Determination method and device for poor-quality user, storage medium and electronic device

Abstract

The embodiment of the invention provides a poor-quality user determination method and device, a storage medium and an electronic device. The method comprises the steps of obtaining matrix data and feature data based on first session level data XDR associated with a user tag; and inputting the matrix data and the feature data into a deep learning model to obtain a poor-quality probability, and determining a poor-quality user according to the poor-quality probability. According to the embodiment of the invention, the poor quality judgment is carried out based on the session level data XDR, so that all information of a user side network can be integrated; matrix data and feature data obtained on the basis of XDR are input into the deep learning model, automatic feature analysis can be performed through the deep learning model, meanwhile, the poor quality probability of the user can be accurately determined through the deep learning method, and more accurate network quality can be obtained. Therefore, the problem of low accuracy of the existing network quality judgment method in the related technology can be solved, and the effect of improving the accuracy of judging the network quality is achieved.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of communications, and in particular, to a method, device, storage medium and electronic device for determining a poor quality user. Background Art

[0002] In the face of a large group of home network users, to ensure high-quality Internet access, it is necessary to judge the quality of home networks in a timely and accurate manner. In the face of the complexity of home networks, there are few existing methods for judging network quality, and there are problems such as high cost, low efficiency, and low accuracy. For example, asking users to rate the network through telephone interviews to confirm whether the user's network is of poor quality is a method with poor operability, low accuracy, and high cost. Another example is to judge the perception of poor quality based on all the user's Internet access information. This method will reduce the accuracy of the judgment result because all Internet access information may include interference information such as keep-alive heartbeats and website advertising pop-ups. Another example is to use the threshold comparison method to compare whether several key indicators are abnormal, and finally determine whether the user is of poor quality. This method has the problem of low accuracy due to incomplete network information. Summary of the invention

[0003] The embodiments of the present invention provide a method, device, storage medium and electronic device for determining poor-quality users, so as to at least solve the problem of low accuracy of the existing network quality judgment methods in the related art.

[0004] According to one embodiment of the present invention, a method for determining poor quality users is provided, comprising: obtaining matrix data and feature data based on first session-level data XDR associated with a user tag; inputting the matrix data and the feature data into a deep learning model to obtain poor quality probability, and determining poor quality users based on the poor quality probability.

[0005] According to another embodiment of the present invention, a device for determining poor quality users is provided, comprising: an acquisition module for acquiring matrix data and feature data based on first session-level data XDR associated with a user tag; a determination module for inputting the matrix data and the feature data into a deep learning model to obtain poor quality probability, and determining poor quality users based on the poor quality probability.

[0006] According to yet another embodiment of the present invention, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to execute the steps of any one of the above method embodiments when run.

[0007] According to yet another embodiment of the present invention, there is provided an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0008] Through the above-mentioned embodiment of the present invention, since the quality judgment is based on the session-level data XDR, all the information of the user-side network can be integrated; and the matrix data and feature data obtained based on XDR are input into the deep learning model, and automatic feature analysis can be performed through the deep learning model. At the same time, the deep learning method can also accurately determine the user's quality probability and obtain more accurate network quality. Therefore, the problem of low accuracy of the existing network quality judgment method in the related art can be solved, and the effect of improving the accuracy of judging network quality can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 It is a hardware structure block diagram of a computer terminal for the method for determining poor-quality users according to an embodiment of the present invention;

[0010] Figure 2 is a schematic diagram of a home network architecture according to an embodiment of the present invention;

[0011] Figure 3 is a flow chart of a method for determining a poor quality user according to an embodiment of the present invention;

[0012] Figure 4 is a structural block diagram of a device for determining a poor quality user according to an embodiment of the present invention;

[0013] Figure 5 is a structural block diagram of a device for determining poor quality users according to another embodiment of the present invention;

[0014] Figure 6 is a structural block diagram of a quality difference judgment system according to an embodiment of the present invention;

[0015] Figure 7 is a flow chart of data processing steps according to an embodiment of the present invention;

[0016] Figure 8 is a schematic diagram of a deep learning process based on a deep learning model according to an embodiment of the present invention;

[0017] Fig. 9 is a flowchart of deep learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings and in combination with the embodiments.

