An online slow user identification method and device and a storage medium

Abstract

The present application relates to the technical field of artificial intelligence, in particular to a slow internet user identification method, device and equipment and computer storage medium. The slow internet user identification method disclosed by the present application uses the random forest and curve fitting method in machine learning to identify the slow internet user of home broadband. The random forest algorithm is used to determine the importance of the index, the curve fitting is used to determine the index threshold, the slow internet user identification rule is obtained, and the slow internet degree of the user identified as the slow internet user is calculated through the formula, so that the quantitative division of the slow internet user is realized, and the accuracy of the slow internet user identification is improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, apparatus, device, and computer storage medium for identifying users with slow internet speeds. Background Technology

[0002] In existing technologies, the identification of potential users with poor home broadband quality, citing "slow internet for several consecutive days," is calculated based on weak light and the signal strength of connected Wi-Fi devices. This indicates a problem with the line or equipment performance leading to poor quality. The diagnostic process for slow Wi-Fi throughout the smart home establishes cause codes including unpaid bills, service suspension, inaccessibility, and gateway mode. The diagnostic results corresponding to these cause codes tend to focus on services and pricing, limiting network quality-related aspects to router equipment indicators. This approach has the following shortcomings: existing technologies related to network quality and fault handling emphasize network quality or fault resolution, neglecting the correlation between network indicators and user perception, and failing to provide clear rules or quantitative definitions of "slow internet." Therefore, accurately identifying users with slow internet access is a problem that needs to be solved. Summary of the Invention

[0003] Therefore, the technical problem to be solved by the present invention is to overcome the problem that the prior art cannot accurately identify users with slow Internet access.

[0004] To solve the above-mentioned technical problems, the present invention provides a method for identifying users with slow internet access, comprising:

[0005] Obtain correlation data between user internet performance metrics and user internet speed evaluations;

[0006] Based on the associated data, correlation analysis and dimensionality reduction are performed on the user internet performance indicators, and the importance of the dimensionality-reduced user internet performance indicators is ranked using the random forest algorithm. User internet performance indicators with an importance index greater than a preset threshold are selected as key indicators of user internet performance.

[0007] Based on the fast and slow samples of user internet performance indicators, calculate the fitting function curves for fast and slow internet for each performance indicator, and determine the critical threshold for evaluating internet speed for each performance indicator based on the inflection point where the two types of fitting function curves intersect on the coordinate axis.

[0008] Obtain key performance data of the user's internet access, and determine whether the user is a slow internet user by comparing the data of each key performance indicator with the corresponding critical threshold.

[0009] Preferably, the associated data for obtaining user internet performance indicators and user internet speed evaluation includes:

[0010] Deep packet inspection is performed on the CR devices of the metropolitan area network unit, and mirroring technology is used to collect broadband users' internet access signaling.

[0011] The broadband user's internet access signaling is parsed into XDR call detail records, and the user's internet access performance indicators are extracted based on the XDR call detail records.

[0012] Based on the aforementioned user internet performance indicators, user internet speed evaluation survey data is collected to generate correlation data between user internet performance indicators and user internet speed evaluation.

[0013] Preferably, the step of using the random forest algorithm to rank the importance of the dimensionality-reduced user internet performance indicators and selecting user internet performance indicators with an importance index greater than a preset threshold as key user internet performance indicators includes:

[0014] Using a classification and regression tree algorithm, the user's online performance indicators after dimensionality reduction are recursively divided to generate multiple decision trees, forming a random forest.

[0015] Calculate the internet performance index x of the j-th user based on the sum of the changes in the Gini coefficient before and after the branch at node k. j The importance score on the w-th decision tree, where k∈U, and U is the feature x. j The set of nodes that appear on the w-th decision tree;

[0016] The importance scores of the internet performance index xj of the j-th user on all decision trees are calculated and normalized to obtain the final importance index of the internet performance index xj of the j-th user.

[0017] Preferably, the step of calculating the fitting function curves for fast and slow internet access for each performance indicator based on the fast and slow internet access samples of the user's internet access performance indicators, and determining the critical threshold for evaluating the internet access speed for each performance indicator based on the inflection point where the two types of fitting function curves intersect on the coordinate axis, includes:

[0018] Based on the fast and slow internet performance samples of users, the fitting function curves for fast and slow internet performance for each performance indicator are calculated using linear fitting, curve fitting, and exponential fitting.

