction method, electronic equipment and a storage medium, and belongs to the technical field of data processing. The method comprises the following steps: determining complaint comprehensive characteristics corresponding to user information and network element information according to complaint data to be processed corresponding to the user information and the network element information; and inputting the complaint comprehensive characteristics into a complaint positioning model of a service domain corresponding to the to-be-processed complaint data to obtain a complaint positioning result corresponding to the user information and the network element information. According to the embodiment of the invention, the complaint positioning result can be quickly and accurately determined, and the processing efficiency of the complaint data is improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a complaint data processing method, an electronic device and a readable storage medium. Background Art

[0002] With the continuous development of mobile Internet services, users have higher and higher requirements for the quality of mobile network services, and there are more and more complaints about mobile communication services. In related technologies, common complaint handling methods include manual handling and rule-based handling. Manual handling refers to the process in which professionals communicate with users to determine the cause of the user's complaint and locate the corresponding service and specific complaint location; rule-based handling refers to the automatic delimitation and location of the cause of the complaint by analyzing the content of the user's complaint according to pre-set rules. Among them, the manual handling method requires a lot of human resources, while the rule-based handling method is easily limited by known experience and is difficult to effectively handle new types of complaints. Summary of the invention

[0003] The present application provides a complaint data processing method, an electronic device and a readable storage medium.

[0004] An embodiment of the present application provides a complaint data processing method, which includes: determining comprehensive complaint features corresponding to the user information and the network element information based on the complaint data to be processed corresponding to the user information and the network element information; inputting the comprehensive complaint features into a complaint location model of the business domain corresponding to the complaint data to be processed, and obtaining a complaint location result corresponding to the user information and the network element information.

[0005] An embodiment of the present application provides an electronic device, including: one or more processors; a memory, on which one or more programs are stored. When the one or more programs are executed by one or more processors, the one or more processors implement any one of the complaint data processing methods in the embodiments of the present application.

[0006] An embodiment of the present application provides a readable storage medium, which stores a computer program. When the computer program is executed by a processor, any one of the complaint data processing methods in the embodiments of the present application is implemented.

[0007] In an embodiment of the present application, based on the complaint data to be processed corresponding to the user information and the network element information, the comprehensive complaint features corresponding to the user information and the network element information are determined; the comprehensive complaint features are input into the complaint location model of the business domain corresponding to the complaint data to be processed, and the complaint location results corresponding to the user information and the network element information are obtained. It can be seen that in the present application, the correlation between the business domain and the corresponding features is fully considered. Therefore, a corresponding complaint location model is set for each business domain. When handling complaints, the extracted comprehensive complaint features are input into the complaint location model of the corresponding business domain for processing, thereby improving the processing capability of the comprehensive complaint features and improving the processing accuracy and efficiency of the complaint data.

[0008] With regard to the above embodiments and other aspects of the present application and their implementation, further description is provided in the accompanying drawings, detailed description and claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 A flowchart of a complaint data processing method provided in an embodiment of the present application is shown.

[0010] Figure 2 A flowchart of a complaint data processing method provided in an embodiment of the present application is shown.

[0011] Figure 3 A schematic diagram showing a complaint data processing method provided in an embodiment of the present application.

[0012] Figure 4 A flowchart of a method for constructing a complaint location model provided in an embodiment of the present application is shown.

[0013] Figure 5 A flowchart of a complaint data processing method provided in an embodiment of the present application is shown.

[0014] Figure 6 A block diagram showing the composition of a complaint data processing device provided in an embodiment of the present application is shown.

[0015] Figure 7 A block diagram showing the composition of an electronic device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the present application more clear, the embodiments of the present application will be described in detail with reference to the accompanying drawings. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other arbitrarily without conflict.

[0017] At present, the processing of demarcation and positioning of online complaints from mobile communication network users mainly relies on two methods: manual processing and expert experience. These two methods have at least the following problems: 1) Complexity: The scenarios involving various problems have different causes and solutions for each problem, which increases the complexity of positioning. 2) Data volume and speed: A large number of user complaints need to be processed quickly and efficiently, and the data processing pressure is relatively high. 3) Adaptability to new problems: The continuous development of the business may lead to the emergence of new problems, and the positioning method needs to be adjusted in time to deal with new problems. 4) Subjectivity: Manual processing is easily affected by subjective judgment, which may lead to inconsistent processing results. 5) Accuracy: Accuracy cannot be guaranteed.

[0018] In view of this, the present application provides a complaint data processing method, an electronic device and a readable storage medium to solve the above problems.

[0019] In an embodiment of the present application, based on the complaint data to be processed corresponding to the user information and the network element information, the comprehensive complaint features corresponding to the user information and the network element information are determined; the comprehensive complaint features are input into the complaint location model of the business domain corresponding to the complaint data to be processed, and the complaint location results corresponding to the user information and the network element information are obtained. It can be seen that in the present application, the correlation between the business domain and the corresponding features is fully considered. Therefore, a corresponding complaint location model is set for each business domain. When handling complaints, the extracted comprehensive complaint features are input into the complaint location model of the corresponding business domain for processing, thereby improving the processing capability of the comprehensive complaint features and improving the processing accuracy and efficiency of the complaint data.

