Method, device, equipment, medium and product for determining target feature parameter

Abstract

The application discloses a target feature parameter determination method, device, equipment, medium and product, wherein the method comprises: obtaining a plurality of historical 5G user samples; for each historical 5G user sample, respectively performing: respectively performing dimensionality reduction and normalization processing on a plurality of quasi-target user feature parameters to obtain normalized feature parameters; generating a plurality of feature parameter sets according to the plurality of normalized feature parameters; for each feature parameter set, respectively performing: using a decision tree model and the feature parameters in the feature parameter set to identify whether the corresponding user is an abnormal user to obtain an identification classification result; and determining the feature parameters in the feature parameter set corresponding to the identification classification result consistent with the user category label as target feature parameters. According to the embodiments of the application, the target feature parameters required for identifying abnormal users can be determined, so that the abnormal users can be more accurately identified, and the efficiency of identifying abnormal users can be ensured.

Description

Technical Field

[0001] This application belongs to the field of big data, and in particular relates to a method, apparatus, equipment, medium and product for determining target feature parameters. Background Technology

[0002] "Dissentful users" refer to individuals who are dissatisfied with a brand, product, or service and are unwilling to recommend the company's products or services to others, thereby reducing the company's sales and profits, damaging its brand image, and allowing competitors to profit. Therefore, in the 5G+AICDE mobile internet era, identifying dissentful users and implementing targeted improvements to service experience and quality has always been a key focus of network optimization efforts.

[0003] In existing technologies, abnormal user prediction or identification models are mostly built based on user package data. The output results only cover users who are dissatisfied with their tariffs, which is biased towards marketing support and has low accuracy and efficiency in identifying users. Summary of the Invention

[0004] This application provides a method, apparatus, device, storage medium, and computer program product for determining target feature parameters, which can determine the target feature parameters required for identifying abnormal users, more accurately identify abnormal users, and ensure the efficiency of identifying abnormal users.

[0005] Firstly, a method for determining target feature parameters is provided, including:

[0006] Multiple historical 5G user samples are obtained. Each historical 5G user sample includes a user category label and multiple quasi-target user feature parameters. The quasi-target user feature parameters are user feature parameters belonging to the OBM domain.

[0007] Perform the following for each historical 5G user sample:

[0008] The feature parameters of multiple quasi-target users are reduced in dimension and normalized respectively to obtain the normalized feature parameters;

[0009] Based on multiple normalized feature parameters, multiple feature parameter sets are generated, and the feature parameters in each feature parameter set are not completely the same;

[0010] For each set of feature parameters, the following steps are performed: using the decision tree model and the feature parameters in the set of feature parameters, the corresponding user is identified as an abnormal user, and the identification and classification results are obtained.

[0011] The feature parameters in the feature parameter set corresponding to the user category label are determined as target feature parameters. These target feature parameters are used to identify whether a user is an abnormal user.

[0012] Secondly, a method for identifying abnormal users is provided, including:

[0013] Obtain the first feature parameter matrix of each user sample from m historical 5G user samples and the second feature parameter matrix of each user to be identified from p users to be identified;

[0014] Normalize each first feature parameter matrix and each second feature parameter matrix respectively to obtain the normalized first feature parameter matrix and second feature parameter matrix;

[0015] Calculate the Euclidean distance between each normalized second feature parameter matrix and each normalized first feature parameter matrix;

[0016] By using the pattern layer in the probabilistic neural network model to process the Euclidean distance, an initial probability matrix is ​​obtained;

[0017] For each user to be identified, perform the following steps:

[0018] Using the summation layer of the probabilistic neural network model, the first initial probability sum of the user to be identified belonging to the abnormal user class and the second initial probability sum of the normal user class are solved respectively.

[0019] Based on the first initial probability and the second initial probability, determine the third probability that the user to be identified belongs to the abnormal user class;

[0020] If the probability of the third step is greater than a preset threshold, the user to be identified is determined to belong to the abnormal user category.

[0021] Thirdly, a device for determining target feature parameters is provided, comprising:

[0022] The first acquisition module is used to acquire multiple historical 5G user samples. Each historical 5G user sample includes a user category label and multiple quasi-target user feature parameters.

[0023] The first processing module is used to perform dimensionality reduction and normalization processing on multiple quasi-target user feature parameters respectively to obtain normalized feature parameters;

[0024] The generation module is used to generate multiple feature parameter sets based on multiple normalized feature parameters, and the feature parameters in each feature parameter set are not completely the same;

[0025] The identification module is used to perform the following for each feature parameter set: using the decision tree model and the feature parameters in the feature parameter set, it identifies whether the corresponding user is an abnormal user and obtains the identification and classification result;

[0026] The first determining module is used to determine the feature parameters in the feature parameter set corresponding to the identification and classification results that are consistent with the user category label as target feature parameters. The target feature parameters are used to identify whether the user is an abnormal user.

[0027] Fourthly, an abnormal user identification device is provided, comprising:

[0028] The second acquisition device is used to acquire the first feature parameter matrix of each user sample in m historical 5G user samples and the second feature parameter matrix of each user to be identified in p users to be identified; the feature parameters in the first feature parameter matrix and the second feature parameter matrix are determined according to the method for determining target feature parameters; m and p are positive integers;

[0029] The second processing module is used to normalize each first feature parameter matrix and each second feature parameter matrix respectively to obtain the normalized first feature parameter matrix and second feature parameter matrix.

[0030] The calculation module is used to calculate the Euclidean distance between each normalized second feature parameter matrix and each normalized first feature parameter matrix, respectively.

[0031] The third processing module is used to process the Euclidean distance using the pattern layer in the probabilistic neural network model to obtain the initial probability matrix;

[0032] The solution module is used to solve the first initial probability sum of the user to be identified belonging to the abnormal user class and the second initial probability sum of the normal user class using the summation layer of the probabilistic neural network model.