[0019] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0020] The method embodiments provided in the embodiments of the present application can be executed in a computer terminal, a computer terminal or a similar computing device. Taking running on a computer terminal as an example, Figure 1 1 is a hardware structure block diagram of a computer terminal for the method for determining poor quality users according to an embodiment of the present invention. Figure 1 As shown, the computer terminal may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the above-mentioned computer terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above-mentioned computer terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0021] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the method for determining poor quality users in the embodiment of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, to implement the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0022] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of a computer terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (Radio Frequency, referred to as RF) module, which is used to communicate with the Internet wirelessly. The embodiment of the present invention is used for user fixed-line home broadband services, mainly for PON (Passive Optical Network, passive optical network) network structure. PON is a network structure that does not contain any electronic devices and electronic power supplies in the optical distribution network.

[0023] The embodiment of the present invention can be run on Figure 2 The home network architecture shown in FIG. 1 includes a passive optical network (PON), such as Figure 2 As shown, the PON includes: an optical line terminal (OLT) installed in a central control station and a group of supporting optical network units (ONU) installed in user locations. The optical distribution network (ODN) between the OLT and the ONU includes optical fibers and passive splitters or couplers.

[0024] Specifically, OLT is one of the devices on the access network service node, which can be used to control, manage, and measure the distance to customer terminal devices such as ONU. It usually includes: control board, DC power board, fan unit, service frame, and other upstream boards.

[0025] One side of the OLT is connected to the switch at the coverage level through a cable and converted into an optical signal, and on the other side, it is connected to the optical splitter at the customer terminal through an optical fiber.

[0026] ODN is a FTTH (Fiber To The Home) optical cable network based on PON equipment. Its function is to provide an optical transmission channel between OLT and ONU. From the functional point of view, ODN can be divided into four parts from the central office to the user end: feeder optical cable subsystem, distribution optical cable subsystem, drop line optical cable subsystem and optical fiber terminal subsystem.

[0027] ONU is divided into active optical network unit and passive optical network unit.

[0028] In ONU, the equipment equipped with optical receiver, upstream optical transmitter, multiple bridge amplifiers and network monitoring is generally called an optical node.

[0029] ONU can provide data, IPTV (interactive Internet protocol television), voice (using IAD (Integrated Access Device)) and other services. Common home ONU is generally called optical modem.

[0030] like Figure 2 As shown, the home network structure includes not only the above structure but also terminals, i.e. Internet access devices such as mobile phones, computers, etc.

[0031] In this embodiment, a method for determining poor quality users running on the above-mentioned computer terminal or home network architecture is provided. Figure 3 is a flowchart of a method for determining a poor quality user according to an embodiment of the present invention, which method can be applied to Figure 2 The OLT in the system is a built-in DPI (Deep Packet Inspection) blade in the OLT to monitor Internet applications and perceive the real business traffic of users.

[0032] like Figure 3 As shown, the process includes the following steps:

[0033] Step S302, acquiring matrix data and feature data based on first session-level data XDR (Extended Detection and Response) associated with the user tag.

[0034] In this embodiment, the first XDR is the user's medium- to long-term (e.g., more than 1 week) XDR cumulative data. In this way, the information contained in the XDR is richer, and when determining poor-quality users, all information on the user-side network can be integrated to obtain a more accurate judgment result.

[0035] Before step S302 of this embodiment, the method further includes: associating a user tag with the first XDR through a user account, wherein the user tag is fuzzy processed.

[0036] Fuzzifying the user label associated with the first XDR can reduce the uncertainty brought by the user's answer. For example, when the user revisits, users who have no feedback on poor quality can be marked as 0.2, those who feedback on poor quality but do not need on-site repair can be marked as 0.8, and those who feedback on poor quality and need on-site repair can be marked as 1.

[0037] In step S302 of this embodiment, matrix data and feature data are obtained based on the first XDR associated with the user tag, including: filtering out non-TCP (Transmission Control Protocol) protocol XDRs in the first XDR to obtain a second XDR; classifying the second XDRs, and performing background traffic filtering on the classified second XDRs respectively; deleting linear correlation features in each type of the second XDRs respectively to obtain third XDRs corresponding to each type of the second XDRs; aggregating different types of the third XDRs to obtain the matrix data; and extracting features from each type of the third XDRs respectively to obtain the feature data.

