Broadband service poor-quality identification method, device, equipment and medium

Abstract

The invention provides a broadband service poor quality identification method, device and equipment and a medium, and relates to the technical field of communication, and the method comprises the steps: obtaining user use data generated when a user uses a broadband network; inputting the user use data into a trained poor-quality identification model to obtain a poor-quality identification result output by the poor-quality identification model; the poor quality identification result is used for indicating a user with a poor quality problem, and the poor quality identification model is obtained by training historical somatosensory poor quality data generated after the user uses a broadband network with preset network loss added to network parameters. According to the invention, active evaluation of the broadband service quality level is realized, so that each potential poor-quality user can be labeled accurately, problems can be found in advance, broadband maintenance can be carried out in time, and user experience and stickiness are improved.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a method, device, equipment and medium for identifying poor quality of broadband services. Background Art

[0002] As the number of broadband users increases, users' expectations for network quality are also increasing. Researching broadband quality differences is crucial to improving network quality and user experience. How to improve user satisfaction when using broadband has become an urgent demand of operators. However, the process of broadband services is complex and involves the user's subjective feelings. Problems such as slow Internet speed and video freezes will affect the user's subjective feelings, and the user's experience of using broadband is the key to affecting satisfaction.

[0003] There is often a certain difference between network quality and actual user service perception. The data generated by users when they are actually online cannot mark which data is of poor quality. For example, although the broadband network indicators are good, users will have a poor experience and complain about services that require low latency and low packet loss rate; the broadband network indicators are poor, but since there are no services on weekdays, users will not complain about poor experience. Therefore, how to accurately find users with poor network quality problems is the premise for improving broadband experience satisfaction. The identification of users with poor quality problems cannot simply rely on network indicator thresholds, but should be centered on user experience, so as to achieve measurable user experience. However, the existing poor quality identification methods cannot accurately evaluate the user's active perception, and there are problems such as the inability to effectively evaluate the quality of user experience and the inability to accurately identify users with poor quality. As a result, the problems complained by users cannot be quickly demarcated and located, resulting in operators lacking the ability to quickly close the loop to improve user experience.

[0004] It can be seen that how to objectively evaluate user perception of using broadband services is very important for fine-grained management of network quality and proactive problem discovery. Therefore, it is necessary to provide a broadband service poor quality identification method that can accurately evaluate user active perception and identify users with truly poor perception. Summary of the invention

[0005] In view of this, the present invention proposes a method, device, equipment and medium for identifying poor quality of broadband services to solve the problem of being unable to accurately evaluate user active perception and identify users with truly poor perception.

[0006] The technical solution of the present invention is achieved in this way:

[0007] According to a first aspect, an embodiment of the present invention provides a method for identifying poor quality of broadband services, the method comprising:

[0008] Obtain user usage data generated by users using broadband networks;

[0009] Inputting the user usage data into a trained quality difference recognition model to obtain a quality difference recognition result output by the quality difference recognition model; the quality difference recognition result is used to indicate users with quality difference problems, and the quality difference recognition model is trained based on historical perceptual quality difference data generated after the user uses a broadband network with a preset network loss added to the network parameters;

[0010] The quality difference recognition model is trained by the following steps:

[0011] Adding preset network losses to network parameters in the broadband network that exceed the abnormal impact level;

[0012] Obtain historical perceived quality difference data generated by users using a broadband network with preset network loss added to network parameters;

[0013] Perform data cleaning on historical sensory quality data to obtain historical network anomaly data;

[0014] The historical network abnormal data is used as input data for training, and preset noise is added to the historical network abnormal data and trained in an unsupervised learning manner, with the goal of controlling the similarity between the input data and the output data within a preset similarity level, thereby obtaining a noise reduction coding layer for outputting decoded data;

[0015] The historical decoded data output by the denoising coding layer is used as input data for training, and training is performed in an unsupervised manner to obtain an isolation forest layer for outputting poor quality recognition results.

[0016] In combination with the first aspect, in a first implementation of the first aspect, adding a preset network loss to a network parameter exceeding an abnormal impact degree in the broadband network specifically includes:

[0017] Acquire user network key performance indicator data of the broadband network, and extract network abnormality data from the user network key performance indicator data;

[0018] The abnormal impact degree of the network abnormal data is obtained, the network parameters corresponding to the network abnormal data exceeding the abnormal impact degree are arranged and combined, and the preset network loss is added to the combined network parameters.

[0019] In combination with the first aspect, in the second implementation of the first aspect, the data cleaning of the historical physical quality difference data to obtain the historical network abnormality data specifically includes:

[0020] Fill the empty values ​​in the historical somatosensory quality difference data;

[0021] The historical network anomaly data is extracted from the historical somatosensory quality difference data after null value filling.