[0033] The second determining module is used to determine a third probability that the user to be identified belongs to the abnormal user class based on the first initial probability sum and the second initial probability sum;

[0034] The third determination module is used to determine that the user to be identified belongs to the abnormal user class when the third probability is greater than a preset threshold.

[0035] Fifthly, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, the computer program performing the method provided by any possible implementation of the first aspect described above.

[0036] A sixth aspect provides a computer storage medium, characterized in that the computer storage medium is subjected to a method provided by any possible implementation of the first aspect by a processor.

[0037] In a seventh aspect, a computer program product is provided, comprising a computer program, wherein the computer program, when executed by a processor, implements a method provided by any possible implementation of the first aspect described above.

[0038] Compared with existing technologies, the target feature parameter determination method, apparatus, device, medium, and product provided in this application obtain multiple historical 5G user samples, perform dimensionality reduction and normalization processing on the quasi-target user feature parameters in each of the multiple historical samples to obtain normalized feature parameters, and then use a decision tree model and feature parameters in the feature parameter set to identify whether the corresponding user is an abnormal user, obtaining an identification classification result. Finally, the feature parameters in the feature parameter set corresponding to the identification classification result consistent with the user category label are determined as target feature parameters. Therefore, the target feature parameters determined through the above process reference multi-dimensional historical user samples, more accurately identifying abnormal users and ensuring high efficiency in identifying abnormal users. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart illustrating a method for determining target feature parameters provided in an embodiment of this application;

[0041] Figure 2 This is a flowchart illustrating another method for determining target feature parameters provided in an embodiment of this application;

[0042] Figure 3 This is a schematic diagram illustrating a specific implementation of embodiment S120 of this application;

[0043] Figure 4 This is a schematic diagram of another specific implementation of embodiment S120 of this application;

[0044] Figure 5 This is a schematic diagram of another specific implementation of embodiment S120 of this application;

[0045] Figure 6 This is a schematic diagram of another specific implementation of embodiment S120 of this application;

[0046] Figure 7 This is a schematic diagram of the probabilistic neural network algorithm model structure;

[0047] Figure 8This is a flowchart illustrating an abnormal user identification method provided in an embodiment of this application;

[0048] Figure 9 This is a flowchart illustrating another abnormal user identification method provided in an embodiment of this application;

[0049] Figure 10 This is a schematic diagram illustrating a specific implementation of embodiment S420 of this application;

[0050] Figure 11 This is a schematic diagram illustrating a specific implementation of embodiment S430 of this application.

[0051] Figure 12 This is a schematic diagram of the structure of a target feature parameter determination device provided in an embodiment of this application;

[0052] Figure 13 This is a schematic diagram of the structure of an abnormal user identification device provided in an embodiment of this application;

[0053] Figure 14 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0054] The features and exemplary embodiments of various aspects of this application will now be described in detail. Numerous specific details are set forth in the following detailed description in order to provide a comprehensive understanding of this application. However, it will be apparent to those skilled in the art that this application can be implemented without some of these specific details. The following description of embodiments is merely intended to provide a better understanding of this application by illustrating examples thereof.

[0055] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0056] Currently, in terms of abnormal user identification, existing technologies mostly rely on user package data to build abnormal user prediction or identification models. The output results only show users who are dissatisfied with their tariffs, and cannot analyze the characteristics of abnormal users from multiple dimensions of information. This approach is biased towards marketing support, resulting in low accuracy and efficiency in identifying users.

[0057] To address the problems of the existing technologies, this application acquires multiple historical 5G user samples and performs dimensionality reduction and normalization on the relevant quasi-target user feature parameters in each historical sample to obtain normalized feature parameters. Then, using a decision tree model and the feature parameters in the feature parameter set, it identifies whether the corresponding user is an abnormal user, obtaining a classification result. Finally, the feature parameters in the feature parameter set corresponding to the classification result consistent with the user category label are determined as the target feature parameters. Therefore, the target feature parameters determined through the above process reference multi-dimensional historical user samples, which can reduce identification costs while more accurately identifying abnormal users, ensuring high efficiency in identifying abnormal users.

[0058] This application provides a method, apparatus, device, storage medium, and computer program product for determining target feature parameters. The method for determining target feature parameters provided in this application will be described first.

[0059] Figure 1 This is a flowchart illustrating a method for determining target feature parameters provided in an embodiment of this application. Figure 1 As shown, the method includes:

[0060] S110: Obtain multiple historical 5G user samples. Each historical 5G user sample includes a user category label and multiple quasi-target user feature parameters.

[0061] S120, dimensionality reduction and normalization are performed on multiple quasi-target user feature parameters respectively to obtain normalized feature parameters;

[0062] S130, generate multiple feature parameter sets based on multiple normalized feature parameters;

[0063] S140: Using the decision tree model and the feature parameters in the feature parameter set, identify whether the corresponding user is an abnormal user and obtain the identification and classification results;

[0064] S150, the feature parameters in the feature parameter set corresponding to the recognition and classification results that are consistent with the user category label are determined as the target feature parameters.

[0065] Therefore, this embodiment of the application acquires multiple historical 5G user samples, performs dimensionality reduction and normalization processing on the relevant quasi-target user feature parameters in each of the multiple historical samples, and then uses a decision tree model and feature parameters in the feature parameter set to identify whether the corresponding user is an abnormal user, obtaining the identification and classification result. The feature parameters in the feature parameter set corresponding to the identification and classification result that matches the user category label are then determined as target feature parameters. Thus, the target feature parameters determined through the above process match the abnormal user results predicted by the model, thereby ensuring a strong correlation between the target parameters and abnormal users. Simultaneously, by referencing multi-dimensional historical user samples, it can more accurately identify abnormal users while reducing identification costs, ensuring high efficiency in identifying abnormal users.

[0066] The specific implementation methods for each of the above steps are described below.