[0038] In one embodiment, the second XDR is classified, including: classifying the video data of the HTTP protocol (Hyper Text Transfer Protocol) in the second XDR into the first category of data; classifying the non-video data of the HTTP protocol in the second XDR into the second category of data; classifying the game data in the second XDR into the third category of data; and classifying the data that cannot be extracted in the second XDR into the fourth category of data.

[0039] In one embodiment, the classified second XDR is subjected to background traffic filtering, including at least one of the following: filtering data of overseas IP (Internet Protocol) addresses; resolving service categories through IP addresses, determining service subcategories of background traffic according to the service categories, and filtering all data of the service subcategories; using the user's nighttime data as a background traffic benchmark, and filtering the user's background traffic throughout the day based on the benchmark.

[0040] In one embodiment, the matrix data includes the following information: data type, data index, and acquisition time of the first XDR.

[0041] In one embodiment, the data indicators include but are not limited to one of the following: the delay of the TCP protocol three-way handshake, the average value of the data upload delay, the average value of the data download delay, and the data upload retransmission rate.

[0042] Step S304: input the matrix data and the feature data into a deep learning model to obtain poor quality probability, and determine poor quality users according to the poor quality probability.

[0043] In step S304 of this embodiment, the matrix data and the feature data are input into a deep learning model to obtain the quality difference probability, including: performing deep feature encoding on the feature data, and performing convolution encoding on the matrix data; splicing the result of the deep feature encoding with the result of the convolution encoding, and performing feature integration on the spliced ​​data through a neural network; and converting the result of the feature integration into the quality difference probability.

[0044] Before step S304 of this embodiment, the method further includes: training a deep learning model.

[0045] Training a deep learning model involves:

[0046] Step 1: Dataset preparation.

[0047] Collect sufficient amount of user XDR data, and associate the tags of returning users with user data (XDR call records) through user accounts.

[0048] Step 2: Model training.

[0049] Input the associated user XDR into a deep learning model to obtain the output result;

[0050] The output results and labels of the deep learning model are subjected to cross entropy loss, and SGD (Stochastic Gradient Descent) is used for back propagation to correct the deep model parameters.

[0051] After multiple iterations, the required model is finally obtained.

[0052] In one embodiment, the XDR starting from the current time node and accumulated for a period of time thereafter is used as a training data set to perform model training.

[0053] In one embodiment, the deep learning model obtained through other training is migrated to the current application site, and the migrated deep learning model is used as the model initialization parameter for subsequent training.

[0054] Through the above steps, since the quality judgment is based on the session-level data XDR, all the information of the user-side network can be integrated; and the matrix data and feature data obtained based on XDR are input into the deep learning model, and automatic feature analysis can be performed through the deep learning model. At the same time, the deep learning method can also accurately determine the user's quality probability and obtain more accurate network quality. Therefore, the problem of low accuracy of the existing network quality judgment method in the relevant technology can be solved, and the effect of improving the accuracy of judging network quality can be achieved.

[0055] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0056] In this embodiment, a device for determining poor quality users is also provided, which is used to implement the above embodiments and preferred implementation modes, and will not be repeated hereafter. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0057] Figure 4 is a structural block diagram of a device for determining a poor quality user according to an embodiment of the present invention. Figure 4 As shown, the device includes: an acquisition module 22 and a determination module 24.

[0058] An acquisition module 22, configured to acquire matrix data and feature data based on first session-level data XDR associated with a user tag;

[0059] The determination module 24 is used to input the matrix data and the feature data into a deep learning model to obtain the poor quality probability, and determine the poor quality user according to the poor quality probability.

[0060] Figure 5 is a structural block diagram of a device for determining poor quality users according to an embodiment of the present invention. Figure 5 As shown, the device includes Figure 4 All modules shown, plus:

[0061] The association module 26 is used to associate the user's tag with the first XDR through the user account, wherein the tag is fuzzy processed.

[0062] It should be noted that the above modules can be implemented by software or hardware. For the latter, it can be implemented in the following ways, but not limited to: the above modules are all located in the same processor; or the above modules are located in different processors in any combination.