[0022] In combination with the first aspect, in a third implementation of the first aspect, the noise reduction coding layer specifically includes:

[0023] Sequentially connected encoder input layer, noise addition layer, encoder hidden layer, and encoder output layer;

[0024] The encoder input layer is used to receive and pass downward the historical network abnormal data; the noise addition layer is used to add preset noise to the historical network abnormal data to obtain damaged data; the encoder hidden layer is used to encode the damaged data through the encoding function g(·) to obtain historical encoded data, and decode the historical encoded data through the decoding function f(·) to obtain historical decoded data; the encoder output layer is used to receive and pass the historical decoded data to the isolation forest layer.

[0025] In combination with the third implementation manner of the first aspect, in the fourth implementation manner of the first aspect, the loss function of the noise reduction coding layer is:

[0026]

[0027] Where L(·) represents the loss function; represents the damaged data of the i-th feature after adding preset noise to the historical network abnormal data; x i represents the historical network anomaly data of the i-th feature; n represents the total number of feature i; λ(ε i ≠0) represents a penalty term. When the damaged data of the i-th feature is not added with noise, a penalty term of size λ is added.

[0028] In combination with the third implementation manner of the first aspect, in the fifth implementation manner of the first aspect, inputting the user usage data into a trained quality difference identification model to obtain a quality difference identification result output by the quality difference identification model specifically includes:

[0029] The user usage data is input into the noise reduction coding layer, and the noise reduction coding layer adds preset noise to the user usage data and then performs encoding and decoding processing to obtain decoded data;

[0030] The decoded data is input into the isolation forest layer to obtain a quality difference identification result output by the isolation forest layer.

[0031] In combination with the first aspect, in a sixth implementation of the first aspect, the historical decoded data is used as input data for training, and training is performed in an unsupervised manner to obtain an isolation forest layer for outputting a poor quality recognition result, specifically including:

[0032] Filtering a preset amount of the historical decoding data and putting them into the root node of the tree;

[0033] A cutting point is randomly generated in the root node; the cutting point is located between the maximum value and the minimum value of the specified dimension in the root node data;

[0034] Divide the root node according to the cutting point, put the sample data below the specified dimension in one of the subspaces of the root node, and put the sample data not less than the specified dimension in another subspace of the root node;

[0035] Each root node is continuously divided until there is only one historical decoding data in the subspace of the root node or the subspace has reached a preset height, and the training of the isolation forest layer is completed.

[0036] According to a second aspect, an embodiment of the present invention further provides a broadband service quality poor identification device, the device comprising:

[0037] A data acquisition module, used to acquire user usage data generated by users using the broadband network;

[0038] A quality difference recognition module is used to input the user usage data into a trained quality difference recognition model to obtain a quality difference recognition result output by the quality difference recognition model; the quality difference recognition result is used to indicate users with quality difference problems, and the quality difference recognition model is trained based on historical perceptual quality difference data generated after the user uses a broadband network with a preset network loss added to the network parameters;

[0039] The quality difference recognition model is trained by the following steps:

[0040] Adding preset network losses to network parameters in the broadband network that exceed the abnormal impact level;

[0041] Obtain historical perceived quality difference data generated by users using a broadband network with preset network loss added to network parameters;

[0042] Perform data cleaning on historical sensory quality data to obtain historical network anomaly data;

[0043] The historical network abnormal data is used as input data for training, and preset noise is added to the historical network abnormal data and trained in an unsupervised learning manner, with the goal of controlling the similarity between the input data and the output data within a preset similarity level, thereby obtaining a noise reduction coding layer for outputting decoded data;

[0044] The historical decoded data output by the denoising coding layer is used as input data for training, and training is performed in an unsupervised manner to obtain an isolation forest layer for outputting poor quality recognition results.

[0045] According to a third aspect, an embodiment of the present invention further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any of the above-mentioned methods for identifying poor broadband service quality are implemented.

[0046] According to a fourth aspect, an embodiment of the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described methods for identifying poor quality of broadband services.

[0047] According to a fifth aspect, an embodiment of the present invention further provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the broadband service poor quality identification method as described in any one of the above items.

[0048] The broadband service poor quality identification method, device, equipment and medium of the present invention have the following beneficial effects compared with the prior art:

[0049] The quality difference problem is identified by the trained quality difference recognition model to obtain the quality difference recognition result. The quality difference recognition model adds a denoising coding layer in the form of an autoencoder before the isolation forest layer, which solves the problem that the traditional isolation forest algorithm can only perform linear segmentation, and retains the advantage that the isolation forest algorithm does not need to calculate the distance and density. It can greatly improve the speed and reduce the system overhead. The denoising coding layer can reduce the impact of reported erroneous data on the subsequent recognition accuracy and improve the anti-noise ability. The data used for training is obtained by adding network loss to the PON network, so that historical somatosensory quality difference data reflecting the user's real physical feeling can be constructed. The historical somatosensory quality difference data contains the user's real physical feeling information, thus filling the gap of unlabeled data in the broadband user quality difference recognition scenario, thereby improving the recognition accuracy and robustness of the quality difference recognition model, and realizing the active evaluation of the broadband service quality level, so as to accurately label each potential poor quality user, thereby discovering problems in advance and repairing the broadband in time, improving user experience and stickiness. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0051] Figure 1 A schematic diagram of a process of identifying poor quality of broadband services according to the present invention;

