Method and device for determining user with poor network quality and computer readable storage medium

Abstract

The embodiment of the invention provides a method for determining a user with poor network quality, electronic equipment and a computer readable storage medium. The method comprises the following steps: acquiring internet data of a target user; a first feature is extracted from the internet surfing data through a model with a Transform structure, and the first feature is used for representing the feature of poor user network quality; and determining the probability that the target user is a user with poor network quality according to the first feature and a preset classification model. In this way, rapid and effective analysis of the user internet surfing data can be completed, so that the user with poor network quality can be efficiently identified.

Description

Technical Field

[0001] This document relates to the field of communication technology, and in particular to a method, device and computer-readable storage medium for determining users with poor network quality. Background Art

[0002] With the rapid development of network technology, the Internet has penetrated into people's lives (for example, home broadband has become a necessary facility for modern families), and surfing the Internet has become an indispensable daily activity for modern people. Accordingly, network quality has become the key to determining the market competitiveness of operators. In order to improve user stickiness, operators need to actively perceive and improve user network quality.

[0003] The most direct way to judge the quality of a user's network is to see how good or bad the user's personal experience is. However, there are many factors that affect the user's Internet experience, such as end-to-end latency, content source quality, and the quality of the home network itself. Moreover, as time goes by, the amount of this data is usually huge. This makes it a recognized problem in the industry to quickly and effectively analyze these factors and efficiently identify users with poor network quality. Summary of the invention

[0004] The embodiments of this specification provide a method, device and computer-readable storage medium for determining users with poor network quality, which are used to solve the problem of how to quickly and effectively analyze user Internet access data to efficiently identify users with poor network quality.

[0005] To solve the above technical problems, the embodiments of this specification are implemented as follows:

[0006] In a first aspect, a method for determining a user with poor network quality is provided, the method comprising:

[0007] Collect the target user’s online data;

[0008] Extracting a first feature from the Internet access data by using a model having a Transformer structure, wherein the first feature is a feature used to characterize poor user network quality;

[0009] The probability that the target user is a user with poor network quality is determined according to the first feature and a preset classification model.

[0010] In a second aspect, a device for determining a user with poor network quality is provided, the device comprising:

[0011] Data collection module, used to collect Internet access data of target users;

[0012] A feature extraction module, used to extract a first feature from the Internet access data by using a model with a Transformer structure, wherein the first feature is a feature used to characterize poor user network quality;

[0013] The first determination module is used to determine the probability that the target user is a user with poor network quality according to the first feature and a preset classification model.

[0014] According to a third aspect, an electronic device is provided, including:

[0015] processor;

[0016] a memory for storing instructions executable by the processor;

[0017] The processor is configured to execute the instructions to implement the method as described in the first aspect.

[0018] In a sixth aspect, a computer-readable storage medium is provided. When instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the method described in the first aspect.

[0019] In the embodiments of the present specification, since a model with a Transformer structure can be used to automatically extract features that can characterize the poor network quality of the user from the Internet access data of the target user, and based on this and a preset classification model, the probability that the target user is a user with poor network quality is obtained, it is possible to quickly and effectively analyze the user's Internet access data, thereby efficiently identifying users with poor network quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of this specification 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 recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0021] Figure 1 It is a schematic diagram of the architecture of a home broadband network provided by an embodiment of this specification.

[0022] Figure 2 It is a schematic diagram of an implementation architecture of a method for determining users with poor network quality provided by an embodiment of this specification.

[0023] Figure 3 It is a flowchart of a method for determining users with poor network quality provided by an embodiment of this specification.

[0024] Figure 4It is a flowchart of a method for determining users with poor network quality provided by another embodiment of the present specification.

[0025] Figure 5A It is a schematic diagram of a list of pre-processed Internet access data - XDR data provided by an embodiment of this specification.

[0026] Figure 5B It is a schematic diagram of data grouping operation provided by an embodiment of this specification.

[0027] Figure 6 It is a schematic diagram of the structure of an electronic device according to an embodiment of the present specification.

[0028] Figure 7 It is a structural diagram of a device for determining users with poor network quality provided by an embodiment of this specification.

[0029] Figure 8 It is a structural diagram of a device for determining users with poor network quality provided by another embodiment of the present specification. DETAILED DESCRIPTION

[0030] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in one or more embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this document.