[0063] To facilitate the understanding of the technical solution provided by the present invention, a detailed description will be given below in conjunction with embodiments of specific scenarios.

[0064] On the network access side, the intelligent big data platform uses unified deep packet inspection (DPI) to perform deep analysis on the control plane and user plane traffic data in the network access pipeline to form XDR call lists. XDR call lists are session-level detailed records of the signaling process and service transmission process generated after processing the full amount of Internet data. This gives traffic data real meaning. DPI has the ability to identify more than 15,000 services, which can provide a more thorough insight into network traffic.

[0065] In the embodiment of the present invention, a deep learning identification method and a poor quality judgment system for home network users are proposed. In this method, the collected session-level data XDR of home network users surfing the Internet (for more than 1 week) is subjected to data preprocessing such as filtering and highly generalized data combination methods, and a feature matrix is ​​formed through automatic feature analysis of a deep model, and the network quality is identified through a deep learning model to find out users with poor network quality, so as to actively operate and complete user rectification and improve the Internet experience of users across the network, which can significantly improve the accuracy of user identification.

[0066] This method is applied to Figure 2 The quality judgment system in OLT, Figure 6 is a structural block diagram of a quality difference judgment system according to an embodiment of the present invention. Figure 6 As shown, the system includes: an original data acquisition device, a data storage device, a data query device and a quality difference judgment device.

[0067] Specifically, the original data collection device is usually deployed in the carrier's bearer equipment to mirror the data passing through the bearer equipment and obtain the user's IP packet header information, and then calculate the delay information including link establishment, transmission, etc. In the embodiment of the present invention, the delay data collected by the OLT is mainly used to judge the poor quality of the user.

[0068] Currently, the data traffic of operators' bearer equipment can reach hundreds of GB per second. Real-time analysis is difficult and unnecessary. Therefore, a timed collection strategy is often used, for example, timed collection at a 5-minute granularity.

[0069] Data storage devices are used to store massive amounts of network indicator information. They use open source distributed file systems, such as FastDFS, Ceph, and HDFS, and integrate the file systems through self-developed platforms to achieve persistence of massive amounts of data.

[0070] The data query device, faced with the query and calculation of massive data, requires the computing power support of multiple cluster servers. The data query device in this embodiment is completed by integrating distributed computing frameworks such as Hadoop3.0 and Spark. The distributed framework uses the computing power of multiple clusters to quickly complete data query and calculation, and at the same time divides the business with high traffic into multiple cycles, and uses a multi-task timing scheduling algorithm to output the results.

[0071] The poor quality judgment device is used to query the original network indicator data, and analyze the indicator data to obtain the poor quality users in the entire network, and give the reasons for the poor quality and rectification suggestions. The final result is presented to the operator in the form of a report, and the user rectification can be completed based on the report to improve the overall network quality and enhance the user's Internet experience.

[0072] The method for identifying poor-quality home broadband users in the embodiment of the present invention includes two stages: a training stage and a verification stage.

[0073] The training phase mainly uses data to learn the model parameters of artificial intelligence; the verification phase is to directly output poor quality users after deployment. The two phases share the same data processing steps and the same deep learning model. Next, we will first explain the data processing steps and deep learning models, and then explain the training phase and the testing phase.

[0074] Figure 7 is a flow chart of data processing steps according to an embodiment of the present invention. Figure 7 As shown, the following steps are included:

[0075] Step S702, performing data processing on the collected XDR according to the protocol;

[0076] Specifically, data processing includes:

[0077] Discard non-TCP protocol data;

[0078] Extract HTTP protocol data from TCP protocol data through protocol packet parsing;

[0079] The TCP protocol data is divided into the following four categories: video, web, game, and other; including: (1) the video data of the HTTP protocol is taken as a large category, recorded as video; (2) the non-video data in the HTTP protocol is taken as a large category, recorded as web; (3) the game data extracted from TCP is taken as a large category, recorded as game; (4) other data that cannot be extracted is taken as a large category, recorded as other.

[0080] Step S704: filtering background traffic.

[0081] Specifically, the background flow is removed in the following three ways:

[0082] (1) Remove data from overseas IP addresses;

[0083] (2) Analyze the service category through the IP address, determine the service category and find the service subcategory of background flow such as heartbeat signal, and remove all data corresponding to the service subcategory;

[0084] (3) Removing nighttime background flow, for example: using the data from 1 a.m. to 5 a.m. as the background flow benchmark, and then removing the background flow of the whole day based on this background flow benchmark.