[0052] Figure 2It is a structural schematic diagram of the noise reduction coding layer in the broadband service quality difference identification method of the present invention;

[0053] Figure 3 It is a schematic diagram of the process of training the quality difference recognition model in the broadband service quality difference recognition method of the present invention;

[0054] Figure 4 It is a schematic diagram of a poor quality identification model in a method for identifying poor quality of broadband services of the present invention when performing service processing;

[0055] Figure 5 It is a structural schematic diagram of a broadband service poor quality identification device of the present invention;

[0056] Figure 6 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0057] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] As the number of broadband users increases, users' expectations for network quality are also increasing. Researching broadband quality differences is crucial to improving network quality and user experience. How to improve user satisfaction when using broadband has become an urgent demand of operators. However, the process of broadband services is complex and involves users' subjective feelings. Problems such as slow Internet access and video freezes will affect users' subjective feelings, and users' experience of using broadband is the key to affecting satisfaction. For example, the network link of home broadband services usually consists of terminal equipment, routers, switches, optical network units (ONU), optical distribution networks (ODN), passive optical networks (PON), optical line terminals (OLT), bearer networks, content source networks, etc. User satisfaction is affected by the use of these users' own terminals, as well as users' subjective behaviors and habits. There is often a certain difference between network quality and real user service perception. The data generated by users when they actually go online cannot mark which data is of poor quality. For example, although the broadband network indicators are good, users will have a poor experience and file complaints for services that require low latency and low packet loss rate; the broadband network indicators are poor, but since there are no services on weekdays, users will not complain due to poor experience. Therefore, how to accurately find users with poor network quality problems is the prerequisite for improving broadband experience satisfaction. The identification of users with poor quality problems cannot simply rely on network indicator thresholds, but should focus on user experience to achieve measurable user experience. However, the existing poor quality identification methods cannot accurately evaluate the user's active perception. There are problems such as the inability to effectively evaluate the quality of the user experience and the inability to accurately identify users with poor physical quality. As a result, the problems complained by users cannot be quickly demarcated and located, resulting in operators lacking the ability to quickly close the loop to improve user experience.

[0059] Existing methods for identifying poor quality cannot accurately evaluate users' active perception and identify users with truly poor perception. In summary, how to objectively evaluate users' perception of using broadband services is very important for fine-grained management of network quality and proactive problem discovery.

[0060] The method for identifying poor broadband service quality provided in this specification can be applied to electronic devices with broadband service communication data processing capabilities, aiming to more accurately identify broadband service poor quality problems and enhance the user experience. The electronic device may include a notebook, a desktop computer, a smart phone, a smart wearable device (virtual reality glasses, smart watches, etc.), a tablet computer, etc. Of course, the method for identifying poor broadband service quality provided in this specification can also be applied to applications running in the above-mentioned electronic devices. For example, the method for identifying poor broadband service quality can be applied to a browser with broadband service communication data processing capabilities, and can also be applied to software with broadband service communication data processing capabilities.

[0061] See also Figure 1 , Figure 1 A flowchart of a method for identifying poor quality of broadband services according to an embodiment of the present invention is shown. The method may include the following steps:

[0062] S101: Obtain user usage data generated by a user using a broadband network.

[0063] In this embodiment, the user usage data may be stored in the electronic device in advance, or may be obtained by the electronic device from the outside. For example, the electronic device obtains the data from an external broadband service collection device. There is no restriction on the specific acquisition form of the user usage data, and it is only necessary to ensure that the electronic device can obtain the user usage data.

[0064] S102. Input the user usage data into a trained quality difference recognition model to obtain a quality difference recognition result output by the quality difference recognition model, wherein the quality difference recognition result is used to indicate users with quality difference problems. The quality difference recognition model is trained based on historical perceived quality difference data generated when users use a broadband network with a preset network loss added to network parameters.

[0065] By adding network loss to the PON network, historical physical quality difference data that reflects the user's real physical feeling can be constructed. The historical physical quality difference data contains the user's real physical feeling information, thus filling the gap of unlabeled data in the broadband user quality difference identification scenario.

[0066] In this embodiment, the quality difference recognition model is trained in the following way:

[0067] S201. Add preset network losses to network parameters exceeding the abnormal impact level in the broadband network.

[0068] Specifically, step S201 includes:

[0069] S2021. Obtain user network key performance indicator (KPI) data of the broadband network, and extract network abnormality data from the user KPI data.