[0031] The terms "first", "second", etc. in this specification and claims are used to distinguish similar objects, and are not used to describe a particular order or precedence. It should be understood that the terms used in this way are interchangeable where appropriate, so that the embodiments of this specification can be implemented in an order other than those illustrated or described herein. In addition, "and / or" in this specification and claims means at least one of the connected objects, and the character " / " generally means that the objects associated with each other are in an "or" relationship.

[0032] In order to solve the problem of how to quickly and effectively analyze the user's online data to efficiently identify users with poor network quality, this specification proposes a method for determining users with poor network quality. The method can be applied to network-side devices to quickly and effectively analyze the user's online data, thereby efficiently identifying users with poor network quality, so that operators can actively improve user network quality, enhance user online experience, and thus increase user stickiness.

[0033] The method for determining a user with poor network quality provided in this specification can be applied to any network. As an example, it can be applied to a home broadband network. For the sake of brevity, the method for determining a user with poor network quality provided in this specification is mainly described below using a home broadband network.

[0034] Figure 1 FIG. 1 shows a schematic diagram of the architecture of a home broadband network. Figure 1 As shown, the home broadband network mentioned in the embodiment of this specification can be a passive optical network (Passive Optical Network, PON). PON is a network that does not contain any electronic devices and electronic power supply.

[0035] like Figure 1 As shown, a PON generally includes: an optical line terminal (Optical Line Terminal, OLT) 1 installed in a central control station, a group of supporting optical network units (Optical Network Unit, ONU) 3 installed in user locations, an optical distribution network (Optical Distribution Network, ODN) 2 located between OLT 1 ​​and ONU 3, and a terminal 4. Among them, ODN 2 includes optical fiber, passive splitter (including primary splitter and secondary splitter) / coupler. As an example, a user with poor network quality proposed in this specification can be executed by OLT 1.

[0036] exist Figure 1 middle:

[0037] OLT 1 ​​is an optical line terminal. OLT 1 ​​usually includes a control board, a DC power board, a fan unit, a service frame, and other upstream boards. One side of OLT 1 ​​is connected to the switch at the coverage level through a cable and converted into an optical signal; the other side is connected to the optical splitter on the customer terminal through an optical fiber. OLT 1 ​​can be used to control, manage, and measure the distance to customer terminal devices such as ONU 3. OLT 1 ​​is one of the devices on the service node of the access network.

[0038] ODU 2 is an optical distribution network. ODN 2 is a fiber-to-the-home (FTTH) optical cable network based on PON equipment. Its function is to provide an optical transmission channel between OLT 1 ​​and ONU 3. From the functional point of view, ODN 2 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.

[0039] ONU3 is called optical network unit. ONU3 is divided into active optical network unit and passive optical network unit. Generally, the equipment equipped with optical receiver, uplink optical transmitter, multiple bridge amplifier network monitoring is called optical node. ONU3 provides data, interactive network television (Internet Protocol Television, IPTV), voice (using integrated access device (Integrated Access Device, IAD)) and other services. Common home ONU is generally called optical modem.

[0040] The terminal 4 refers to an Internet access device at the user end, such as a set-top box (STB) 41, an Internet TV 42, a computer 43, and a mobile phone 44.

[0041] In practical applications, the method for determining users with poor network quality provided in the embodiments of this specification can be implemented by a device for determining users with poor network quality set in an OLT, such as Figure 2 As shown, the device may include: a data acquisition device 21, a data storage device 22, a data query device 23 and a quality difference feature analysis device 24.

[0042] The data collection device 21 is used to collect the user's Internet access data. Specifically, a deep packet inspection (DPI) probe can be set in the OLT to monitor the Internet application (Application, APP) used by the user and perceive the user's real Internet access data (original Internet access data). Then the original data is processed and analyzed, including initialization decoding, packet header decompression, service layer decoding, etc., and the data is extracted from the result of packet decoding, and the required fields are extracted. According to the method of session association and synthesis, the user's Internet access session-level data - user detail record (Xdetail record, XDR) data is generated. More specifically, the DPI probe divides the end-to-end home broadband network into two: the upper network above the OLT and the lower network below the OLT, and calculates the upstream network round trip time (Round Trip Time, RTT), the downstream network RTT and network indicator information such as service indicators through two-way service traffic (tcp link establishment three-way handshake process and data transmission process, etc.), thereby laying a data foundation for perceiving user service experience and identifying poor quality (abbreviation for poor network quality) users.

[0043] The data storage device 22 is used to store the massive amount of user online data (including many network indicator information) collected and processed by the data collection device 21 - XDR data. At present, due to the huge user network traffic, it may reach hundreds of Gb per second. Therefore, a strategy of periodic collection at intervals is generally adopted to collect and persist data for several weeks for subsequent analysis.