[0020] In a first aspect, an embodiment of the present application provides a method for processing complaint data.

[0021] Figure 1 A flowchart of a method for processing complaint data provided by an embodiment of the present application is shown. Figure 1 As shown, the complaint data processing method in the embodiment of the present application includes but is not limited to the following steps.

[0022] Step S101, determining the comprehensive complaint features corresponding to the user information and the network element information according to the complaint data to be processed corresponding to the user information and the network element information.

[0023] In some optional implementations, the user information and network element information are information used to determine the user and network element (or network element group) corresponding to the complaint data to be processed.

[0024] Exemplarily, the user information includes an International Mobile Subscriber Identity (IMSI), and the network element information includes a network element identifier (ID).

[0025] In some optional implementations, the complaint data to be processed is data determined based on the user's complaint behavior. The complaint data to be processed is processed accordingly to determine the complaint location result, clarify the user's complaint cause and the problem that caused the complaint (for example, network coverage problem, terminal configuration problem, etc.).

[0026] Exemplarily, the complaint data to be processed may include a plurality of indicator data. For example, the indicator data includes some key performance indicator (Key Quality Indicator, KQI) data generated during the user's mobile communication process.

[0027] For example, some KQI data in the Global System for Mobile Communications (GSM) include: signal strength good rate, end-to-end voice service connection rate, end-to-end voice service connection time, end-to-end voice service drop rate, single-talk ratio (single-talk means only the caller or only the called party can hear the voice), crosstalk ratio (crosstalk means hearing a third party's voice during a call between the caller and the called party), echo ratio (echo means that the user hears his own voice during a call), etc.

[0028] It should be noted that, in some optional implementations, the user's complaint request carries user information, network element information, and corresponding complaint content (for example, indicator data), based on which the complaint data to be processed corresponding to the user information and network element information can be clearly identified; in some optional implementations, the user's complaint request does not directly carry user information and network element information, therefore, the user information and network element information corresponding to the complaint data to be processed can be determined by information query and other methods (for example, locating the user and network element through the identifier of the complaint request, and querying the user information and network element information from a preset database). The embodiments of the present application are not limited to this.

[0029] In some optional implementations, determining comprehensive complaint features corresponding to the user information and network element information based on the complaint data to be processed corresponding to the user information and network element information may include: extracting features of the complaint data to be processed based on preset feature dimensions, and determining feature parameters corresponding to the user information and network element information; and determining comprehensive complaint features corresponding to the user information and network element information based on the feature parameters.

[0030] In some optional implementations, the characteristic parameter may include at least one of a mean, a standard deviation, a kurtosis, a skewness, and an autocorrelation of the time series.

[0031] For example, the preset feature dimensions include the delay dimension, and the complaint data to be processed includes call delay index data. Then, the mean calculation, standard deviation calculation, kurtosis calculation, skewness calculation, time series autocorrelation calculation, etc. can be performed based on the call delay index data to obtain the delay mean, delay standard deviation, delay kurtosis, delay skewness, and delay time series, and the above-mentioned feature parameters are used as the comprehensive complaint features corresponding to the user information and network element information.

[0032] In some optional implementations, after obtaining the characteristic parameters such as delay mean, delay standard deviation, delay kurtosis, delay skewness, autocorrelation of delay time series, etc. through mean calculation, standard deviation calculation, kurtosis calculation, skewness calculation, and autocorrelation calculation of time series, these characteristic parameters can also be normalized to obtain more accurate and standardized comprehensive complaint characteristics.

[0033] In some optional implementations, after obtaining the characteristic parameters of the delay mean, delay standard deviation, delay kurtosis, delay skewness, autocorrelation of the delay time series, etc. for the above-mentioned complaint data to be processed, these characteristic parameters can also be encoded based on a preset encoding method, and the generated code can be used as the corresponding comprehensive complaint feature.

[0034] In some optional implementations, feature extraction is performed on the complaint data to be processed based on preset feature dimensions to determine feature parameters corresponding to the user information and network element information, including: when there is no indicator data corresponding to the preset feature dimension in the complaint data to be processed, feature fusion of indicator data of at least two different indicator types of the complaint data to be processed according to the preset feature dimension to obtain comprehensive indicator data; and extracting feature parameters of the comprehensive indicator data as feature parameters corresponding to the user information and network element information.

[0035] From this, we can see that by fusing different types of indicator data through cross-pollination, combination, splicing, etc., new indicator data can be obtained. The new indicator data can reflect new features. Therefore, the utilization rate of complaint data to be processed can be improved, and more accurate and reasonable complaint location results can be obtained.

[0036] For example, indicator data includes system throughput and user response time. Throughput can only determine the number of requests processed by the system per unit time, and response time can only determine the time it takes for the system to respond to user requests. Combining the two (for example, calculating the average response time per user through throughput and response time) can reflect whether the system's resource allocation is reasonable and balanced.

[0037] In other words, by combining different indicator data, new indicator data can be obtained, and the new indicator data may reflect new features. This new feature enriches the data sources involved in data processing and can obtain better processing results.