[0085] Step S706, removing the linear correlation features in the XDR.

[0086] Among them, linear association is defined as the feature that can be obtained through a series of operations such as addition, subtraction, multiplication and division from other features.

[0087] Step S708, the data of different business categories are aggregated to form a three-dimensional matrix, which includes the four business categories of the user, the adoption data of the business categories, and the collection time. This method greatly reduces the size of the data, while the discarded information is limited, and the arrangement of the features is also regular.

[0088] Among them, the data used in major business categories include: the delay of the TCP protocol three-way handshake, the average value of data upload delay, the average value of data download delay, the data upload retransmission rate, and the collection time.

[0089] In this embodiment, data is sorted by a three-dimensional matrix, the number of XDRs is not excessively required, and the input data can be variable in length.

[0090] Step S710: manual feature extraction, extracting common features such as mean, standard deviation, maximum value, etc. for each indicator in the time dimension.

[0091] Figure 8 is a schematic diagram of a deep learning process based on a deep learning model according to an embodiment of the present invention, such as Figure 8 As shown in the figure, the process includes steps such as manual feature extraction, data processing, deep coding, hole convolution, splicing, feature integration, and output analysis results.

[0092] The deep learning model fuses the artificial features (obtained through step S710 in data processing) and the data matrix (obtained through step S708 in data processing) to output poor quality users. This method can control the generalization ability through artificial features and can also automatically process features through deep models. Combining the two methods can enhance the accuracy and generalization of the model.

[0093] Fig. 9is a flowchart of deep learning according to an embodiment of the present invention. Fig. 9 As shown in Figure 1, the deep learning process includes the following steps:

[0094] Step S902: deep feature coding, using a two-layer neural network to perform deep feature coding on artificial features.

[0095] Step S904, convolution coding, performs three module convolutions on the data matrix, each module consists of three layers of dilated convolution (a deep learning processing operation).

[0096] Step S906, concatenate the deep feature encoding result of step S902 and the convolution encoding result of step S904, and then use a 2-layer neural network to perform feature integration.

[0097] Step S908: Convert the result of step S906 into poor quality probability through the softmax layer and output the poor quality probability. The poor quality probability ranges from 0 to 1.

[0098] Regarding the training phase and the validation phase, the training phase can be a training model obtained by running the training steps based on the training data set; or it can be a training model obtained by migrating the model of other sites to the current site. The latter requires less data, and the migrated model is used as the model initialization parameter for subsequent training. The former requires a large amount of data and can be trained directly.

[0099] The training phase mainly includes the following two steps:

[0100] Step 1: Dataset preparation.

[0101] Specifically, the label of the revisited user is associated with the user data (XDR call record) through the user account. For example, when the user is revisited, the user who has no feedback on poor quality is marked as 0.2, the user who feedbacks poor quality but does not want on-site repair is marked as 0.8, and the user who feedbacks poor quality and requires on-site repair is marked as 1.

[0102] Because home users have uncertainty when answering, for example, the person making a phone call is not the main network user, etc., blurring the labels can reduce the uncertainty brought about by the user's answer.

[0103] In one embodiment, the above label can also be selected as 0.3, 0.7, and 0.9. For example, when users visit again, users who do not give feedback on poor quality are marked as 0.3, users who give feedback on poor quality but do not want on-site repair are marked as 0.7, and users who give feedback on poor quality and need on-site repair are marked as 0.9.

[0104] The tag combinations in the embodiment of the present invention are not limited to the above two types. Specific examples of other tag combinations can refer to the above embodiments, which will not be described in detail in this embodiment.

[0105] Step 2: Training.

[0106] Specifically, the cross entropy loss is applied to the output results and labels of the deep learning model, and SGD is used for back propagation to correct the deep model parameters. After multiple iterations, the required model is finally obtained.

[0107] The process of the testing phase mainly includes the following steps:

[0108] Step 1: Process the XDR data. The specific data processing steps are as described in steps S702-S710.

[0109] Step 2: Load the model in the training phase.