[0070] Similarly, in this embodiment, the user KPI data may be stored in advance in the electronic device, or may be obtained by the electronic device from the outside. For example, the electronic device obtains the data from an external broadband service collection device. There is no restriction on the specific acquisition form of the user KPI data, and it is only necessary to ensure that the electronic device can obtain the user KPI data.

[0071] Since the user KPI data contains a lot of data, it is also necessary to extract network anomaly data with a high correlation with network anomalies. In this embodiment, the network anomaly data includes eight types of data: average round-trip time on the user side (networkSideAverageRtt), average jitter of the network side round-trip time (networkSideRttAverageJitter), average round-trip time on the user side (userSideAverageRtt), average round-trip time on the user side (userSideRttAverageJitter), downlink network side packet loss rate (flowDownLinkNetworkSideDiscardRatio), uplink network side packet loss rate (flowUplinkNetworkSideDiscardRatio), downlink user side packet loss rate (flowDownLinkUserSideDiscardRatio) and uplink user side packet loss rate (flowUplinkUserSideDiscardRatio).

[0072] Preferably, the user KPI data is derived from the report of the OLT board. The user_kpi data reported by the OLT board has many fields, so it is also necessary to extract the fields highly correlated with the network anomaly.

[0073] The fields with high correlation with network anomalies include user side average round-trip time (networkSideAverageRtt), network side round-trip time average jitter (networkSideRttAverageJitter), user side average round-trip time (userSideAverageRtt), user side round-trip time average jitter (userSideRttAverageJitter), total number of downstream packets (flowDownLinkTotalPackets), number of downstream network side packet losses (flowDownLinkNetworkSideDiscardPackets), number of downstream user side packet losses (flowDownLinkUserSideDiscardPackets), total number of upstream packets (flowUplinkTotalPackets), number of upstream network side packet losses (upstream network side packet losses), and number of upstream user side packet losses (upstream user side packet losses).

[0074] Then, according to the number of lost packets and the total number of packets, the required packet loss rate data is obtained, where:

[0075]

[0076] In formula (1), flowDownLinkNetworkSideDiscardRatio represents the packet loss rate on the downlink network side.

[0077]

[0078] In formula (2), flowDownLinkUserSideDiscardRatio represents the downlink user side packet loss rate.

[0079]

[0080] In formula (3), flowUpLinkNetworkSideDiscardRatio represents the uplink network side packet loss rate.

[0081]

[0082] In formula (4), flowUpLinkUserSideDiscardRatio represents the uplink user side packet loss rate.

[0083] S2022. Obtain the abnormal impact degree of the network abnormal data, arrange and combine the network parameters corresponding to the network abnormal data exceeding the abnormal impact degree, and add a preset network loss to the combined network parameters.

[0084] Among the network anomaly data, the average jitter and uplink packet loss rate have a smaller impact on the perceived quality, while the other four data have a greater impact on the perceived quality. The network parameters that generate these four data (including round-trip time and packet loss rate) are then arranged and combined, and a total of 15 combinations can be obtained. In each combination of network parameters, network loss parameters with multiple gradient parameter values ​​are set. For example, the round-trip time is increased from 100ms to 300ms by a network damage meter, with each increase span being 50ms or 100ms, and the packet loss rate is increased from 10% to 25% by a network damage meter, with each increase span being 5% or 10%.

[0085] By arranging and combining the network abnormality data that exceeds the preset abnormal impact level and adding the preset network loss, the network loss with specific parameters can be added to the PON network. When the ONU is connected to the router, the user terminal can experience the network status under the preset network loss with the parameter after connecting to the broadband.

[0086] S202: Obtain historical perceived quality difference data generated after the user uses the broadband network with a preset network loss added to the network parameters. It can be understood that the historical perceived quality difference data is also the obtained user KPI data.

[0087] When users use various applications in a broadband network with added network loss, various types of user data will be generated. By reading the user data of the application from systems such as Kafka and recording the time period when the physical quality is poor, the data is marked as 1, and the data of the time period when the physical quality is not poor is marked as 0. In this way, the historical physical quality difference data is obtained.

[0088] S203: Clean the historical sensory quality data to obtain historical network abnormal data.

[0089] In this embodiment, data cleaning includes null value filling and feature screening. Specifically, step S203 includes:

[0090] S2031, fill in the empty values ​​in the historical somatosensory quality difference data. For the empty values ​​that may exist in the historical somatosensory quality difference data, fill in the data with 0 values.

[0091] S2032. Extract historical network anomaly data from the historical somatosensory quality differences after null value filling.

[0092] In this embodiment, when obtaining the packet loss rate of abnormal network data in step S202, the upper limit of the packet loss rate is set to 1 to prevent dirty data from causing the packet loss rate to be greater than 1. For other extraction processes, please refer to step S2021.