[0038] In summary, new features can be constructed through feature interaction and combination, and the correlation between different features can be extracted and expressed, thereby revealing more information and potential patterns of complaint data, which is conducive to improving the expressive ability and prediction accuracy of the model.

[0039] In some optional implementations, considering the differences in business attributes of a certain business, it can be divided into multiple business domains, and then complaint location can be performed based on the business domain, which can achieve complaint location quickly and efficiently.

[0040] In some optional implementations, for a certain service, according to its service attributes, the service can be divided into multiple service domains, and different service domains have different service scopes. For example, for a mobile communication service, its corresponding service domains may include: user service domain, wireless service domain, core network service domain, service provider (SP) service domain and bearer service domain.

[0041] Accordingly, when processing complaint data, the corresponding features of the complaint data to be processed can be extracted for each business domain respectively, so as to facilitate the determination of the corresponding complaint location results based on the comprehensive characteristics of complaints in different business domains, thereby making the complaint location more accurate, detailed and comprehensive, and facilitating more targeted complaint processing.

[0042] In some optional implementations, before determining the comprehensive complaint characteristics corresponding to the user information and the network element information based on the complaint data to be processed corresponding to the user information and the network element information, the method also includes: obtaining initial complaint data to be processed, the initial complaint data to be processed including multiple indicator data; dividing the initial complaint data to be processed into corresponding business domains according to the indicator types of the multiple indicator data, and obtaining domain data corresponding to the business domains; for each domain data, dividing the domain data into multiple data groups according to the user information and the network element information corresponding to the domain data, each data group corresponding to the complaint data to be processed with the same user information and the same network element information.

[0043] Exemplarily, the initial complaint data to be processed includes multiple data columns, which respectively correspond to user information, network element information, indicator data, etc. First, according to the indicator type of the indicator data, the initial complaint data to be processed can be divided into corresponding business domains to obtain multiple domain data, each of which may correspond to multiple users and multiple network elements. Therefore, each domain data can also be divided into multiple data groups according to user information and network element information, so that the complaint data to be processed in each data group corresponds to one user and one network element, which is convenient for complaint location and delimitation.

[0044] For example, Table 1 shows a certain domain data, which includes three data columns, corresponding to user information, network element information and indicator data respectively.

[0045] Table 1 Domain data

[0046] User Information Network element information Indicator data user1 id1 KQI11 user2 id3 KQI31 user1 id2 KQI21 user4 id3 KQI32 user3 id1 KQI12 user2 id3 KQI33 user1 id1 KQI13

[0047] After the domain data in Table 1 are grouped based on the user information and the network element information, multiple data groups as shown in Table 2 can be obtained.

[0048] Table 2 Data set

[0049] Data Group User Information Network element information Indicator data Data set 1 user1 id1 KQI11, KQI13 Data set 2 user1 id2 KQI21 Data set 3 user2 id3 KQI31, KQI33 Data set 4 user3 id1 KQI12 Data set 5 user4 id3 KQI32

[0050] As shown in Table 2, each data group corresponds to a user information, a network element information, and one or more indicator data. For example, data group 1 is the pending complaint data about user user1 and network element id1 (including KQI11 and KQI13); data group 2 is the pending complaint data about user user1 and network element id2 (including KQI21); data group 3 is the pending complaint data about user user2 and network element id3 (including KQI31 and KQI33); data group 4 is the pending complaint data about user user3 and network element id1 (including KQI12); data group 5 is the pending complaint data about user user4 and network element id3 (including KQI32).

[0051] In some optional implementations, after obtaining the initial complaint data to be processed, the method further includes: performing data preprocessing on the initial complaint data to be processed, wherein the data preprocessing includes filling missing values ​​and / or removing abnormal values. For the initial complaint data to be processed after data preprocessing, the business domain division, data group division, etc. can be performed.

[0052] In some optional implementations, missing value filling can be implemented based on the upper and lower value filling method of adjacent values, thereby effectively solving the data discontinuity caused by missing data. Among them, the upper and lower value filling method based on adjacent values ​​can make full use of the information of time series data, reduce the deviation caused by filling, and also consider the trend and periodicity of the data. In addition, compared with other filling methods, the upper and lower value filling method based on adjacent values ​​relies on adjacent values ​​and data change trends, does not require additional calculations and iterative processes, and has a faster filling speed and higher efficiency.

[0053] It should be noted that the above method for filling missing values ​​is only an example and the embodiments of the present application do not limit this.

[0054] Exemplarily, the initial complaint data to be processed includes KQI data, as well as corresponding time information, network element identifier (ID), International Mobile Subscriber Identity (IMSI) and other necessary data.

[0055] First, some data preprocessing can be performed on the above-mentioned initial complaint data to be processed, including but not limited to the removal of illegal fields, filling in missing items, etc., so as to obtain new initial complaint data to be processed, including IMSI, network element id and some KQI data. Further, according to the indicator type of KQI data, the initial data to be processed is divided into corresponding business domains to obtain domain data of each business domain. For each domain data, grouping operation is performed according to its network element id and IMSI, so that each data group corresponds to one user and one network element (or one network element group).

[0056] Step S102: input the comprehensive complaint features into the complaint location model of the service domain corresponding to the complaint data to be processed, and obtain the complaint location result corresponding to the user information and the network element information.