[0044] The data query device 23 is used to quickly query and calculate the target data required later from the massive data stored in the data storage device 22. It can be understood that in the face of the query and calculation of massive data, the computing power support of multiple cluster servers is required, which can be achieved by integrating distributed computing frameworks such as Hadoop3.0 and Spark through the self-developed data query device 23. The distributed framework uses the computing power of multiple clusters to quickly complete the query and calculation of data, 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.

[0045] The poor quality feature analysis device 24 is the core of a method for determining poor quality network users provided in this specification. The poor quality feature analysis device 24 is used to query the XDR data of the target user from the data storage device 22 through the data query device 23, and automatically extract the poor quality feature (i.e., the first feature hereinafter) from it using a model with a Transformer structure, and then derive the probability of the target user having poor quality based on the poor quality feature, and provide potential poor quality users based on the probability.

[0046] Optionally, the quality difference feature analysis device 24 can also present the final result in the form of a report to the operator, and the operator can complete the user network rectification according to the report, thereby improving the user's Internet experience.

[0047] Optionally, the quality difference feature analysis device 24 also provides the ability to update and iterate the model, that is, iteratively update the parameters of the model with the Transformer structure according to the second feature below (including user preferences, user tendencies, user evaluations, etc. in the area where the target user is located), so that the quality difference features extracted by using the model are more accurate, so that the final quality difference user analysis results are more consistent with the actual situation in the area, thereby increasing the reliability of the quality difference user analysis results.

[0048] It should be noted that the processing of XDR data is a recognized problem in the industry. On the one hand, these XDR data are simplified relative to the actual online situation of users. Since they are only a data representation of the user's online situation, they are bound to be limited by the objectively huge amount of online data and the limited collection and storage capabilities of the equipment, so this data representation is usually simplified. How to extract effective and key features from these simplified data is a must-go test for all intelligent algorithms. On the other hand, these simplified XDR data are still extremely complex. In order to restore the user's network quality as much as possible, it is necessary to introduce dozens or hundreds of basic indicator dimensions and their derivative dimensions including latency and retransmission rate into each XDR data. In addition, the data that affects the user's online perception may appear at any possible time in the past period of time, which means that the data to be analyzed has a considerable time span. If only the data of the past week is used as a benchmark for analysis, the number of XDRs will usually exceed 30,000, and most of the data may be almost invalid. The ultimate goal is to analyze and determine the user's overall online experience from these long-term data. In actual operation, the reliability of the algorithm will be verified by calling candidate poor-quality users.

[0049] This specification proposes a method for determining users with poor network quality, which cleverly introduces a model with a Transformer structure, making it possible to intelligently infer and extract user quality characteristics (i.e., the first feature below) from XDR data. By collecting past user data and outbound call feedback (i.e., the second feature below), a reliable algorithm for intelligently extracting data features can be established, thereby training a model with the ability to identify poor quality characteristics. If this solution provided in this specification is applied to the target network device, it can play a key role in automatically and efficiently identifying poor quality users.

[0050] A method for determining users with poor network quality provided by an embodiment of this specification is described in detail below with reference to the accompanying drawings.

[0051] In one embodiment, the present specification provides a method for determining users with poor network quality, which can be applied to network devices such as OLT, such as Figure 3 As shown, the method may include:

[0052] Step 301: Collect Internet access data of target users.

[0053] The network used by the target user may be, but is not limited to, a home broadband network.

[0054] As mentioned above, the Internet applications used by users can be monitored through pre-set DPI probes to perceive the user's real Internet data (original Internet data). Then the original data is processed and analyzed, including initialization decoding, packet header decompression, business layer decoding, etc., and the data is extracted from the packet decoding results, and the required fields are extracted. According to the session association and synthesis method, the user's Internet session-level data - user detail record (XDR) data is generated.

[0055] Step 302: extract a first feature from the Internet access data by using a model with a Transformer structure, wherein the first feature is a feature used to characterize poor user network quality.

[0056] Transformer is a deep learning model based on the self-attention mechanism, which is often used in natural language processing tasks such as machine translation. Compared with traditional recurrent neural networks (RNN) and convolutional neural networks (CNN), Transformer uses a global self-attention mechanism, which can process input sequences in parallel and is not limited by the length of the input sequence. This gives it better parallel computing performance and can better capture long-distance dependencies in the sequence.