[0093] S204, using the historical network anomaly data as input data for training, adding preset noise to the historical network anomaly data and training in an unsupervised learning manner, with the goal of controlling the similarity between the input data and the output data within a preset similarity level, thereby obtaining a noise reduction coding layer for outputting decoded data.

[0094] There are two problems with the isolation forest in the existing technology: first, abnormal data can only be segmented on a single feature plane at a time, which leads to the problem of deviation in the uncovered area of ​​the data; second, when processing high-dimensional data, the distance between data points becomes blurred, making it difficult to accurately detect abnormal points.

[0095] An autoencoder is a deep learning model used for unsupervised learning. The goal of an autoencoder is to compress input data into a low-dimensional representation and then reconstruct the original data from the low-dimensional representation. In this embodiment, the denoising autoencoder is based on the autoencoder. In order to prevent overfitting problems, noise is added to the input data of the input layer, so that the trained denoising encoder has better robustness.

[0096] Since the collected historical data may contain a small amount of errors, the source data is mixed with noise data. The noise factor is added to the loss function of the denoising autoencoder. In this way, in addition to ensuring that the reconstructed data is close to the original data, the denoising autoencoder will also provide additional rewards when reconstructing the original data with a lot of noise, thereby improving the model's denoising ability.

[0097] The quality difference recognition model adds an improved noise reduction encoder to the isolation forest algorithm, which is used to learn nonlinear data relationships and compress high-dimensional data. The isolation forest algorithm with the noise reduction encoder can effectively solve the problem that the existing isolation forest algorithm can only segment abnormal data on a single feature plane at a time and the distance between data points becomes blurred.

[0098] See also Figure 2 ,The network structure of the denoising autoencoder includes the encoder input layer, the noise adding layer, the encoder hidden layer and the encoder output layer.

[0099] Among them, the encoder input layer is used to input historical network abnormal data, and the noise addition layer is used to add preset noise to the historical network abnormal data. Specifically:

[0100]

[0101] In formula (5), represents the damaged data of the i-th feature with the preset noise added; x irepresents the original historical network anomaly data of the i-th feature; p represents the preset probability, that is, the noise adding layer has a probability of p to set the historical network anomaly data to 0, and a probability of 1-p to keep the original value unchanged.

[0102] The encoder hidden layer is used to encode the damaged data through the encoding function g(·) to obtain the historical encoded data, and decode the historical encoded data through the decoding function f(·) to obtain the historical decoded data. Specifically, the encoding process is:

[0103]

[0104] In formula (6), Indicates damaged data; h indicates damaged data The historical encoded data (features) of the encoding process; u represents the activation function; w1 represents the first weight, which is the weight from the encoder input layer to the encoder hidden layer; b1 represents the first bias term, which is the bias term of the encoder hidden layer; the activation function u is specifically:

[0105] u(z)=(1+e -z ) -1 (7)

[0106] The decoding process is:

[0107] y=f(h)=u(w2h+b2) (8)

[0108] In formula (8), y represents the reconstructed historical decoding data (features); w2 represents the second weight, which is the weight from the encoder hidden layer to the encoder output layer; b2 represents the second bias term, which is the bias term of the encoder output layer.

[0109] In this embodiment, in order to determine that the similarity or proximity between the data received and transmitted downward by the encoder input layer and the subsequent data input by the encoder output layer is within a preset similarity range, in this embodiment, a loss function L(·) is set in the denoising encoder, specifically:

[0110]

[0111] In formula (9), n represents the total number of feature i; λ(ε i ≠0) indicates that when the damaged data of the i-th feature has no noise added, a penalty term of size λ is added. In this way, by adding additional penalty terms, a balance point of the model training process can be found, so that the model can effectively denoise and maintain the stability of model training.

[0112] Assuming that the parameters θ = (w1, b1, w2, b2), the parameters θ can be obtained by minimizing the loss function L(·). The objective function J(·) of the improved denoising autoencoder is set to:

[0113] J(θ)=∑ x∈s L(x,f(g(x))) (10)

[0114] In formula (10), S represents the total number of damaged data.

[0115] S205, using the historical decoding data as input data for training, and training in an unsupervised manner to obtain an isolation forest layer for outputting poor quality recognition results

[0116] See also Figure 3 After the denoising encoder is trained, the improved denoising encoder is used as a feature extractor, and the historical decoding data output by the denoiser output layer is used as the input data of the subsequent network structure. In this embodiment, the subsequent network structure is an isolation forest layer, and the processing method of the isolation forest layer is:

[0117] S2051. Randomly select a preset number of N sample data from the historical decoded data as sub-samples, and put them into the root node of the tree.

[0118] S2052. Randomly set a specified dimension and randomly generate a cutting point in the root node. The cutting point is generated between the maximum value and the minimum value of the specified dimension in the current root node data.