[0019] Find the inflection point where the curves of the two types of fitted functions intersect on the coordinate axis, and select the inflection point where the performance index changes from good to bad as the critical threshold for the performance index to change from fast to slow.

[0020] Preferably, the step of acquiring key performance indicators of the user's internet access and determining whether the user is a slow internet user includes:

[0021] If all key performance indicators exceed their corresponding critical thresholds, the user is identified as a slow internet user.

[0022] Preferably, after obtaining the key performance indicators of the user's internet access and determining whether the user is a slow internet user, the process further includes:

[0023] The importance index of key performance indicators for users with slow internet speeds is calculated by summing the data that exceed the corresponding critical threshold within a given period to determine the degree of slowness of the user's internet speed.

[0024] Preferably, determining the degree of internet slowness for a user by summing the importance indices of key performance indicators exceeding the corresponding critical threshold within a given period includes:

[0025] Calculate the internet speed of a slow user in hour h:

[0026]

[0027] ξ j =S j (W)η j

[0028]

[0029] Where J represents the total number of key performance indicators, and S... j (W) represents the importance index of the j-th key performance indicator;

[0030] Calculate the internet speed of a user with slow internet connection on a given day:

[0031]

[0032] Wherein, λ1, λ2, and λ3 are the weights corresponding to the busy / sluggishness of business in the current time period.

[0033] The present invention also provides a device for identifying slow internet users, comprising:

[0034] The data acquisition module is used to acquire related data on user internet performance indicators and user internet speed evaluation;

[0035] The indicator screening module is used to perform correlation analysis and dimensionality reduction on the user's internet performance indicators based on the associated data, and to sort the importance of the dimensionality-reduced user internet performance indicators using the random forest algorithm, and to screen user internet performance indicators with an importance index greater than a preset threshold as key indicators of user internet performance.

[0036] The threshold calculation module is used to calculate the fitting function curves of fast and slow internet access for each performance indicator based on the fast and slow internet access samples of the user's internet access performance indicators, and to determine the critical threshold for evaluating the speed of internet access for each performance indicator based on the inflection point when the two types of fitting function curves intersect on the coordinate axis.

[0037] The slow internet access identification module is used to obtain key performance indicators of the user's internet access and to determine whether the user is a slow internet user by comparing the data of each key performance indicator with the corresponding critical threshold.

[0038] The present invention also provides a device for identifying slow internet users, comprising:

[0039] Memory, used to store computer programs;

[0040] A processor is used to implement the steps of the above-described method for identifying slow internet users when executing the computer program.

[0041] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described method for identifying slow internet users.

[0042] The technical solution of the present invention has the following advantages compared with the prior art:

[0043] The slow internet user identification method described in this invention utilizes random forest and curve fitting methods from machine learning to identify slow internet users in home broadband. By employing the random forest algorithm to determine the importance of indicators and using curve fitting to determine indicator thresholds, a rule for identifying slow internet users is derived. The degree of slow internet speed for users identified as slow is then calculated using a formula, thereby achieving quantitative classification of slow internet users and improving the accuracy of slow internet user identification. Attached Figure Description

[0044] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein:

[0045] Figure 1 This is a flowchart illustrating the implementation of a method for identifying slow internet users provided by the present invention.

[0046] Figure 2 It is the set of independent variables for internet performance indicators;

[0047] Figure 3 This is a diagram showing the order of importance of the indicators;

[0048] Figure 4 This is a statistical diagram of the various performance index functions obtained after fitting;

[0049] Figure 5 This is a schematic diagram of the fitted curves for some performance indicators;

[0050] Figure 6 This is a diagram illustrating the importance and thresholds of the output metrics;

[0051] Figure 7 This is a feedback form from some users in Henggang Village, Qili Avenue South, Leping Town, Sanshui District, Foshan, regarding slow internet speeds.

[0052] Figure 8 These are sample images of some complaint work orders. Detailed Implementation

[0053] The core of this invention is to provide a method, apparatus, device, and computer storage medium for identifying users with slow internet access, which effectively improves the accuracy of identifying such users.