[0057] In some optional implementations, the complaint location model has a corresponding relationship with the business domain. For example, a corresponding complaint location model can be set for each business domain. After the complaint comprehensive features are input into the corresponding complaint location model, a corresponding complaint location result can be obtained after a series of data processing and analysis.

[0058] In some optional implementations, the complaint location model is constructed based on a decision tree algorithm; moreover, the complaint location result includes the cause of the complaint, and the cause of the complaint has a location level, wherein the location level is used to characterize the size of the location range corresponding to the cause of the complaint.

[0059] In other words, the complaint location model is essentially a decision tree model. The complaint location result can not only reflect the cause of the complaint, but also reflect the size of the corresponding location range. For example, if the complaint location result is poor network coverage, the location level can be low 4G (4-Generation) network coverage, which further clarifies that there is a problem with the 4G network coverage.

[0060] In some optional implementations, the complaint location model is a model based on a decision tree for determining the complaint location result based on the comprehensive characteristics of the complaint. Among them, the decision tree is a model for decision-making based on a tree structure, which classifies the data set through multiple conditional discrimination processes and finally obtains the required results. Usually, the starting point of the decision tree is the root node, the intermediate decision process is the internal node, and the classification result is the leaf node.

[0061] In some optional implementations, the leaf nodes of the complaint location model may correspond to some complaint reason labels. After the comprehensive complaint features are input into the complaint location model, the corresponding leaf nodes can be finally determined through the judgment and processing of the internal nodes, and the complaint location results can be determined through the complaint reason labels corresponding to the leaf nodes.

[0062] In summary, the decision tree algorithm can effectively classify and divide a large amount of complaint data, thereby improving the accuracy of locating problems. Through the learning and fitting of decision trees, the correspondence between user complaints and problems can be better matched, the situation of incorrect positioning can be reduced, and the accuracy can be improved. In addition, the decision tree algorithm has good adaptability and can cope with the emergence of new problems. For new complaint problems, complaints can be handled by updating the decision tree model without large-scale system changes. Moreover, the decision tree is an algorithm with strong interpretability. Therefore, it can show the process of complaint classification and positioning more clearly, which is convenient for operators and other business personnel to understand the cause of the problem and develop a more appropriate solution.

[0063] In addition, the decision tree method has the ability to automatically learn and generate rules without human intervention. Therefore, this automated feature helps to reduce the burden of manual processing, improve processing efficiency, and effectively reduce the impact of errors that may be caused by human subjective judgment. Furthermore, as an efficient data processing algorithm, the decision tree can classify and locate complaint data in a short time, respond to user complaints more quickly, and improve user satisfaction and experience.

[0064] In some optional implementations, the complaint location model can be a model built using a lightweight gradient boosting machine learning algorithm (Light Gradient Boosting Machine, LightGBM). The basic principle of LightGBM is the histogram algorithm, where the essence of the histogram algorithm is bucket sorting, and the feature segmentation points are determined by bucket sorting. In addition, the decision tree algorithm based on LightGBM can process large-scale data, and can effectively handle nonlinear relationships, while having the functions of feature selection and importance evaluation. Therefore, based on the above characteristics, it enables it to quickly and accurately locate the root causes of problems in the communication business field and improve network quality and user experience.

[0065] In related technologies, when performing feature point segmentation, it is usually necessary to pre-sort the original data of each feature, while the histogram algorithm "buckets" the original data of the feature, divides the data into different discrete areas, and then traverses the discrete data to find the optimal division point. Among them, each "bucket" divided for the feature value has two meanings, one is the number of samples in each bucket, and the other is the gradient sum of the samples in each bucket.

[0066] In the embodiment of the present application, in the complaint location model based on LightGBM, the histogram algorithm can be used to obtain the suboptimal solution of the segmented features and the splitting points. For example, for continuous features, k buckets can be determined first, and then the feature values ​​are placed in the corresponding buckets to construct a histogram with a width of k, and the values ​​of each bucket are counted, and the optimal splitting point is found by traversing the above k feature values; for discrete features, the first-order gradient sum of the category samples (or the second-order gradient sum of the category samples) and the regularization term (ie, the accumulated value) can be sorted before constructing the histogram.

[0067] In summary, in the complaint location model based on LightGBM, the model complexity is reduced, and only discrete values ​​are saved after feature bucketing, which can effectively reduce memory occupancy. In addition, this bucketing method is equivalent to adding a regularization process to the model (the number of buckets k determines the intensity of the regularization penalty), so it can avoid overfitting of the model and improve the robustness of the model.

[0068] It should be noted that the above complaint location model is only an example. Any complaint location model constructed based on the decision tree and its variant algorithm can be used in the complaint data processing method of the embodiment of the present application, and the embodiment of the present application does not limit this.

[0069] In an embodiment of the present application, based on the complaint data to be processed corresponding to the user information and the network element information, the comprehensive complaint features corresponding to the user information and the network element information are determined; the comprehensive complaint features are input into the complaint location model of the business domain corresponding to the complaint data to be processed, and the complaint location results corresponding to the user information and the network element information are obtained. It can be seen that in the present application, the correlation between the business domain and the corresponding features is fully considered. Therefore, a corresponding complaint location model is set for each business domain. When handling complaints, the extracted comprehensive complaint features are input into the complaint location model of the corresponding business domain for processing, thereby improving the processing capability of the comprehensive complaint features and improving the processing accuracy and efficiency of the complaint data.