[0067] In some embodiments, in S110, the historical 5G user samples can be data information stored in a large data wide table domain. User category labels can be labels such as derogatory users or highly satisfied users. The quasi-target user feature parameters are user feature parameters belonging to the OBM domain. During acquisition, key data can be selected from multiple data information in the OBM domain based on model running capabilities and condition complexity settings. Specifically, the O domain (operation support system) primarily includes application wireless signaling and network metrics; the B domain (business support system) primarily includes application terminal information and consumption habits; and the M domain (management support system) primarily includes application user trajectories and location information.

[0068] As an example, when obtaining a historical 5G user sample, the key data should include at least twelve categories of user information fields under the OMB domain, which can be used to obtain the user's category label as a derogatory user and related information under the user's OMB domain.

[0069] For example, in the O domain, key data includes: video initial playback success rate, video initial buffer latency, video stuttering rate, and average video download speed information in video services; page response success rate, page opening latency, and page download speed information in web browsing information; instant messaging response success rate, instant messaging response latency, and rates for receiving and sending text / voice / images / videos in instant messaging information; game interaction latency information in game services; message response latency information in payment services; EPSFB fallback success rate, average EPSFB fallback latency, voice connection rate, call setup latency, voice call drop rate, RTP uplink MOS, and RTP downlink MOS information in voice services; and the number of user voice quality issues, user internet access quality issues, user network coverage quality issues, and user overall service quality issues in all services.

[0070] In Domain B, key data includes user identifier, phone number, gender, age, star rating, network tenure (months), total monthly call billing duration (minutes) (MOU), total monthly data usage (MB) (DOU), 2G DOU (MB), 3G DOU (MB), 4G DOU (MB), 4G network tenure, name of main tariff product, monthly voice call fee (RMB), out-of-plan data fee (RMB), out-of-plan voice communication fee (RMB), total payment amount (RMB), number of payments, data service fee ratio, and speed limit information.

[0071] In the M domain, key data includes workplace, permanent residence, 4G permanent cell CI, and 4G active cell CI information.

[0072] In some embodiments, in S120, the dimensionality reduction and normalization processing of multiple quasi-target user feature parameters is performed respectively. This can be done by substituting the data in the parameters into the normalization calculation formula, and the result of the calculation is the normalized user feature parameters.

[0073] As an example, the normalization calculation formula can be:

[0074] y=(x-MinValue) / (MaxValue-MinValue) (1)

[0075] Where x and y are the values ​​of the user feature parameters before and after the transformation, respectively, and MaxValue and MinValue are the maximum and minimum values ​​of the user feature parameters in this group, respectively.

[0076] In some embodiments, in S130, the multiple normalized feature parameters are from historical 5G sample users, and the feature parameters in each feature parameter set are random, and their number can be one or more, and their types can be one or more.

[0077] As an example, for a historical 5G sample user with the feature label of "disparaging user", the following parameter set can be generated: {user identifier, phone number, gender, age, star rating}, {number of poor voice quality events, number of poor internet access events, number of poor network coverage events, and number of poor service events across all services}, {work location, permanent residence, 4G permanent cell CI, and 4G active cell CI information}.

[0078] In some embodiments, in S140, the feature parameters in the feature parameter set are input into the decision tree model to obtain the identification and classification result, which may be a derogatory user or a normal user.

[0079] As an example, consider the feature parameter set {user identifier, phone number, gender, age, star rating} in the example above, which corresponds to the data {0, 1, 0, 1, 1}. The data for {number of poor voice quality events, number of poor internet access events, number of poor network coverage events, and number of poor overall service events} corresponds to the data {1, 1, 1, 1}. The data set for {work location, permanent residence, 4G permanent cell CI, and 4G active cell CI information} corresponds to the data set {0, 1, 0, 0}. The data from each set is sequentially input into a decision tree for classification. The decision tree continues inputting data as it recognizes a 1, until a 0 is encountered, at which point the decision tree terminates and outputs the classification result: the user is a derogatory user. If the final result of the decision tree is 1, then the user is classified as a normal user. Therefore, for the three sets mentioned above, the decision tree can obtain the results as derogatory user, normal user, and derogatory user.

[0080] In some embodiments, in S150, the feature parameters in the feature parameter set corresponding to the recognition and classification results that are consistent with the user category label are determined as target feature parameters.

[0081] As an example, it can be seen that the user category label of the above historical 5G sample users is derogatory user. The classification result obtained in the above decision tree model is derogatory user, normal user, derogatory user. Therefore, it can be determined that the feature parameters in the feature parameter set corresponding to the decision tree with the classification result of derogatory user are the target feature parameters, that is, {user identifier, phone number, gender, age, star rating}, {work location, permanent residence, 4G permanent cell CI, 4G active cell CI information} are the target feature parameters.

[0082] Therefore, by matching the determined target feature parameters with the abnormal user results predicted by the model, a strong correlation between the target parameters and abnormal users is ensured. At the same time, by referring to multi-dimensional historical user samples, abnormal users can be identified more accurately while reducing identification costs, thus ensuring the high efficiency of abnormal user identification.

[0083] To obtain the characteristic parameters of the target user more accurately, this application also provides another implementation of the target characteristic parameter determination method.

[0084] like Figure 2 As shown, before performing dimensionality reduction and normalization on multiple quasi-target user feature parameters to obtain the normalized feature parameters, this implementation method may also include the following steps:

[0085] S210, Analyze the correlation between multiple target user feature parameters;

[0086] S220, Based on the Pearson correlation algorithm and the aforementioned correlation, set the weights of each quasi-target user feature parameter respectively;

[0087] S230, the corresponding target user feature parameters are weighted using weights to obtain the weighted target user feature parameters.

[0088] In some embodiments, in S210, the correlation between multiple quasi-target user feature parameters is analyzed. This can be done by statistically analyzing the correlation trends between the feature parameters and presenting them in the form of bar charts, line charts, etc.

[0089] To better understand the scheme of this application, as an example, Figure 3 This is a schematic diagram of a specific implementation of S120. For example... Figure 3 As shown, when the target feature parameters are user age, business volume, package, consumption status, etc., the analysis shows that the older the user is, the less the average monthly data usage (dou) is, and conversely, the older the user is, the longer the average monthly call duration (mou).