[0119] S2053. A hyperplane is generated based on this cutting point, and then the current root node data space is divided into two sub-nodes, i.e., sub-spaces, based on the cutting point: sample data below the specified dimension is placed on the left side of the current root node, and sample data not less than the specified dimension is placed on the right side of the current root node.

[0120] S2054, recursively perform steps S2052 and S2053 in the subspace of the current root node, and continuously construct new subspaces until only one sample data in the subspace cannot be cut further or the subspace has reached a preset height. After obtaining t subtrees, the training is finished.

[0121] In this embodiment, the isolation forest layer does not output true or false, but outputs a value of 0-1, which represents the degree of abnormality of the data. The larger the value, the higher the degree of abnormality.

[0122] Accordingly, step S102 specifically includes:

[0123] S1021, inputting the user usage data into the noise reduction coding layer, the noise reduction coding layer adds preset noise to the user usage data and then performs encoding and decoding processing to obtain decoded data;

[0124] S1022, input the decoded data into the isolation forest layer to obtain the quality difference recognition result output by the isolation forest layer.

[0125] See also Figure 4 Among them, the user usage data can be stored in Kafka, and then the quality difference recognition model can be integrated into the Spark task. The quality difference recognition model obtains the quality difference recognition result by consuming the data in Kafka and stores it in ElasticSearch.

[0126] The broadband service quality difference identification method of the present invention identifies the quality difference problem through a trained quality difference identification model to obtain the quality difference identification result. The quality difference identification model is to add a noise reduction coding layer in the form of an autoencoder before the isolation forest layer, which solves the problem that the traditional isolation forest algorithm can only perform linear segmentation, and retains the advantage that the isolation forest algorithm does not need to calculate the relevant distance and density, can greatly improve the speed, reduce system overhead, and the noise reduction coding layer can reduce the influence of reported erroneous data on the subsequent recognition accuracy, improve the anti-noise ability, and the data used for training is obtained by adding network loss to the PON network, so that historical somatosensory quality difference data reflecting the real somatosensory of the user can be constructed, and the historical somatosensory quality difference data contains the real somatosensory information of the user, thereby making up for the blank of data without label in the broadband user quality difference identification scenario, thereby improving the recognition accuracy and robustness of the quality difference identification model, and realizing the active evaluation of the broadband service quality level, so as to accurately label each potential quality difference user, thereby discovering problems in advance and performing broadband maintenance in time, and improving user experience and stickiness.

[0127] The following describes an apparatus provided by an embodiment of the present invention. The apparatus described below and the method described above can refer to each other.

[0128] See also Figure 5 , Figure 5 A schematic diagram showing the structure of a device for identifying poor quality of broadband services according to an embodiment of the present invention is shown. The device may include:

[0129] The data acquisition module 10 is used to acquire user usage data generated by users using the broadband network.

[0130] In this embodiment, the user usage data may be stored in the electronic device in advance, or may be obtained by the electronic device from the outside. For example, the electronic device obtains the data from an external broadband service collection device. There is no restriction on the specific acquisition form of the user usage data, and it is only necessary to ensure that the electronic device can obtain the user usage data.

[0131] The quality difference identification module 20 is used to input user usage data into a trained quality difference identification model to obtain a quality difference identification result output by the quality difference identification model, wherein the quality difference identification result is used to indicate users with quality difference problems. The quality difference identification model is trained based on historical perceived quality difference data generated after users use a broadband network with a preset network loss added to the network parameters.

[0132] By adding network loss to the PON network, historical physical quality difference data that reflects the user's real physical feeling can be constructed. The historical physical quality difference data contains the user's real physical feeling information, thus filling the gap of unlabeled data in the broadband user quality difference identification scenario.

[0133] The broadband service quality difference identification device of the present invention identifies the quality difference problem through a trained quality difference identification model to obtain the quality difference identification result. The quality difference identification model is to add a noise reduction coding layer in the form of an autoencoder before the isolation forest layer, which solves the problem that the traditional isolation forest algorithm can only perform linear segmentation, and retains the advantage that the isolation forest algorithm does not need to calculate the relevant distance and density, which can greatly improve the speed and reduce the system overhead. The noise reduction coding layer can reduce the influence of reported erroneous data on the subsequent recognition accuracy and improve the anti-noise ability. The data used for training is obtained by adding network loss to the PON network, so that historical somatosensory quality difference data reflecting the real somatosensory of the user can be constructed. The historical somatosensory quality difference data contains the real somatosensory information of the user, thus making up for the blank of data without label in the broadband user quality difference identification scenario, thereby improving the recognition accuracy and robustness of the quality difference identification model, and realizing the active evaluation of the broadband service quality level, so as to accurately label each potential quality difference user, thereby discovering problems in advance and performing broadband maintenance in time, and improving user experience and stickiness.