[0070] Figure 2 A flowchart of a method for processing complaint data provided by an embodiment of the present application is shown. Figure 2 As shown, the complaint data processing method in the embodiment of the present application includes but is not limited to the following steps.

[0071] Step S201, obtaining initial complaint data to be processed, and dividing the initial complaint data to be processed into corresponding business domains according to the indicator types of multiple indicator data in the initial complaint data to be processed, to obtain domain data corresponding to the business domains.

[0072] Among them, each domain data may correspond to multiple user information and multiple network element information, and it is necessary to further divide it into complaint data to be processed for each user information and each network element information.

[0073] Step S202: grouping the domain data according to the user information and network element information corresponding to the domain data to obtain at least one data group.

[0074] Among them, the data group includes at least one data entry, and since one data group corresponds to the same user information and the same network element information, each data group is equivalent to a "pending complaint data corresponding to the user information and network element information".

[0075] For example, the domain data includes several indicator data of user A about network element 1, several indicator data of user B about network element 1, and several indicator data of user A about network element 2. Then, based on the user information and network element information, the domain data needs to be divided into three data groups. The first data group includes the indicator data of user A about network element 1, the second data group includes the indicator data of user B about network element 1, and the third data group includes the indicator data of user A about network element 2.

[0076] Step S203, extracting features from the data group to obtain comprehensive complaint features corresponding to the data group.

[0077] Among them, feature extraction is performed on the data group, which is equivalent to feature extraction of the pending complaint data corresponding to a certain user regarding a certain network element. The obtained comprehensive complaint features of the data group are essentially the comprehensive complaint features corresponding to the user and the network element.

[0078] For example, feature extraction is performed on the first data group, and the comprehensive complaint features of the first data group are obtained. In fact, they are the comprehensive complaint features of user A regarding network element 1. Similarly, feature extraction is performed on the second data group, and the comprehensive complaint features of the second data group are obtained. In fact, they are the comprehensive complaint features of user B regarding network element 1. Feature extraction is performed on the third data group, and the comprehensive complaint features of the third data group are obtained. In fact, they are the comprehensive complaint features of user A regarding network element 2.

[0079] Step S204: input the comprehensive complaint features into the complaint location model of the corresponding business domain to obtain the complaint location results corresponding to the user information and the network element information.

[0080] Figure 3 A schematic diagram of a method for processing complaint data provided by an embodiment of the present application is shown. Figure 3 As shown in , after obtaining the initial complaint data to be processed, the initial complaint data to be processed can be preprocessed by removing illegal fields, filling in missing items, etc., to improve the data quality of the initial complaint data to be processed. Furthermore, according to the indicator type of the indicator data in the initial data to be processed, the domain data corresponding to each business domain can be determined. Figure 3 As shown, domain data 1 corresponding to business domain 1, ..., domain data j corresponding to business domain j can be obtained.

[0081] It should be noted that there may be some service domains that do not have corresponding domain data, which indicates that there is no complaint data to be processed in these service domains and no subsequent data processing is required. For example, if there is no domain data in the core network service domain and the bearer service domain for the complaint data to be processed, it means that these two service domains can provide services to users normally and users have not filed any complaints against these two service domains.

[0082] For the domain data of any service domain, the domain data can be grouped according to the user information and network element information corresponding to the domain data, and each data group obtained corresponds to one user and one network element group. For example, for domain data 1, after data grouping and screening, n data groups are obtained (n is an integer greater than or equal to 1), namely data group 11, data group 12, ..., data group 1n; similarly, for domain data j, m data groups can be obtained, namely data group j1, data group j2, ..., data group jm; wherein n and m are both integers greater than or equal to 1, and n and m can be the same or different.

[0083] After obtaining the data group of each domain data, feature extraction can be performed on the data group, so as to obtain at least one comprehensive complaint feature corresponding to each data group. For example, for data group 1, feature extraction is performed on the complaint data to be processed in data group 1 based on the preset feature dimension, and the comprehensive complaint feature of data group 1 can be obtained. Similarly, the comprehensive complaint feature of each data group can be obtained.

[0084] It should be noted that for each data group, the number of comprehensive complaint features obtained can be one or more, and the embodiments of the present application do not limit this.

[0085] Furthermore, a corresponding complaint location model is pre-set for each business domain. For example, complaint location model 1 is set for business domain 1, ..., and complaint location model j is set for business domain j, where j is the identifier of the complaint location model and j is an integer greater than or equal to 1. Therefore, for data groups 11 to 1n corresponding to business domain 1, the corresponding complaint comprehensive feature 11, complaint comprehensive feature 12, ..., complaint comprehensive feature 1p are input into complaint location model 1, and the complaint location results for each data group can be obtained. Similarly, complaint comprehensive feature j1, complaint comprehensive feature j2, ..., complaint comprehensive feature jk (i.e., the kth complaint comprehensive feature of business domain j) are input into complaint location model j, and the complaint location results for each data group can be obtained.