[0090] Figure 4 This is a schematic diagram of another specific implementation of S120; for example... Figure 4 As shown, considering the consumption patterns of users in different age groups, younger users prefer to choose plans with lower prices and more data, while older users tend to prefer plans with more talk time, and are not very price-sensitive.

[0091] Figure 5 This is a schematic diagram of another specific implementation of S120; for example... Figure 5 As shown, poor network quality events are negatively correlated with user satisfaction; when the number of poor quality events exceeds 300, user satisfaction deteriorates significantly.

[0092] Figure 6 This is a schematic diagram of another specific implementation of S120; for example... Figure 6 As shown, there is a negative correlation between the number of poor community quality events and user satisfaction; when the number of poor community quality events exceeds 2000, user satisfaction is below the mean.

[0093] In some embodiments, in S220, the weight values ​​of each quasi-target user feature parameter are obtained by inputting the related feature parameters into the Pearson correlation algorithm formula.

[0094] In some embodiments, in S230, the corresponding quasi-target user feature parameters are summed with the weight values ​​to obtain the weighted quasi-target user feature parameters.

[0095] Therefore, by further analyzing the relationship between the feature parameters of the target users, and then obtaining the weight of each feature parameter according to the Pearson correlation coefficient algorithm, the original feature parameters can be weighted to ensure the accuracy and strong correlation of the feature parameters.

[0096] Based on the target feature parameters determined in the above embodiments, this application also provides a method for identifying abnormal users. To obtain accurate and rapid identification results, this application employs a probabilistic neural network algorithm model to identify abnormal users.

[0097] To better understand the scheme of this application, the probabilistic neural network algorithm model will be introduced below.

[0098] like Figure 7 As shown, Figure 7 This is a schematic diagram of the probabilistic neural network algorithm model structure.

[0099] Probabilistic Neural Networks (PNNNs), first proposed by Dr. DFSpeeht in 1989, are a branch of radial basis function networks and a type of feedforward network. They offer advantages such as simple learning processes, fast training speeds, more accurate classification, and good fault tolerance. Essentially, they are a type of supervised network classifier based on the Bayesian minimum risk criterion.

[0100] A probabilistic neural network typically has four layers: an input layer, a pattern layer, a summation layer, and an output layer. The input layer has the same number of neurons as the number of feature vectors, where X represents the input value. The pattern layer has the same number of neurons as the number of training samples, where y represents the Gaussian function. The summation layer has the same number of neurons as the number of categories, where g represents the summation function. The output layer has only one neuron. The input layer feeds the feature vectors into the network, and the pattern layer is connected to the input layer via connection weights. The similarity between the input feature vector and each pattern in the training set is calculated, and the distance between them is fed into the Gaussian function to obtain the output of the pattern layer.

[0101] To better understand the solution proposed in this application, the method of identifying abnormal users using a probabilistic neural network algorithm model is described in detail below.

[0102] Figure 8 This is a flowchart illustrating a method for identifying abnormal users provided in this application. Figure 3 As shown, the identification method may specifically include:

[0103] S810, obtain the first feature parameter matrix of each user sample in m historical 5G user samples and the second feature parameter matrix of each user to be identified in p users to be identified;

[0104] S820, normalize each first feature parameter matrix and each second feature parameter matrix respectively;

[0105] S830, calculate the Euclidean distance between each normalized second feature parameter matrix and each normalized first feature parameter matrix respectively;

[0106] S840 uses the pattern layer in the probabilistic neural network model to process the Euclidean distance and obtain the initial probability matrix;

[0107] S850 uses the summation layer of the probabilistic neural network model to solve for the first initial probability sum of the user to be identified belonging to the abnormal user class and the second initial probability sum of the normal user class, respectively.

[0108] S860, based on the first initial probability sum and the second initial probability sum, determine the third probability that the user to be identified belongs to the abnormal user class;

[0109] S870 determines that the user to be identified belongs to the abnormal user category if the third probability is greater than the preset threshold.

[0110] In S810, for example, when m historical 5G user samples are obtained, and each user sample has n first feature parameters, a set of n rows and m columns of first feature parameter matrix can be formed. When p user feature parameters to be identified are obtained, and each user sample has p first feature parameters, a set of n columns and p rows of second feature parameter matrix can be formed. m, n, and p are all positive integers.

[0111] In S820, each first feature parameter matrix and each second feature parameter matrix can be normalized according to the normalization formula to obtain a normalized m-column first parameter matrix and an n-column second parameter matrix. The normalized first feature parameter matrix and second parameter matrix are then input into the input layer of the probabilistic neural network algorithm model.

[0112] As an example, the feature parameter matrix can be normalized according to the following formulas (2) and (3):

[0113]

[0114] C m×n =B m×1 [1 1 … 1] 1×n ·X m×n (3)

[0115] Among them B T is the normalization coefficient, C is the normalized sample matrix, m is the number of historical user samples or the number of users to be identified, x is the feature parameter in the matrix, and n is the number of feature parameters.

[0116] In S830, the Euclidean distance between each normalized second feature parameter matrix and each normalized first feature parameter matrix is ​​calculated in the input layer to obtain a matrix composed of Euclidean distance values. The obtained Euclidean distance matrix is ​​then input into the pattern layer of the probabilistic neural network algorithm model.

[0117] In S840, for example, the initial probability matrix can be calculated according to the following formula (4):

[0118]

[0119] Where P represents the initial probability matrix, E pm Let e ​​represent the Euclidean distance between the p-th sample to be classified and the m-th training sample, where e and σ are constants in the Gaussian function.

[0120] In some embodiments, in S850, the summation layer of the probabilistic neural network model is used. This summation layer can be a matrix composed of summation formulas. The first initial probability that the user to be identified belongs to the abnormal user class and the second initial probability that the user belongs to the normal user class are input into the formulas to obtain the summation matrix.