[0054] To enable those skilled in the art to better understand the present invention, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] Please refer to Figure 1 , Figure 1 The flowchart illustrates the implementation of a method for identifying slow internet users provided by this invention; the specific operation steps are as follows:

[0056] S101: Obtain the correlation data between user internet performance indicators and user internet speed evaluation;

[0057] S102: Based on the associated data, perform correlation analysis and dimensionality reduction on the user internet performance indicators, and use the random forest algorithm to sort the importance of the dimensionality-reduced user internet performance indicators, and select user internet performance indicators with an importance index greater than a preset threshold as key indicators of user internet performance.

[0058] S103: Calculate the fitting function curves for fast and slow internet access for each performance indicator based on the fast and slow internet access samples of the user's internet access performance indicators, and determine the critical threshold for evaluating the speed of internet access for each performance indicator based on the inflection point where the two types of fitting function curves intersect on the coordinate axis.

[0059] S104: Obtain key performance data of the user's internet access, and determine the magnitude of each key performance data point relative to the corresponding critical threshold to ascertain whether the user is a slow internet user.

[0060] Based on the above embodiments, this embodiment will provide a detailed description of step S101:

[0061] • Perform deep packet inspection on CR devices in metropolitan area network units and use mirroring technology to collect broadband user internet access signaling;

[0062] • Parse the broadband user's internet access signaling into XDR call detail records, and extract the user's internet access performance indicators based on the XDR call detail records;

[0063] Based on the aforementioned user internet performance indicators, collect user internet speed evaluation survey data and generate correlation data between user internet performance indicators and user internet speed evaluation.

[0064] In one embodiment, the present invention performs deep packet inspection on the CR device of the metropolitan area network (MAN) network unit and uses mirroring technology to output broadband user internet access signaling to the signaling acquisition device in near real-time. The MAN is a TCP / IP network, and the communication protocols used include AAA, TCP, and HTTP. The signaling obtained from the signaling acquisition device is then parsed into XDR call detail records (CDRs) using a protocol parsing tool and output to the signaling platform sharing layer. The sharing layer calculates various data in the XDRs, including key fields such as time, traffic, authentication success, and accessed websites, and outputs a statistical table of 375 user-dimensional internet access performance indicators, including traffic, TCP two-way and three-way handshake latency, and HTTP response success rate. Generally, existing technologies directly use performance indicator statistical tables to conduct network quality analysis and mining for users. This approach can discover potential quality issues before user complaints, but it cannot correlate with user evaluations of perceived speed. Therefore, based on the user performance index statistics table, this invention conducted a survey on the internet speed evaluation of 1.4 million users across the entire network, collected 5,630 evaluation results of internet speed / slowness, added the results data to the user performance index statistics table, and output a correlation table between user performance index and internet speed evaluation.

[0065] Based on the above embodiments, this embodiment will provide a detailed description of step S102:

[0066] This invention utilizes the bootstrap resampling method to repeatedly sample with replacement from user internet speed evaluations obtained across the entire network. The association table from step S101 is used as the input for this step's calculation. 375 user-level internet performance indicators are used as independent variables X, and user internet speed evaluation results are used as dependent variables Y. Correlation analysis is applied to the 375 independent variables to reduce dimensionality, retaining 53 user-level internet performance indicators with correlation coefficients above 0.5 as the feature set X = (x0, x1, x2, ..., x...). 52 ),See Figure 2 .

[0067] A random forest is constructed by dividing the training set K and the test set N in an 8:2 ratio. The random forest consists of multiple decision trees, constructed using the CART algorithm (Classification and Regression Tree). CART recursively divides the 53-dimensional space into non-overlapping rectangles. The partitioning steps are as follows:

[0068] (1) Select an independent variable x i Then select x i A value v i v i Divide the 53-dimensional space into two parts, where all samples in one part satisfy x i ≤v i All samples in the other part satisfy x i >v i For discrete variables, the attribute value can only take two values: equal to the value or not equal to the value. For continuous variables, discretization is required first.