[0086] In some optional implementations, the complaint location result may include a corresponding complaint reason label, a problem object, and a confidence level, wherein the problem object is the specific network element or device being complained about, and the confidence level is used to characterize the reliability or certainty of the complaint reason label.

[0087] Furthermore, after obtaining multiple complaint location results, relevant technical personnel can perform fault analysis and other processing based on the multiple complaint location results, clarify the reasons why the network element or equipment was complained about, and make corresponding adjustments and upgrades to the network element or equipment to reduce the occurrence of related complaints.

[0088] Figure 4 A flowchart of a method for constructing a complaint location model provided in an embodiment of the present application is shown. Figure 4 As shown, the construction method includes but is not limited to the following steps.

[0089] Step S401, obtaining complaint training set data of multiple business domains and complaint labels corresponding to the complaint training set data.

[0090] In some optional implementations, the complaint training set data of each service domain may include some KQI indicator data corresponding to user information and network element information, and the complaint label is used to characterize the complaint cause of the corresponding KQI indicator data.

[0091] Step S402: extract features from the complaint training set data to obtain comprehensive training features for each business domain.

[0092] In some optional implementations, feature extraction is performed on the complaint training set data to obtain training comprehensive features of the business domain, including: dividing the complaint data in the complaint training set data into multiple data groups based on user information and network element information, so that each data group corresponds to the same user information and the same network element information; performing feature extraction on the data groups to obtain grouped comprehensive features of the data groups; and obtaining the training comprehensive features of the business domain based on the grouped comprehensive features of multiple data groups.

[0093] In some optional implementations, the complaint data can be divided into multiple data groups according to user information and network element information, each data group corresponds to one user and one network element group, and then the data groups with fewer data entries are deleted, and only the data groups with more data entries are retained to extract features from the retained data groups. In this way, data groups with insufficient data volume for accurate analysis can be excluded, while data groups with sufficient data volume can be retained as valid samples for subsequent analysis.

[0094] In some optional implementations, extracting features from a data group to obtain grouped comprehensive features of the data group includes: determining feature parameters of the data group based on a preset feature dimension; and determining grouped comprehensive features of the data group based on the feature parameters of the data group. The feature parameters include at least one of mean, standard deviation, kurtosis, skewness, and autocorrelation of the time series.

[0095] Step S403, constructing a complaint location model for each business domain based on the training comprehensive features and complaint labels.

[0096] In some optional implementations, the complaint location model has a tree structure, including a root node, multiple intermediate nodes, and multiple leaf nodes. The model construction process is actually also a feature selection process and a feature parameter determination process, that is, for multiple training comprehensive features, it is necessary to select the root node, the next level node of the root node, etc., until the leaf node is determined, and the characteristic value classification position of each node is determined, so as to obtain the final complaint location model. Among them, when selecting the node of the current level from multiple indicator features, the selection can be made by information gain, Gini impurity, etc., and the embodiment of the present application does not limit this.

[0097] Exemplarily, the construction process of the complaint location model for a certain business domain may include the following: first, for multiple training comprehensive features, find the one with the largest benefit after classification as the root node, then based on the remaining training indicator features, find the one with the largest benefit after classification as the next level node of the root node, and then further split it, and so on, and finally grow into a complaint decision tree with a tree structure. Among them, for each node that needs to be classified, it is necessary to find the best classification position, that is, it is necessary to find the best split feature value of the current node according to the indicator data of the current node to split it to form the left node and right node of the next level.

[0098] For example, in the process of building a complaint location model based on LightGBM, when determining the classification position, the optimal split feature value is determined by the histogram method to perform node splitting.

[0099] It should be noted that since the indicator data of mobile communication services are usually large-scale, the histogram-based decision tree learning method in the LightGBM algorithm can efficiently process large amounts of data, has a high training speed and low memory consumption, and the LightGBM algorithm can capture the complex nonlinear relationships in communication problems, thereby generating a complaint location model with strong fitting capabilities; secondly, since the complaint location model processes data through loss function optimization and gradient boosting methods, the complaint location results obtained based on the indicator data have high accuracy; in addition, the complaint location model is equivalent to having a certain feature selection capability and performance evaluation function built in when determining the node, so for a large amount of indicator data, key features can be selected as nodes of the corresponding level, which improves the model generalization ability. In summary, the complaint location model based on LightGBM has many advantages in the demarcation and location of complaints in mobile network services, and can quickly and accurately locate the source of the problem, which is conducive to improving network quality and user experience.

[0100] Figure 5 A flowchart of a method for processing complaint data provided by an embodiment of the present application is shown in FIG. Figure 5 As shown, the method includes but is not limited to the following steps.

[0101] Step S501, obtaining historical complaint data with complaint tags.

[0102] For example, historical complaint data of user service domain, wireless service domain, core network service domain, service provider service domain and bearer service domain with complaint labels are obtained, which are Duser, Dci, Dcn, Dapp and Dbn respectively. The historical complaint data of each service domain includes the corresponding KQI data and necessary data such as time information, network element id, IMSI, etc.