[0121] As an example, the sum of the first initial probability of the user to be identified belonging to the abnormal user class and the second initial probability of the user belonging to the normal user class can be calculated using the following formula (5):

[0122]

[0123] Where S represents the probability matrix, K represents the number of users in each class after classification, m represents the total number of samples, P represents the initial probability matrix, and l represents the number of classifications. In this embodiment, l can be a constant 2, i.e., the abnormal user category and the normal user category.

[0124] For example, if the user samples used in the embodiments of this application are known to be abnormal user categories and normal user categories, then it is known that K = 2 / m, l = 2. For each user to be identified, the following operation is performed: the probability matrix calculated by formula (4) is summed with the initial probability calculated using abnormal user samples to obtain P. p1 To obtain the first initial probability sum, sum the initial probabilities calculated using normal user samples from the probability sum matrix obtained by formula (4) above, and get P. p2 Let be the second initial probability sum, and p be the p-th user to be identified.

[0125] In some embodiments, in S860, the sum of the probabilities of abnormal users and the sum of the probabilities of normal users is added to obtain the total probability, and then the proportion of the first initial probability in the total probability is calculated to obtain the third probability that the user to be identified belongs to the abnormal user class.

[0126] For example, for the user p to be identified above, the probability that this user belongs to the abnormal user class is:

[0127] In some embodiments, in S870, if the third probability is greater than a preset threshold, it is determined that the user to be identified belongs to the abnormal user class.

[0128] For example, if the preset threshold is 60% and the third probability value of the above user is 70%, then the user to be identified is considered an abnormal user.

[0129] As an example, when p is the user to be identified, and the quantity is 1, and the number of sample users is 2, where sample user x1 is a derogatory user and sample user x2 is a derogatory user, if the number of features is 2, according to S810, the first feature parameter matrix can be obtained as follows: The second feature parameter matrix is ​​[p] 11 p 12 ];

[0130] According to S820, the normalized first eigenparameter matrix is: The normalized second characteristic parameter matrix is ​​[p1];

[0131] According to S830, the Euclidean distance matrix is ​​[d1 d2];

[0132] According to step S840, the initial probability matrix is ​​[P1 P2];

[0133] According to step S850, sample user x1 is a derogatory user. At this time, P1 is the first initial probability sum of the derogatory user and P2 is the second initial probability sum of the normal user.

[0134] According to step S860, the third probability that the user to be identified belongs to the abnormal user category is...

[0135] When the third probability value reaches the preset value, the user to be identified is considered an abnormal user.

[0136] Therefore, abnormal users were identified through a probabilistic neural network model. The user's feature parameters were correlated with the user's category classification through calculation. A large number of known sample users were used as model variables, which ensured that the identification of abnormal users was more accurate. At the same time, since multiple users to be identified can be input into the probabilistic neural network model at one time, the efficiency of abnormal user identification was ensured.

[0137] Because of the need for broad coverage of abnormal user identification, and to ensure that abnormal users are processed from multiple dimensions, adjustments are required to the abnormal users identified by the above model. As an example, this application provides a flowchart of another abnormal user identification method.

[0138] like Figure 9 As shown, after determining that the user to be identified belongs to the abnormal user class, this implementation method can also include:

[0139] S910, Based on the M-domain feature parameters of each abnormal user, extract the trajectory feature points of the abnormal user to obtain the movement trajectory of the abnormal user;

[0140] S920 divides the movement trajectory into multiple trajectory segments to obtain a user trajectory set;

[0141] S930, segments and clusters the trajectory segments in the user trajectory set to generate a linear trajectory clustering map;

[0142] S940 renders the region center points of the linear trajectory clustering graph to obtain a clustering graph containing abnormal users.

[0143] In S910, the M-domain feature parameters can be information such as the location and movement trajectory of abnormal users. The extraction method can be to directly read from the big data wide table and display the movement trajectory of abnormal users in the form of a map.

[0144] As an example, user trajectory feature points can be extracted by integrating the user's permanent / active location and cell information in the M domain, and adding call detail records (CDRs) and cell information when necessary.

[0145] In S920, the movement trajectory is divided into multiple trajectory segments. The segmentation method can be the use of the traclus (trajectory segmentation) algorithm. The user's movement trajectory is input into the algorithm model, and the output result is the user trajectory set.

[0146] As an example, in order to obtain a more accurate set of user trajectory features, such as Figure 10 As shown, the S920 described above may specifically include:

[0147] S921, calculate the distance between different trajectory feature points on the moving trajectory;

[0148] S922, based on the distance, the trajectory feature points on the movement trajectory are divided into multiple trajectory segments according to the minimum description to obtain the user trajectory set.

[0149] In S921, the distance between any two trajectory feature points can be calculated using a distance formula, which can be a perpendicular distance, a parallel distance, or an angular distance.

[0150] In S922, the principle of minimal description includes:

[0151] aL(H): Describes the length required for the compression model (or encoding method);

[0152] bL(D|H): Describes the length required for data encoded using a compression model.

[0153] Based on the principle of minimal description, multiple feature trajectory segments can be obtained, thus forming a set of trajectories for abnormal users.

[0154] Therefore, the location information of abnormal users is converted into a set of trajectories by the traclus algorithm, and the complex movement trajectories of abnormal users are cut into standard length (minimum principle) trajectory units, which facilitates the calculation of the clustering process of degraded users.

[0155] In S930, multiple trajectory segments in the user trajectory set are segmented and clustered. The clustering method can be the DBSCAN (density clustering) algorithm. The trajectory segments are input into the algorithm model, and the output result is the user trajectory set.

[0156] As an example, to generate more accurate linear trajectory clustering graphs, such as Figure 11 As shown, the above-mentioned S930 may specifically include:

[0157] S931, Calculate the average vector of the multiple trajectory segments;

[0158] S932, rotate the trajectory segments in the user trajectory set according to the average vector until the coordinates of the trajectory segment in the first direction are parallel to the average vector, and use the second direction coordinate axis to sweep along the first direction coordinate axis to determine the intersection point of the second direction coordinate axis with each trajectory segment; the first direction is perpendicular to the second direction.