[0057] Transformer consists of two parts: Encoder and Decoder. The encoder is responsible for encoding the input sequence into a series of hidden state representations, while the decoder generates the output sequence based on these hidden states.

[0058] In the encoder, Transformer uses a multi-layer self-attention mechanism so that each word can pay attention to all other words in the input sequence and weight them according to their relevance. This self-attention mechanism can help the model understand the contextual information of the input sequence and provide better feature representation.

[0059] This specification takes into account the characteristics of user Internet data such as long time span and long-term correlation between data, which are similar to the data that can be processed by Transformer. Therefore, a model with a Transformer structure is introduced, and the encoder therein is used to encode the Internet data to obtain the first feature.

[0060] As mentioned above, when users use the Internet, they often have a bad experience when using a certain APP. When users open a certain APP, they will naturally think of the previous bad experience. Such experience accumulation may eventually produce a qualitative change. In other words, users have a certain attention, which is generated through long-term associations. This is also the scenario that the attention mechanism tries to describe. The encoding-decoding structure can be understood as a process of information compression (encoding) and then decompression (decoding).

[0061] This specification only uses the encoder in the model with a Transformer structure, because the purpose of this specification is to extract the key features related to quality difference in the XDR data (i.e., the first feature) as efficiently and accurately as possible, which is equivalent to performing necessary (lossy) compression on the Internet data, retaining the valid information and eliminating redundant information.

[0062] Therefore, step 302 may specifically include: encoding the online data to obtain the first feature through an encoder in a model with a Transformer structure. In other words, the encoding result of the online data is the first feature.

[0063] Optionally, before executing the above step 302, the Internet access data of the target user collected in step 301 may be preprocessed. This part is a pre-step for the model with the Transformer structure to perform intelligent reasoning to obtain the first feature, and is also an important part of optimizing data quality. In an actual production environment, the collected Internet access data usually has information redundancy or is affected by outliers. Therefore, it is necessary to effectively preprocess or standardize the collected Internet access data.

[0064] Specifically, the pretreatment may include but is not limited to at least one of the following:

[0065] 1) Eliminate Internet access data corresponding to unnecessary background traffic

[0066] This part of data may include: traffic with a number of packets received and sent less than a certain threshold, heartbeat data generated by each networked device when no one is using broadband at night, and data from certain apps or small business categories that clearly only generate background traffic. It can be expected that after such data screening, most of the remaining data can reflect the user's actual perception of surfing the Internet. The user's real perception may exist in many forms, such as alternating over a long period of time, overlapping and coupling between different apps, and occasional occurrence of poor quality.

[0067] 2) Eliminate abnormal peaks

[0068] As mentioned above, in actual production environments, the Internet access data collected by devices usually contain abnormal values. Such abnormal values ​​are often manifested as unreasonable peaks of certain indicators. It is understandable that since the indicators that affect the Internet access experience are usually related to latency and retransmission, the impact of these indicators on perception is not linear. According to laboratory measurement results, latency exceeding 5000ms is usually unbearable. Therefore, if an abnormally large or far exceeding 5000ms latency value appears in the Internet access data, the impact on the user experience has actually exceeded the temporary tolerance limit. If the actual algorithm overemphasizes such abnormal values ​​at a certain point in time, it will significantly increase the possibility of users being judged as poor quality, but this is contrary to the evaluation goal of long-term experience. Therefore, it is necessary to perform peak shaving operations on important indicators, for example: the value exceeding 5000ms in the latency indicator is preset to 5000ms. In this way, the final feature extraction algorithm - the model with a Transformer structure, can achieve a better balance in the judgment of long-term experience and short-term experience.

[0069] 3) Determine statistical characteristics

[0070] The so-called statistical features are the data features obtained by applying traditional statistical functions such as mean, peak, standard deviation, median, etc. From the laboratory data, these statistical features can indeed reflect the network quality or user experience from a certain perspective. Using these statistical features, an accuracy rate of over 90% can be obtained on laboratory data. Although the laboratory scenario will only collect experience in a relatively short period of time, at least it can be confirmed that these statistical features are reliable in short-term experience evaluation. Therefore, in the online algorithm, long-term data will be grouped into small granularity (every hour), and after extracting statistical features respectively, the model with Transformer structure comprehensively infers these statistical features reflecting short-term experience and Internet data (Internet data preprocessed by items 1) and 2) above) to extract the first feature. In other words, these statistical features can be used to provide data enhancement effects, so that the neural network algorithm can grasp the data features more easily and accurately.