[0103] Step S502, preprocessing the historical complaint data, including but not limited to removing illegal fields and filling in missing items, to obtain new historical complaint data.

[0104] Step S503: performing a grouping operation on the historical complaint data of each service domain based on the user information and the network element information to obtain at least one data group corresponding to each service domain.

[0105] Step S504, deleting the data group whose data entries are smaller than the preset value, and retaining the data group whose data entries are greater than or equal to the preset value.

[0106] For example, for each service domain, grouping operation is performed based on network element id and IMSI as two data columns, each obtained data group corresponds to a network element id and an IMSI, and data groups with less than 10 data entries are deleted, and only data groups Di with greater than or equal to 10 data entries are retained.

[0107] Step S505, extracting characteristic parameters from the data group based on preset characteristic dimensions, and determining grouping comprehensive characteristics of the data group based on the characteristic parameters.

[0108] For example, for each data group, the mean, standard deviation, kurtosis, skewness, time series autocorrelation, etc. of the KQI indicator data can be extracted as the grouping comprehensive features of the data group. In addition, a comprehensive feature construction method can be used to construct new indicator data based on existing KQI indicator data and extract the features of the new indicator data.

[0109] Step S506, obtaining the training comprehensive features of the corresponding business domain according to the group comprehensive features belonging to the same business domain.

[0110] Step S507, based on the LightGBM algorithm, training is performed according to the training comprehensive features and complaint labels of each business domain to build a complaint location model corresponding to each business domain.

[0111] For example, for the above user service domain, wireless service domain, core network service domain, service provider service domain and bearer service domain, the corresponding complaint location models can be obtained: Modeluser, Modelci, Modelcn, Modelapp and Modelbn. After obtaining the complaint location model of each service domain, the complaint location model can be used to process the complaint data.

[0112] Step S508, obtaining initial complaint data to be processed.

[0113] For example, the complaint data to be processed may include KQI indicator data of any one or more business domains and their corresponding necessary data. It should be noted that the complaint data to be processed does not contain complaint labels, and it is necessary to determine the complaint labels matching the complaint data to be processed through subsequent processing to determine the complaint location results.

[0114] Step S509: grouping the initial complaint data to be processed according to the user information and the network element information to obtain multiple data groups.

[0115] Step S510, extracting characteristic parameters of each data group, and determining comprehensive complaint characteristics of each data group based on the characteristic parameters.

[0116] Step S511, input the comprehensive complaint features into the complaint location model of the corresponding business domain to obtain the corresponding complaint location results.

[0117] Step S512: Analyze the complaint location results and determine the complaint handling plan.

[0118] In a second aspect, an embodiment of the present application provides a complaint data processing device.

[0119] Figure 6 FIG. 1 is a block diagram showing a complaint data processing device provided by an embodiment of the present application. Figure 6 As shown, the complaint data processing device 600 in the embodiment of the present application includes but is not limited to the following modules.

[0120] The determination module 601 is used to determine the comprehensive complaint features corresponding to the user information and the network element information according to the complaint data to be processed corresponding to the user information and the network element information;

[0121] The processing module 602 is used to input the comprehensive complaint features into the complaint location model of the business domain corresponding to the complaint data to be processed, and obtain the complaint location result corresponding to the user information and the network element information.

[0122] In an embodiment of the present application, a determination module is used to determine the complaint comprehensive features corresponding to the user information and the network element information based on the complaint data to be processed corresponding to the user information and the network element information; and a processing module is used to input the complaint comprehensive features into the complaint location model of the business domain corresponding to the complaint data to be processed, so as to obtain the complaint location results corresponding to the user information and the network element information. It can be seen that in the present application, the correlation between the business domain and the corresponding features is fully considered. Therefore, a corresponding complaint location model is set for each business domain. When handling complaints, the extracted complaint comprehensive features are input into the complaint location model of the corresponding business domain for processing, thereby improving the processing capability of the complaint comprehensive features and improving the processing accuracy and efficiency of the complaint data.

[0123] It should be clear that the present invention is not limited to the specific configurations and processes described in the above embodiments and shown in the figures. For the convenience and brevity of description, a detailed description of the known methods is omitted here, and the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, which will not be repeated here.

[0124] In a third aspect, an embodiment of the present application provides an electronic device.

[0125] Figure 7 FIG. 1 is a block diagram showing the composition of an electronic device provided in an embodiment of the present application. Figure 7 As shown, the electronic device 700 includes an input device 701, an input interface 702, a central processing unit 703, a memory 704, an output interface 705, an output device 706, and a bus 707. The input interface 702, the central processing unit 703, the memory 704, and the output interface 705 are connected to each other through the bus 707, and the input device 701 and the output device 706 are connected to the bus 707 through the input interface 702 and the output interface 705, respectively, and then connected to other components of the electronic device 700.

[0126] Specifically, the input device 701 receives input information from the outside and transmits the input information to the central processing unit 703 through the input interface 702; the central processing unit 703 processes the input information based on the computer executable instructions stored in the memory 704 to generate output information, temporarily or permanently stores the output information in the memory 704, and then transmits the output information to the output device 706 through the output interface 705; the output device 706 outputs the output information to the outside of the electronic device 700 for user use.