[0159] S933, if the number of intersection points is greater than or equal to a preset threshold, then generate a target point based on the average value of the coordinates of the intersection points in the second direction, and return to step B until there are no vector endpoints on the side of the scan line away from the origin in the first direction.

[0160] S934, rotate the target point back to its original angle, connect the positions of the target point before and after the rotation, and generate the target trajectory;

[0161] S935 generates a linear trajectory clustering map based on each target trajectory.

[0162] In S931, multiple trajectory segments can be vectorized separately to obtain vectorized values, and then the average vector of multiple trajectory segments can be calculated using the average vector calculation formula.

[0163] For example, the average vector can be calculated using the following formula (6):

[0164]

[0165] Where V represents the average vector, V n This represents the vectorized trajectory vector, where n represents the number of trajectory segments in the set.

[0166] For example, when the trajectory segment is a unit vector in the horizontal and vertical coordinate directions, the average vector of the two vectors is a unit vector that is 45 degrees upward and to the right.

[0167] In S932, the trajectory segments in the user trajectory set are rotated according to the average vector until the coordinates of the trajectory segments in the first direction are parallel to the average vector. The second direction coordinate axis is used to sweep along the first direction coordinate axis to determine the intersection point of the second direction coordinate axis with each trajectory segment; the first direction is perpendicular to the second direction.

[0168] For example, the coordinate axes can be rotated using the following formula (7) to obtain the coordinate values ​​of each trajectory vector after rotation:

[0169]

[0170] in, It can be obtained by the inner product of the average direction vector and the unit vector, where x and y are the coordinates of the vector before rotation, and x′ and y′ are the coordinates of the vector after rotation.

[0171] In S933, for example, the preset threshold is set to 3. When the number of intersection points is greater than 3, the coordinates of 3 intersection points are selected, and the average of their ordinate values ​​is calculated. At this time, an abscissa value of x′ and a ordinate value of x′ can be obtained. The intersection points.

[0172] In S934, for example, the coordinate values ​​in the original coordinate axis are calculated according to formula (7) for the multiple intersection points obtained above, and the intersection points in the original coordinate axis before rotation are obtained. Connecting these points can obtain a target trajectory.

[0173] In S935, for example, the target trajectories formed by multiple identified abnormal users can be integrated onto a map to form a linear trajectory clustering map.

[0174] Therefore, by segmenting and clustering all identified abnormal users from multiple dimensions, and further processing the behavioral trajectories of abnormal users, the results are more accurate and have greater breadth and applicability in identifying abnormal users.

[0175] Based on the same inventive concept, this application also provides a device for determining target feature parameters. (Specifically combined with...) Figure 12 Please provide a detailed explanation.

[0176] Figure 12 This is a schematic diagram of the structure of a target feature parameter determination device provided in one embodiment of this application. Figure 12 As shown, the device 1200 may include:

[0177] The first acquisition module 1201 is used to acquire multiple historical 5G user samples. Each historical 5G user sample includes a user category label and multiple quasi-target user feature parameters. The quasi-target user feature parameters are user feature parameters belonging to the OBM domain.

[0178] The first processing module 1202 is used to perform dimensionality reduction and normalization processing on multiple quasi-target user feature parameters respectively to obtain normalized feature parameters.

[0179] The generation module 1203 is used to generate multiple feature parameter sets based on multiple normalized feature parameters, and the feature parameters in each feature parameter set are not completely the same.

[0180] The identification module 1204 is used to perform the following for each feature parameter set: using the decision tree model and the feature parameters in the feature parameter set, to identify whether the corresponding user is an abnormal user and obtain the identification and classification result;

[0181] The first determining module 1205 is used to determine the feature parameters in the feature parameter set corresponding to the identification and classification results that are consistent with the user category label as target feature parameters. The target feature parameters are used to identify whether the user is an abnormal user.

[0182] Therefore, by acquiring multiple historical 5G user samples, the relevant quasi-target user feature parameters in each historical sample are subjected to dimensionality reduction and normalization. Then, using a decision tree model and feature parameters from the feature parameter set, the system identifies whether the corresponding user is an anomalous user, obtaining the identification and classification results. The feature parameters from the feature parameter set corresponding to the identification and classification results that match the user category label are then determined as the target feature parameters. Thus, the target feature parameters determined through the above process match the anomalous user results predicted by the model, ensuring a strong correlation between the target parameters and anomalous users. Furthermore, by referencing multi-dimensional historical user samples, the system can more accurately identify anomalous users while reducing identification costs, ensuring high efficiency in anomalous user identification process.

[0183] In some embodiments, to improve the accuracy of determining the target feature parameters, the first processing module may include the following modules:

[0184] The analysis submodule analyzes the correlation between multiple target user feature parameters.

[0185] The submodule is set up to determine the weights of each target user feature parameter based on the Pearson correlation algorithm and association relationships.

[0186] The weighted submodule uses weights to weight the corresponding target user feature parameters to obtain the weighted target user feature parameters.

[0187] Based on the same inventive concept, this application also provides a device for identifying abnormal users. Specifically, in conjunction with... Figure 8 Please provide a detailed explanation.

[0188] Figure 13 This is a schematic diagram of the structure of an abnormal user identification device provided in one embodiment of this application. Figure 13 As shown, the device 1300 may include:

[0189] The second acquisition module 1301 is used to acquire the first feature parameter matrix of each user sample in m historical 5G user samples and the second feature parameter matrix of each user to be identified in p users to be identified; the feature parameters in the first feature parameter matrix and the second feature parameter matrix are determined according to the method in device 1300; m and p are positive integers.

[0190] The second processing module 1302 is used to normalize each first feature parameter matrix and each second feature parameter matrix respectively to obtain the normalized first feature parameter matrix and the second feature parameter matrix.