[0069] (2) Recursive processing: The two parts obtained above are re-selected according to step (1) to continue the division until the entire 53-dimensional space is divided. During the division process, for a variable attribute, its split point is the midpoint of a pair of continuous variable attribute values. Assuming that a set of D samples has D continuous values ​​for an attribute, there will be D-1 split points, each of which is the average of two adjacent continuous values. The division of each attribute is sorted according to the amount of impurity that can be reduced, and the amount of impurity reduction is defined as the sum of the impurity measure before the division minus the impurity measure of each node after the division. The Gini coefficient is commonly used to measure the impurity of the training dataset K. The Gini coefficient is calculated at the branch node. If a certain purity is satisfied, the data is divided into the left subtree; otherwise, it is divided into the right subtree, and finally a binary decision tree is generated. The smaller the Gini coefficient, the higher the "purity" of the sample. Let the Gini coefficient be G. Assume that a sample has L classes and the probability of belonging to class l is p. l Then the Gini coefficient of a node k is:

[0070]

[0071] Equation (1) maximizes G(k) when G(k) = 0, all samples belong to the same class, and all classes appear in the nodes with equal probability. This invention is divided into two categories: fast internet access and slow internet access, i.e., L=2. In this case, equation (1) can be simplified to:

[0072]

[0073] In the formula, p1 + p2 = 1.

[0074] During the recursive partitioning process, if all samples of the current node do not belong to the same class or there is only one sample left, then this node is a non-leaf node. Therefore, this invention attempts to find the partition with the largest impurity variable by trying to find each attribute of the sample and the corresponding split point. The subtree of this attribute partition is the optimal branch.

[0075] Step (2) The CART algorithm uses post-pruning. In this invention, post-pruning employs a cost-complexity pruning method: τ(q) is the error rate of node q, ρ(q) is the proportion of data at node q to all data, and if the node is pruned, σ(q) is the error cost of node q.

[0076] σ(q)=τ(q)×σ(q) ​​(3)

[0077] If the node is not pruned, σ(q) is the subtree T. t The error cost is equal to that of subtree T. t The sum of the error costs of all leaf nodes; The number of leaf nodes contained in the subtree is given by equation (3). For each non-leaf node in the classification regression tree, the surface error rate gain value α is given by equation (3).

[0078]

[0079] Each decision tree consists of M training samples K and a random vector θ. k Generate a random vector sequence {θ} k The set of all independent and identically distributed decision trees {h(X,θ)} constitutes a random forest. k ), k=1,2,...,K}, each decision tree h(X,θ) k Each variable has one vote to select the input variable x. j The classification results, i.e.

[0080]

[0081] In the formula, H(x) j ) represents the classification result of the random forest, h i (x j Y represents the classification result of a single decision tree, I represents the classification objective, and I represents the indicator function.

[0082] Suppose there are W decision trees, and feature x j The importance of node k in the w-th tree, i.e., the change in the Gini coefficient before and after node k branches, is the indicator importance score, denoted by S.

[0083] S jk (w)=G(k)-G(q)-G(f) (5)

[0084] In the formula, G(q) and G(f) represent the Gini coefficients of the two new nodes after branching, respectively. Characteristic x j If the nodes appearing in the decision tree w form a set U, then the feature x j The importance of decision tree w is:

[0085]

[0086] For all W decision trees, feature x j Its importance is:

[0087]

[0088] From equation (7), all features X = (x0, x1, x2, ..., x) are obtained. 52 The importance of ) was normalized for the importance of all features, resulting in 53 importance indices for performance indicators (see below). Figure 3 .

[0089] Based on the above embodiments, this embodiment will provide a detailed description of step S103:

[0090] For the feature set X = (x0, x1, x2, ..., x... 52 Feature x in ) j After dividing the D samples into two categories, fast and slow internet access, according to equation (4), the sample data is further processed using fitting functions including linear fitting, curve fitting, and exponential fitting. where a is the slope and b is the intercept, the linear fitting function is:

[0091] Z(x) = ax + b (8)

[0092] m <N-1,a0,a1,a2…,a m The polynomial fitting function with slope b and intercept b and order no higher than N-1 is:

[0093] Z(x) = a0 + a1x + a2x 2 +…+a m x m +b (9)

[0094] Let a be the coefficient, b be the exponent, and the exponential fitting function be:

[0095] Z(x) = ae bx (10)

[0096] where a is the coefficient and b is the intercept, the logarithmic fitting function is:

[0097] Z(x)=a ln x+b (11)

[0098] Equations (8), (9), (10), and (11) are used to fit the performance indicators of the D samples for fast and slow internet access, respectively. The fitting functions for each performance indicator corresponding to fast / slow internet access are calculated as follows: Figure 4 .