[0127] In one embodiment, Figure 7 The electronic device shown may include: a memory configured to store a program; and a processor configured to run the program stored in the memory to execute the complaint data processing method described in the above embodiment.

[0128] In one embodiment, Figure 7 The electronic device shown can be implemented as a complaint data processing device, which may include: a memory configured to store a program; a processor configured to run the program stored in the memory to execute the complaint data processing method described in the above embodiment.

[0129] In one example, an embodiment of the present application further provides a readable storage medium, which stores a computer program. When the computer program is executed by a processor, any one of the complaint data processing methods in the embodiments of the present application is implemented.

[0130] The above is only an exemplary embodiment of the present disclosure and is not intended to limit the scope of protection of the present disclosure. In general, various embodiments of the present disclosure can be implemented in hardware or dedicated circuits, software, logic or any combination thereof. For example, some aspects can be implemented in hardware, while other aspects can be implemented in firmware or software that can be executed by a controller, microprocessor or other computing device, although the present disclosure is not limited thereto.

[0131] Embodiments of the present disclosure may be implemented by a processor executing computer program instructions, such as in a processor entity, or by hardware, or by a combination of software and hardware. The computer program instructions may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages.

[0132] The block diagram of any logic flow in the drawings of the present disclosure may represent program steps, or may represent interconnected logic circuits, modules and functions, or may represent a combination of program steps and logic circuits, modules and functions. The computer program may be stored on a memory. The memory may have any type suitable for the local technical environment and may be implemented using any suitable data storage technology, such as but not limited to read-only memory (ROM), random access memory (RAM), optical memory devices and systems (digital versatile discs DVD or CD discs), etc. Computer-readable media may include non-transient storage media. The processor may be any type suitable for the local technical environment, such as but not limited to a general-purpose computer, a special-purpose computer, a microprocessor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a programmable logic device (FGPA) and a processor based on a multi-core processor architecture.

[0133] By way of exemplary and non-limiting examples, a detailed description of exemplary embodiments of the present disclosure has been provided above. However, various modifications and adjustments to the above embodiments will be apparent to those skilled in the art when considered in conjunction with the accompanying drawings and claims, but will not depart from the scope of the present disclosure. Therefore, the proper scope of the present disclosure will be determined according to the claims.

Claims

1. A method for processing complaint data, characterized in that: The method comprises: Determine, according to the complaint data to be processed corresponding to the user information and the network element information, the comprehensive features of the complaints corresponding to the user information and the network element information; The comprehensive complaint features are input into a complaint location model of the business domain corresponding to the complaint data to be processed, and a complaint location result corresponding to the user information and the network element information is obtained.

2. The method according to claim 1, characterized in that Before determining the complaint comprehensive features corresponding to the user information and the network element information based on the complaint data to be processed corresponding to the user information and the network element information, the method further includes: Acquire initial complaint data to be processed, wherein the initial complaint data to be processed includes a plurality of indicator data; According to the indicator types of the plurality of indicator data, the initial complaint data to be processed is divided into corresponding business domains to obtain domain data corresponding to the business domains; For each of the domain data, the domain data is divided into a plurality of data groups according to the user information and network element information corresponding to the domain data, and each of the data groups corresponds to the complaint data to be processed with the same user information and the same network element information.

3. The method according to claim 2, characterized in that After obtaining the initial complaint data to be processed, the method further includes: The initial complaint data to be processed is subjected to data preprocessing, wherein the data preprocessing includes filling in missing values ​​and / or removing outliers.

4. The method according to any one of claims 1 to 3, characterized in that: The determining, based on the complaint data to be processed corresponding to the user information and the network element information, the comprehensive complaint features corresponding to the user information and the network element information includes: Extract features of the complaint data to be processed based on preset feature dimensions, and determine feature parameters corresponding to the user information and the network element information; According to the characteristic parameters, comprehensive complaint characteristics corresponding to the user information and the network element information are determined.

5. The method according to claim 4, characterized in that The extracting features of the complaint data to be processed based on the preset feature dimension to determine feature parameters corresponding to the user information and the network element information includes: In the case that there is no indicator data corresponding to the preset feature dimension in the complaint data to be processed, characteristic fusion is performed on the indicator data of at least two different indicator types of the complaint data to be processed according to the preset feature dimension to obtain comprehensive indicator data; Extract characteristic parameters of the comprehensive indicator data as characteristic parameters corresponding to the user information and the network element information.

6. The method according to claim 4, characterized in that The characteristic parameters include at least one of mean, standard deviation, kurtosis, skewness, and autocorrelation of the time series.

7. The method according to claim 1, characterized in that The complaint location model is constructed based on a decision tree algorithm; The complaint location result includes a complaint reason, and the complaint reason has a location level, wherein the location level is used to characterize the size of a location range corresponding to the complaint reason.

8. The method according to claim 7, characterized in that The decision tree algorithm includes a lightweight gradient boosting machine learning algorithm.

9. An electronic device, characterized in that: include: one or more processors; A memory having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the complaint data processing method as described in any one of claims 1 to 8.

10. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and when the computer program is executed by a processor, the complaint data processing method as described in any one of claims 1 to 8 is implemented.

Images