[0191] The calculation module 1303 is used to calculate the Euclidean distance between each normalized second feature parameter matrix and each normalized first feature parameter matrix respectively;

[0192] The third processing module 1304 is used to process the Euclidean distance using the pattern layer in the probabilistic neural network model to obtain the initial probability matrix;

[0193] The solver module 1305 is used to solve the first initial probability sum of the user to be identified belonging to the abnormal user class and the second initial probability sum of the normal user class using the summation layer of the probabilistic neural network model.

[0194] The second determining module 1306 is used to determine a third probability that the user to be identified belongs to the abnormal user class based on the first initial probability sum and the second initial probability sum;

[0195] The third determination module 1307 is used to determine that the user to be identified belongs to the abnormal user class when the third probability is greater than a preset threshold.

[0196] Therefore, by identifying abnormal users and referencing multi-dimensional historical user samples, it is possible to identify abnormal users more accurately while reducing identification costs and ensuring the efficiency of abnormal user identification.

[0197] In some embodiments, to more accurately identify abnormal users, the device 1300 may further include the following modules:

[0198] The extraction module is used to extract the trajectory feature points of each abnormal user based on the M-domain feature parameters, so as to obtain the movement trajectory of the abnormal user.

[0199] The cutting module is used to cut the movement trajectory into multiple trajectory segments to obtain a set of user trajectories.

[0200] The clustering module is used to segment and cluster the trajectory segments in the user trajectory set to generate a linear trajectory clustering graph.

[0201] The rendering module is used to render the region center points of the linear trajectory clustering graph to obtain a clustering graph containing the abnormal user.

[0202] To obtain more accurate user trajectory segments, the above-mentioned segmentation module also includes:

[0203] The calculation submodule is used to calculate the distance between different trajectory feature points on the movement trajectory;

[0204] The cutting submodule is used to cut the trajectory feature points on the movement trajectory into multiple trajectory segments according to the minimum description based on the distance, so as to obtain the user trajectory set.

[0205] To obtain a more accurate clustering diagram, the above clustering module also includes:

[0206] The calculation submodule is used to calculate the average vector of multiple trajectory segments;

[0207] The rotation submodule is used to rotate the trajectory segments in the user trajectory set according to the average vector until the coordinates of the trajectory segment in the first direction are parallel to the average vector.

[0208] The determination submodule is used to scan along the first direction coordinate axis using the second direction coordinate axis to determine the intersection points of the second direction coordinate axis and each of the trajectory segments;

[0209] The generation submodule is used to generate a target point based on the average value of the coordinates in the second direction of the intersection points if the number of intersection points is greater than or equal to a preset threshold, and then return to step B until there are no vector endpoints on the side of the scan line away from the origin in the first direction.

[0210] The connection submodule is used to rotate the target point back to its original angle, connect the positions of multiple rotated target points, and generate the target trajectory.

[0211] The generation submodule is used to generate a linear trajectory clustering diagram based on each target trajectory.

[0212] Figure 14 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application is shown. The electronic device may include a processor 1401 and a memory 1402 storing computer program instructions.

[0213] Specifically, the processor 1401 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0214] Memory 1402 may include mass storage for data or instructions. For example, and not limitingly, memory 1402 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 1402 may include removable or non-removable (or fixed) media. Where appropriate, memory 1402 may be internal or external to an electronic device. In a particular embodiment, memory 1402 is a non-volatile solid-state memory.

[0215] Memory 1402 may include read-only memory (ROM), flash memory device, random access memory (RAM), disk storage medium device, optical storage medium device, electrical, optical, or other physical / tangible memory storage device. Therefore, typically, memory 1402 includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the methods described above according to the foregoing aspects of this disclosure.

[0216] The processor 1401 implements any of the methods described above by reading and executing computer program instructions stored in the memory 1402.

[0217] In one example, the device may also include a communication interface 1403 and a bus 1410. Wherein, as... Figure 14 As shown, the processor 1401, memory 1402, and communication interface 1403 are connected through bus 1410 and complete communication with each other.

[0218] The communication interface 1403 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0219] Bus 1410 includes hardware, software, or both, that couples components of an electronic device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 1410 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0220] This electronic device can be based on a method to achieve integration Figures 1 to 13 The methods and apparatus described.

[0221] Furthermore, in conjunction with the methods in the above embodiments, this application embodiment can provide a computer storage medium for implementation. This computer storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the methods in the above embodiments.

[0222] In addition, this application also provides a computer program product, including a computer program, which, when executed by a processor, can implement the steps and corresponding content of the aforementioned method embodiments.

[0223] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0224] It should be understood that in the embodiments of this application, "B corresponding to A" means that B is associated with A, and B can be determined based on A. However, it should also be understood that determining B based on A does not mean that B is determined solely based on A; B can also be determined based on A and / or other information.

[0225] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for determining target feature parameters, characterized in that, include: Multiple historical 5G user samples are obtained. Each historical 5G user sample includes a user category label and multiple quasi-target user feature parameters. The quasi-target user feature parameters are user feature parameters belonging to the OBM domain of the Operation Business Management Domain. Perform the following for each historical 5G user sample: The dimensionality reduction and normalization processes are performed on the multiple quasi-target user feature parameters respectively to obtain the normalized feature parameters; Based on multiple normalized feature parameters, multiple feature parameter sets are generated, and the feature parameters in each feature parameter set are not completely the same; For each set of feature parameters, the following steps are performed: using the decision tree model and the feature parameters in the set of feature parameters, the corresponding user is identified as an abnormal user, and the identification and classification results are obtained. The feature parameters in the feature parameter set corresponding to the identification and classification results that are consistent with the user category label are determined as target feature parameters, which are used to identify whether the user is an abnormal user.