[0099] Figure 4 The corresponding plottable fitting curves are shown below, taking two performance metrics—total TCP retransmission packets and video packet latency—as examples. Figure 5 .

[0100] Depend on Figure 4 Therefore, the function fitted to the samples with fast internet speeds is Z1(x), and the function fitted to the samples with slow internet speeds is Z2(x). Figure 5 Z1(x) = Z2(x), and the fitting curves for the two classes of samples show inflection points on the coordinate axes. Due to the characteristics of curve fitting, the two curves will have multiple inflection points on the coordinate axes. The inflection point where the performance index changes from good to bad is defined as the threshold F for the user's perception of various performance indicators in the internet browsing process from fast to slow. j .

[0101] Therefore, the importance and thresholds of 53 indicators are derived as follows: Figure 6 :

[0102] Based on the above embodiments, this embodiment will provide a detailed description of step S104:

[0103] Compare the j-th performance metric of a user to be identified with Figure 6 The threshold comparison is denoted as η. j ,but:

[0104]

[0105] In the formula, j = 0, 1, 2, ..., 52

[0106] From equations (7) and (12), the rule for identifying a user as having slow internet speed is: For a user, when S j (W)≥0.05, where the comparison result of any performance index for j=0,1,2,...,52 is η j =0, then the user is not identified as a slow internet user; otherwise, the user is identified as a slow internet user.

[0107] Further, the degree of identified slow internet access is quantitatively divided. Let h be the time of day, and take one hour as the observation and statistical period. From equations (7) and (12), the contribution of the j-th performance indicator to the degree of slow internet access for a certain user in the h-th hour is obtained by multiplying the importance of the j-th performance indicator with the threshold comparison result:

[0108] ξ j =S j(W)η j (13)

[0109] In the formula, j = 0, 1, 2, ..., 52

[0110] From equation (13), the internet speed of the 53 indicators in the h-th hour can be obtained as follows:

[0111]

[0112] From equation (14), the internet speed of a user, composed of 53 performance indicators with a daily observation period, can be calculated as follows:

[0113]

[0114] In the formula, λ1, λ2, and λ3 are weights, initially assigned values ​​of λ1 = 0.1, λ2 = 0.3, and λ3 = 0.5 based on empirical values ​​of the difference between user activity during peak and off-peak hours. As user internet performance data accumulates, the weights can be adaptively adjusted according to the ratio of user volume to service volume in different time periods.

[0115] By statistically analyzing the internet speed of users identified as having slow internet access over a month, a quarter, or a year, and taking the average or recording the number of times, priority and focused processing can be applied to these identified users.

[0116] This invention relates to a machine learning method for identifying slow internet users. It uses the random forest algorithm in machine learning to rank the importance of network quality performance indicators, then uses curve fitting to obtain the degradation threshold of each indicator, and combines importance and common network knowledge to determine key indicators and thresholds. It quantitatively obtains the specific values ​​of key indicators that characterize slow internet access, and finally identifies the degree of slow internet access of users by determining judgment rules, thereby realizing the quantitative classification of slow internet users.

[0117] Compared with existing technologies that do not assign importance to network metrics but instead select metrics based on experience, this invention uses the random forest algorithm of machine learning to quantitatively assign importance to metrics, realizing a shift from qualitative to quantitative, and from subjective to objective, and better reflecting the user's actual usage perception in network performance metrics.

[0118] Compared to existing technologies that do not provide a specific numerical value for "slow internet access," this invention uses curve fitting to determine a threshold, which better reflects the actual user experience of network performance degradation and is closer to the user's real-world perception. Furthermore, by combining weighted calculations based on the importance of the indicators, the degree of internet slowness is determined, allowing for flexible output and identification of users with different levels of slow internet access, thus overcoming the deficiency of existing technologies in clearly quantifying the degree of "slow internet access."