[0134] Figure 6 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 6 As shown, the electronic device may include: a processor 610 (processor), a communication interface 620 (Communications Interface), a memory 630 (memory) and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 may call the logic command in the memory 630 to execute the broadband service quality difference identification method, which includes:

[0135] Obtain user usage data generated by users using broadband networks;

[0136] Inputting the user usage data into a trained quality difference recognition model to obtain a quality difference recognition result output by the quality difference recognition model; the quality difference recognition result is used to indicate users with quality difference problems, and the quality difference recognition model is trained based on historical perceptual quality difference data generated after the user uses a broadband network with a preset network loss added to the network parameters;

[0137] The quality difference recognition model is trained by the following steps:

[0138] Adding preset network losses to network parameters in the broadband network that exceed the abnormal impact level;

[0139] Obtain historical perceived quality difference data generated by users using a broadband network with preset network loss added to network parameters;

[0140] Perform data cleaning on historical sensory quality data to obtain historical network anomaly data;

[0141] The historical network abnormal data is used as input data for training, and preset noise is added to the historical network abnormal data and trained in an unsupervised learning manner, with the goal of controlling the similarity between the input data and the output data within a preset similarity level, thereby obtaining a noise reduction coding layer for outputting decoded data;

[0142] The historical decoded data output by the denoising coding layer is used as input data for training, and training is performed in an unsupervised manner to obtain an isolation forest layer for outputting poor quality recognition results.

[0143] In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memor), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0144] On the other hand, the present invention further provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, when the program instructions are executed by a computer, the computer can execute the broadband service quality difference identification method provided by the above methods, the method comprising:

[0145] Obtain user usage data generated by users using broadband networks;

[0146] Inputting the user usage data into a trained quality difference recognition model to obtain a quality difference recognition result output by the quality difference recognition model; the quality difference recognition result is used to indicate users with quality difference problems, and the quality difference recognition model is trained based on historical perceptual quality difference data generated after the user uses a broadband network with a preset network loss added to the network parameters;

[0147] The quality difference recognition model is trained by the following steps:

[0148] Adding preset network losses to network parameters in the broadband network that exceed the abnormal impact level;

[0149] Obtain historical perceived quality difference data generated by users using a broadband network with preset network loss added to network parameters;

[0150] Perform data cleaning on historical sensory quality data to obtain historical network anomaly data;

[0151] The historical network abnormal data is used as input data for training, and preset noise is added to the historical network abnormal data and trained in an unsupervised learning manner, with the goal of controlling the similarity between the input data and the output data within a preset similarity level, thereby obtaining a noise reduction coding layer for outputting decoded data;

[0152] The historical decoded data output by the denoising coding layer is used as input data for training, and training is performed in an unsupervised manner to obtain an isolation forest layer for outputting poor quality recognition results.

[0153] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for identifying poor quality of broadband services provided above is implemented, and the method comprises:

[0154] Obtain user usage data generated by users using broadband networks;

[0155] Inputting the user usage data into a trained quality difference recognition model to obtain a quality difference recognition result output by the quality difference recognition model; the quality difference recognition result is used to indicate users with quality difference problems, and the quality difference recognition model is trained based on historical perceptual quality difference data generated after the user uses a broadband network with a preset network loss added to the network parameters;

[0156] The quality difference recognition model is trained by the following steps:

[0157] Adding preset network losses to network parameters in the broadband network that exceed the abnormal impact level;

[0158] Obtain historical perceived quality difference data generated by users using a broadband network with preset network loss added to network parameters;

[0159] Perform data cleaning on historical sensory quality data to obtain historical network anomaly data;

[0160] The historical network abnormal data is used as input data for training, and preset noise is added to the historical network abnormal data and trained in an unsupervised learning manner, with the goal of controlling the similarity between the input data and the output data within a preset similarity level, thereby obtaining a noise reduction coding layer for outputting decoded data;

[0161] The historical decoded data output by the denoising coding layer is used as input data for training, and training is performed in an unsupervised manner to obtain an isolation forest layer for outputting poor quality recognition results.

[0162] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0163] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying poor quality of broadband services, characterized in that: The method comprises: Obtain user usage data generated by users using broadband networks; Inputting the user usage data into a trained quality difference recognition model to obtain a quality difference recognition result output by the quality difference recognition model; the quality difference recognition result is used to indicate users with quality difference problems, and the quality difference recognition model is trained based on historical perceptual quality difference data generated after the user uses a broadband network with a preset network loss added to the network parameters; The quality difference recognition model is trained by the following steps: Adding preset network losses to network parameters in the broadband network that exceed the abnormal impact level; Obtain historical perceived quality difference data generated by users using a broadband network with preset network loss added to network parameters; Perform data cleaning on historical sensory quality data to obtain historical network anomaly data; The historical network abnormal data is used as input data for training, and preset noise is added to the historical network abnormal data and trained in an unsupervised learning manner, with the goal of controlling the similarity between the input data and the output data within a preset similarity level, thereby obtaining a noise reduction coding layer for outputting decoded data; The historical decoded data output by the denoising coding layer is used as input data for training, and training is performed in an unsupervised manner to obtain an isolation forest layer for outputting poor quality recognition results.