2. The method according to claim 1, characterized in that, Before performing dimensionality reduction and normalization on the plurality of quasi-target user feature parameters to obtain the normalized feature parameters, the method further includes: Analyze the correlation between the multiple quasi-target user feature parameters; Based on the Pearson correlation algorithm and the aforementioned correlation, the weights of each quasi-target user feature parameter are set respectively; The corresponding target user feature parameters are weighted using the weights to obtain the weighted target user feature parameters. The step of performing dimensionality reduction and normalization processing on the multiple quasi-target user feature parameters respectively to obtain normalized feature parameters includes: The weighted quasi-target user feature parameters are subjected to dimensionality reduction and normalization respectively to obtain the normalized feature parameters.

3. A method for identifying abnormal users, characterized in that, The method includes: Obtain the first feature parameter matrix of each user sample from m historical 5G user samples and the second feature parameter matrix of each user to be identified from p users to be identified; the feature parameters in the first and second feature parameter matrices are determined according to the method described in claim 1 or 2; m and p are positive integers; Normalize each first feature parameter matrix and each second feature parameter matrix respectively to obtain the normalized first feature parameter matrix and second feature parameter matrix; Calculate the Euclidean distance between each normalized second feature parameter matrix and each normalized first feature parameter matrix; The Euclidean distance is processed using the pattern layer in the probabilistic neural network model to obtain the initial probability matrix; For each user to be identified, perform the following steps: Using the summation layer of the probabilistic neural network model, the first initial probability sum of the user to be identified belonging to the abnormal user class and the second initial probability sum of the normal user class are respectively solved; Based on the first initial probability and the second initial probability, a third probability is determined that the user to be identified belongs to the abnormal user class; If the third probability is greater than a preset threshold, the user to be identified is determined to belong to the abnormal user category.

4. The method according to claim 3, characterized in that, After determining that the user to be identified belongs to the abnormal user class, the method further includes: Based on the M-domain feature parameters of each abnormal user, the trajectory feature points of the abnormal user are extracted to obtain the movement trajectory of the abnormal user. The movement trajectory is divided into multiple trajectory segments to obtain a user trajectory set; The trajectory segments in the user trajectory set are segmented and clustered to generate a linear trajectory clustering graph.

5. The method according to claim 4, characterized in that, The step of dividing the movement trajectory into multiple trajectory segments to obtain a user trajectory set includes: Calculate the distance between different trajectory feature points on the movement trajectory; Based on the distance, the trajectory feature points on the movement trajectory are divided into multiple trajectory segments according to the minimum description to obtain a user trajectory set.

6. The method according to claim 5, wherein the distance comprises at least one of the following: Vertical distance, horizontal distance, angular distance.

7. The method according to claim 4, characterized in that, The user trajectory set is segmented and clustered to generate a linear trajectory clustering graph, including: For each abnormal user, perform steps A through D to obtain the target trajectory for each abnormal user: Step A: Calculate the average vector of the multiple trajectory segments; Step B: Rotate the trajectory segments in the user trajectory set according to the average vector until the coordinates of the trajectory segment in the first direction are parallel to the average vector. Use the second direction coordinate axis to sweep along the first direction coordinate axis to determine the intersection point of the second direction coordinate axis with each trajectory segment; the first direction is perpendicular to the second direction. Step C: If the number of intersection points is greater than or equal to a preset threshold, then generate a target point based on the average value of the coordinates of the intersection points in the second direction, and return to step B until there are no vector endpoints on the side of the scan line away from the origin in the first direction. Step D: Rotate the target point back to its original angle, connect the positions of the multiple rotated target points, and generate the target trajectory; Generate a linear trajectory clustering diagram based on each target trajectory.

8. A device for determining target feature parameters, characterized in that, The device includes: The first acquisition module is used to acquire multiple historical 5G user samples. Each historical 5G user sample includes a user category label and multiple quasi-target user feature parameters. The quasi-target user feature parameters are user feature parameters belonging to the OBM domain. The first processing module is used to perform dimensionality reduction and normalization processing on the multiple quasi-target user feature parameters respectively to obtain normalized feature parameters; The generation module is used to generate multiple feature parameter sets based on multiple normalized feature parameters, and the feature parameters in each feature parameter set are not completely the same; The identification module is used to perform the following for each feature parameter set: using the decision tree model and the feature parameters in the feature parameter set, it identifies whether the corresponding user is an abnormal user and obtains the identification and classification result; The first determining module is used to determine the feature parameters in the feature parameter set corresponding to the identification and classification results that are consistent with the user category label as target feature parameters, and the target feature parameters are used to identify whether the user is an abnormal user.

9. An abnormal user identification device, characterized in that, The device includes: The second acquisition module is used to acquire the first feature parameter matrix of each user sample in m historical 5G user samples and the second feature parameter matrix of each user to be identified in p users to be identified; the feature parameters in the first feature parameter matrix and the second feature parameter matrix are determined according to the method of claim 1 or 2; m and p are positive integers; The second processing module is used to normalize each first feature parameter matrix and each second feature parameter matrix respectively to obtain the normalized first feature parameter matrix and the second feature parameter matrix. The calculation module is used to calculate the Euclidean distance between each normalized second feature parameter matrix and each normalized first feature parameter matrix, respectively. The third processing module is used to process the Euclidean distance using the pattern layer in the probabilistic neural network model to obtain the initial probability matrix; The solution module is used to solve the first initial probability sum of the user to be identified belonging to the abnormal user class and the second initial probability sum of the normal user class by using the summation layer of the probabilistic neural network model. The second determining module is used to determine a third probability that the user to be identified belongs to the abnormal user class based on the first initial probability sum and the second initial probability sum; The third determining module is used to determine that the user to be identified belongs to the abnormal user class when the third probability is greater than a preset threshold.

10. An electronic device, characterized in that, The electronic device includes a memory, a processor, and a computer program stored in the memory and running on the processor, the computer program performing the method as described in any one of claims 1 to 7.

11. A computer storage medium, characterized in that, When the computer storage medium is executed by the processor, it implements the method according to any one of claims 1 to 7.

12. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the method as described in any one of claims 1 to 7.

Images