[0119] Compared to existing technologies that focus on poor quality due to line or equipment performance issues, or slow internet speeds due to service and pricing problems, without considering actual user experience and internet performance metrics, this invention combines network performance metrics and user internet experience evaluation results to generate training and testing sets for machine learning. This achieves a correlation between objective data and subjective perception, making the results closer to actual user experience. It provides a data foundation for accurate output and allows for targeted optimization of resource allocation, offering guidance for improving user internet experience and on-site work efficiency.

[0120] Based on the above embodiments, in order to verify the accuracy of the present invention in identifying potential slow internet users, Figure 7 To identify 34 users with slow internet speeds from 1350 broadband subscribers in Henggang Village, Qili Avenue South, Leping Town, Sanshui District, Foshan, the daily internet speed was recorded and averaged over a month. User feedback was then gathered through phone calls and contact with installation and maintenance personnel. 28 users reported slow internet speeds. Verification results show that the accuracy rate of this invention is 28 / 34 * 100% = 82.35%. For the 28 users identified as having slow internet speeds requiring on-site repair, work orders were dispatched to installation and maintenance personnel. Follow-up calls were made the following day to assess user experience. Excluding 3 users who could not be contacted multiple times, 19 users reported improvements. This invention's method has a significant effect on improving users' internet experience.

[0121] Figure 8 Examples of user identification for the 7293 service tickets filed by home broadband users in March 2022 regarding slow internet speeds are provided. This invention can identify 5999 users with slow internet speeds corresponding to the service tickets, with an accuracy rate of 82.26%.

[0122] This invention also provides a device for identifying users with slow internet access; the specific device may include:

[0123] The data acquisition module is used to acquire related data on user internet performance indicators and user internet speed evaluation;

[0124] The importance ranking module is used to perform correlation analysis and dimensionality reduction on the user's internet performance indicators based on the associated data, and to use the random forest algorithm to rank the importance of the dimensionality-reduced user internet performance indicators, and to select user internet performance indicators with an importance index greater than a preset threshold as key indicators of user internet performance.

[0125] The threshold calculation module is used to calculate the fitting function curves of fast and slow internet access for each performance indicator based on the fast and slow internet access samples of the user's internet access performance indicators, and to determine the critical threshold for evaluating the speed of internet access for each performance indicator based on the inflection point when the two types of fitting function curves intersect on the coordinate axis.

[0126] The indicator filtering module is used to obtain key performance indicator data of the user's Internet access, and to determine whether the user is a slow Internet user by comparing the data of each key performance indicator with the corresponding critical threshold.

[0127] The slow internet identification module is used to calculate the sum of the importance indices of key performance indicators that exceed the corresponding critical threshold within a period for users with slow internet speeds, thereby determining the degree of slowness of the user's internet speed.

[0128] The slow internet user identification device in this embodiment is used to implement the aforementioned slow internet user identification method. Therefore, the specific implementation of the slow internet user identification device can be found in the previous embodiment section of the slow internet user identification method. For example, the data acquisition module, importance ranking module, threshold calculation module, indicator filtering module, and slow internet identification module are used to implement steps S101, S102, S103, S104, and S105 in the above-mentioned slow internet user identification method. Therefore, its specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.

[0129] A specific embodiment of the present invention also provides a device for identifying slow internet users, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the above-described method for identifying slow internet users.

[0130] A specific embodiment of the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described method for identifying slow internet users.

[0131] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0132] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0133] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0134] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0135] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for identifying users with slow internet speeds, characterized in that, include: Obtain correlation data between user internet performance metrics and user internet speed evaluations; Based on the associated data, correlation analysis and dimensionality reduction are performed on the user internet performance indicators, and the importance of the dimensionality-reduced user internet performance indicators is ranked using the random forest algorithm. User internet performance indicators with an importance index greater than a preset threshold are selected as key indicators of user internet performance. Based on the fast and slow samples of user internet performance indicators, calculate the fitting function curves for fast and slow internet for each performance indicator, and determine the critical threshold for evaluating internet speed for each performance indicator based on the inflection point where the two types of fitting function curves intersect on the coordinate axis. Obtain key performance data of the user's internet access, and determine whether the user is a slow internet user by comparing the data of each key performance indicator with the corresponding critical threshold.