2. The method for identifying poor broadband service quality as claimed in claim 1, wherein: The adding of preset network loss to the network parameters exceeding the abnormal impact degree in the broadband network specifically includes: Acquire user network key performance indicator data of the broadband network, and extract network abnormality data from the user network key performance indicator data; The abnormal impact degree of the network abnormal data is obtained, the network parameters corresponding to the network abnormal data exceeding the abnormal impact degree are arranged and combined, and the preset network loss is added to the combined network parameters.

3. The method for identifying poor quality of broadband services as claimed in claim 1, wherein: The data cleaning of the historical sensory quality data to obtain the historical network abnormal data specifically includes: Fill the empty values ​​in the historical somatosensory quality difference data; The historical network anomaly data is extracted from the historical somatosensory quality difference data after null value filling.

4. The method for identifying poor quality of broadband services as claimed in claim 1, wherein: The noise reduction coding layer specifically includes: Sequentially connected encoder input layer, noise addition layer, encoder hidden layer, and encoder output layer; The encoder input layer is used to receive and pass downward the historical network abnormal data; the noise addition layer is used to add preset noise to the historical network abnormal data to obtain damaged data; the encoder hidden layer is used to encode the damaged data through the encoding function g(·) to obtain historical encoded data, and decode the historical encoded data through the decoding function f(·) to obtain historical decoded data; the encoder output layer is used to receive and pass the historical decoded data to the isolation forest layer.

5. The method for identifying poor quality of broadband services as claimed in claim 4, characterized in that: The loss function of the denoising coding layer is: Where L(·) represents the loss function; represents the damaged data of the i-th feature after adding preset noise to the historical network abnormal data; x i represents the historical network anomaly data of the i-th feature; n represents the total number of feature i; λ(ε i ≠0) represents a penalty term. When the damaged data of the i-th feature is not added with noise, a penalty term of size λ is added.

6. The method for identifying poor quality of broadband services as claimed in claim 4, characterized in that: The step of inputting the user usage data into a trained quality difference recognition model to obtain a quality difference recognition result output by the quality difference recognition model specifically includes: The user usage data is input into the noise reduction coding layer, and the noise reduction coding layer adds preset noise to the user usage data and then performs encoding and decoding processing to obtain decoded data; The decoded data is input into the isolation forest layer to obtain a quality difference identification result output by the isolation forest layer.

7. The method for identifying poor quality of broadband services as claimed in claim 1, wherein: The method of using the historical decoded data as input data for training and training in an unsupervised manner to obtain an isolation forest layer for outputting poor quality recognition results specifically includes: Filtering a preset amount of the historical decoding data and putting them into the root node of the tree; A cutting point is randomly generated in the root node; the cutting point is located between the maximum value and the minimum value of the specified dimension in the root node data; Divide the root node according to the cutting point, put the sample data below the specified dimension in one of the subspaces of the root node, and put the sample data not less than the specified dimension in another subspace of the root node; Each root node is continuously divided until there is only one historical decoding data in the subspace of the root node or the subspace has reached a preset height, and the training of the isolation forest layer is completed.

8. A broadband service poor quality identification device, characterized in that: The device comprises: A data acquisition module, used to acquire user usage data generated by users using the broadband network; A quality difference recognition module is used to input the user usage data into a trained quality difference recognition model to obtain a quality difference recognition result output by the quality difference recognition model; the quality difference recognition result is used to indicate users with quality difference problems, and the quality difference recognition model is trained based on historical perceptual quality difference data generated after the user uses a broadband network with a preset network loss added to the network parameters; The quality difference recognition model is trained by the following steps: Adding preset network losses to network parameters in the broadband network that exceed the abnormal impact level; Obtain historical perceived quality difference data generated by users using a broadband network with preset network loss added to network parameters; Perform data cleaning on historical sensory quality data to obtain historical network anomaly data; The historical network abnormal data is used as input data for training, and preset noise is added to the historical network abnormal data and trained in an unsupervised learning manner, with the goal of controlling the similarity between the input data and the output data within a preset similarity level, thereby obtaining a noise reduction coding layer for outputting decoded data; The historical decoded data output by the denoising coding layer is used as input data for training, and training is performed in an unsupervised manner to obtain an isolation forest layer for outputting poor quality recognition results.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method for identifying poor quality of broadband services as described in any one of claims 1 to 7 are implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for identifying poor quality of broadband services as claimed in any one of claims 1 to 7 are implemented.

Images