[0110] Step three, input the data processed in step one into the model in step two, and finally obtain the probability of poor quality users.

[0111] Step 4: According to the probability of poor quality users, sort them from large to small according to the probability, take the specified number of users, and get a list of poor quality users.

[0112] Through the above steps of the present invention, the method for identifying poor quality users of home networks in the embodiment of the present invention has an accuracy rate 15% higher than the method currently used in the product line, and an accuracy rate 45% higher than the operator's method. At the same time, the deep learning algorithm is introduced into the identification of poor quality users. Through the deep learning model, multiple features can be automatically learned to reduce the trial and error of manual feature extraction; and the deep learning algorithm does not need to be trained separately for each object. The trained model will adapt to all objects, so for the client, it only needs to wait for an inference time; in addition, the deep learning algorithm in this embodiment requires fewer layers and the operation is very fast.

[0113] An embodiment of the present invention further provides a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute the steps of any of the above method embodiments when running.

[0114] In an exemplary embodiment, the computer-readable storage medium may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0115] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0116] In an exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0117] For specific examples in this embodiment, reference may be made to the examples described in the above embodiments and exemplary implementation modes, and this embodiment will not be described in detail herein.

[0118] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0119] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for determining a poor quality user, characterized in that: include: Acquire matrix data and feature data based on first session-level data XDR associated with the user tag; The matrix data and the feature data are input into a deep learning model to obtain poor quality probability, and poor quality users are determined according to the poor quality probability.

2. The method according to claim 1, characterized in that Before acquiring the matrix data and the feature data based on the first session-level data XDR associated with the user tag, the method further includes: The user tag is associated with the first XDR through a user account, wherein the user tag is fuzzy processed.

3. The method according to claim 1, characterized in that Acquiring matrix data and feature data based on a first XDR associated with the user tag includes: Screening out non-TCP protocol XDRs in the first XDR to obtain a second XDR; Classifying the second XDRs, and performing background traffic filtering on the classified second XDRs respectively; Deleting linear correlation features in each type of the second XDR respectively, to obtain a third XDR corresponding to each type of the second XDR; The third XDRs of different types are aggregated to obtain the matrix data; and features are extracted from each type of the third XDR to obtain the feature data.

4. The method according to claim 3, characterized in that The second XDR is classified into: Classify the video data of the HTTP protocol in the second XDR into first category data; Classifying the non-video data of the HTTP protocol in the second XDR into second category data; Classifying the game data in the second XDR into third category data; The data that cannot be extracted from the second XDR is classified as fourth category data.

5. The method according to claim 3, characterized in that: The classified second XDRs are respectively subjected to background traffic filtering, which includes at least one of the following: Filter data from overseas IP addresses; Parsing the service category through the IP address, determining the service subcategory of the background traffic according to the service category, and filtering all data of the service subcategory; The user's nighttime data is used as a background traffic benchmark, and the user's background traffic throughout the day is filtered based on the benchmark.

6. The method according to claim 3, characterized in that: in, The matrix data includes the following information: data type, data index, and acquisition time of the first XDR.

7. The method according to claim 6, characterized in that in, The data indicators include at least one of the following: the delay of the TCP protocol three-way handshake, the average value of the data upload delay, the average value of the data download delay, and the data upload retransmission rate.

8. The method according to claim 1, characterized in that Inputting the matrix data and the feature data into a deep learning model to obtain the probability of poor quality includes: Performing deep feature encoding on the feature data, and performing convolution encoding on the matrix data; Splicing the result of the deep feature encoding with the result of the convolutional encoding, and performing feature integration on the spliced ​​data through a neural network; The result of integrating the features is converted into a quality difference probability.

9. A device for determining a poor quality user, characterized in that: include: An acquisition module, configured to acquire matrix data and feature data based on first session-level data XDR associated with a user tag; A determination module is used to input the matrix data and the feature data into a deep learning model to obtain a poor quality probability, and determine a poor quality user according to the poor quality probability.

10. The device according to claim 9, characterized in that Also includes: An association module is used to associate a user tag with the first XDR through a user account, wherein the user tag is fuzzy processed.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program implements the steps of the method described in any one of claims 1 to 8 when executed by a processor.

12. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method described in any one of claims 1 to 8 are implemented.

Images