2. The method for identifying slow internet users according to claim 1, characterized in that, The associated data for obtaining user internet performance metrics and user internet speed evaluations includes: Deep packet inspection is performed on the CR devices of the metropolitan area network unit, and mirroring technology is used to collect broadband users' internet access signaling. The broadband user's internet access signaling is parsed into XDR call detail records, and the user's internet access performance indicators are extracted based on the XDR call detail records; Based on the aforementioned user internet performance indicators, user internet speed evaluation survey data is collected to generate correlation data between user internet performance indicators and user internet speed evaluation.

3. The method for identifying slow internet users according to claim 1, characterized in that, The step of using the random forest algorithm to rank the importance of user internet performance indicators after dimensionality reduction, and selecting user internet performance indicators with an importance index greater than a preset threshold as key user internet performance indicators includes: Using a classification and regression tree algorithm, the user's online performance indicators after dimensionality reduction are recursively divided to generate multiple decision trees, forming a random forest. Calculate the internet performance index x of the j-th user based on the sum of the changes in the Gini coefficient before and after the branch at node k. j The importance score on the w-th decision tree, where k∈U, and U is the feature x. j The set of nodes that appear on the w-th decision tree; The importance scores of the internet performance index xj for the j-th user on all decision trees are calculated and normalized to obtain the final importance index of the internet performance index xj for the j-th user.

4. The method for identifying slow internet users according to claim 1, characterized in that, The process of calculating the fitting function curves for fast and slow internet access for each performance indicator based on fast and slow internet access samples, and determining the critical threshold for evaluating internet access speed for each performance indicator based on the inflection point where the two types of fitting function curves intersect on the coordinate axis, includes: Based on the fast and slow internet performance samples of users, the fitting function curves for fast and slow internet performance for each performance indicator are calculated using linear fitting, curve fitting, and exponential fitting. Find the inflection point where the curves of the two types of fitted functions intersect on the coordinate axis, and select the inflection point where the performance index changes from good to bad as the critical threshold for the performance index to change from fast to slow internet access.

5. The method for identifying slow internet users according to claim 4, characterized in that, The process of obtaining key performance indicators of the user's internet access and determining whether a user is a slow internet user involves comparing each key performance indicator with its corresponding threshold. If all key performance indicators exceed their corresponding critical thresholds, the user is identified as a slow internet user.

6. The method for identifying slow internet users according to claim 1, characterized in that, The process of obtaining key performance indicators of the user's internet access and determining whether a user is a slow internet user, after comparing each key performance indicator with its corresponding threshold, also includes: The importance index of key performance indicators for users with slow internet speeds is calculated by summing the data that exceed the corresponding critical threshold within a given period to determine the degree of slowness of the user's internet speed.

7. The method for identifying slow internet users according to claim 6, characterized in that, The determination of the user's internet slowness level is based on the sum of the importance indices of key performance indicators that exceed the corresponding critical threshold within a given period. Calculate the internet speed of a slow user in hour h: x j =S j (W)h j Where J represents the total number of key performance indicators, and S... j (W) represents the importance index of the j-th key performance indicator; Calculate the internet speed of a user with slow internet connection on a given day: Wherein, λ1, λ2, and λ3 are the weights corresponding to the busy / sluggishness of business in the current time period.

8. A device for identifying slow internet users, characterized in that, include: The data acquisition module is used to acquire related data on user internet performance indicators and user internet speed evaluation; The indicator screening module is used to perform correlation analysis and dimensionality reduction on the user's internet performance indicators based on the associated data, and to sort the importance of the dimensionality-reduced user internet performance indicators using the random forest algorithm, and to screen user internet performance indicators with an importance index greater than a preset threshold as key indicators of user internet performance. The threshold calculation module is used to calculate the fitting function curves of fast and slow internet access for each performance indicator based on the fast and slow internet access samples of the user's internet access performance indicators, and to determine the critical threshold for evaluating the speed of internet access for each performance indicator based on the inflection point when the two types of fitting function curves intersect on the coordinate axis. The slow internet access identification module is used to obtain key performance indicators of the user's internet access and to determine whether the user is a slow internet user by comparing the data of each key performance indicator with the corresponding critical threshold.

9. A device for identifying slow internet users, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the method for identifying slow internet users as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for identifying slow internet users as described in any one of claims 1 to 7.

Images