[0071] After preprocessing, correspondingly, step 302 may specifically include: encoding the preprocessed Internet access data through an encoder in a model with a Transformer structure to obtain a first feature.

[0072] In step 302, the model with the Transformer structure is a model that has been trained and converged. During training, a supervised training method can be used to train the model with the Transformer structure according to a preset ratio of negative samples and positive samples, wherein the negative samples are the Internet access data of known users who are not of poor network quality, and the positive samples are the Internet access data of known users of poor network quality.

[0073] The model with a Transformer structure may include but is not limited to a neural network with a Transformer structure.

[0074] Step 303: Determine the probability that the target user is a user with poor network quality based on the first feature and a preset classification model.

[0075] Among them, the preset classification model can be a binary classification model, such as softmax function, logistic regression, support vector machine (SVM), random forest, naive Bayes or neural network. Optionally, a binary classification model that can consider the correlation between features (such as a neural network) is selected as the preset classification model, because the user's Internet experience may be correlated at different times (such as when the user uses the Internet, he often has a poor experience whenever he uses a certain APP, so when the user opens a certain APP, he will naturally think of the previous bad experience, and such a sense of experience is constantly superimposed and may eventually produce a qualitative change effect).

[0076] Optionally, it may be further determined whether the probability exceeds a preset threshold; when the probability exceeds the preset threshold (such as 50%), the target user is determined as a user with poor network quality.

[0077] A method for determining users with poor network quality proposed in an embodiment of this specification can use a model with a Transformer structure to automatically extract features that can characterize the poor network quality of the user from the Internet access data of the target user, and based on this and a preset classification model, obtain the probability that the target user is a user with poor network quality. Therefore, it is possible to quickly and effectively analyze the user's Internet access data, thereby efficiently identifying users with poor network quality.

[0078] In another embodiment, if Figure 4 As shown, a method for determining a user with poor network quality provided in this specification can be applied to an OLT. The method may include:

[0079] Step 301: Collect Internet access data of target users.

[0080] The description of step 301 is the same as above and will not be repeated here.

[0081] Step 304 , determine whether the current model with the Transformer structure is applicable; if yes, execute step 306 , otherwise, execute step 305 .

[0082] This step aims to determine whether the model with the Transformer structure is consistent with the common characteristics of users in the target user's area. If not, the current model with the Transformer structure needs to be adjusted according to the common characteristics of users in the target user's area.

[0083] Step 305: Adjust the current model with the Transformer structure.

[0084] Specifically, the second feature of the target area users can be obtained first, wherein the target area is the area where the target users are located, and the second feature is a common feature of the target area users using the target network; the second feature is used as the optimization target, and the parameters of the model with the Transformer structure are updated.

[0085] The second feature may include but is not limited to one or more of the following:

[0086] User preferences;

[0087] User preferences;

[0088] User evaluation, also known as outbound feedback.

[0089] Step 306: pre-process the Internet access data.

[0090] This part is the pre-step for the model with Transformer structure to perform intelligent reasoning to obtain the first feature, and it is also an important part of optimizing data quality. In actual production environments, the collected online data usually has information redundancy or is affected by outliers. Therefore, it is necessary to effectively preprocess or standardize the collected online data.

[0091] As described above, the pretreatment may include but is not limited to at least one of the following (see above for details):

[0092] 1) Eliminate Internet access data corresponding to unnecessary background traffic;

[0093] 2) Eliminate abnormal peaks;

[0094] 3) Determine statistical characteristics.

[0095] Step 307: group the pre-processed Internet access data to obtain a plurality of data groups.

[0096] For the above preprocessed or standardized Internet data, in order to better use the model with the Transformer structure to extract the quality difference feature (ie, the first feature), the embodiment of this specification further performs grouping processing.

[0097] like Figure 5A As shown, the preprocessed Internet access data - XDR data is a two-dimensional list. The row indicators in the list mark the order of the XDR data in terms of generation time, and the column indicators in the list are various network indicator dimensions.

[0098] like Figure 5B As shown, the grouping described in the embodiments of this specification refers to dividing the preprocessed two-dimensional list into a number of continuous sub-lists in form, and then arranging these sub-lists in sequence on the timeline. The columns contained in these sub-lists are usually consistent, but the rows are inconsistent. This grouping is to achieve the idea of ​​segmented short-term experience. For the XDR data within each group, it can be considered that they have a short-term impact on the user's experience, and all the groups combined together form a long-term user experience.

[0099] Specifically, the pre-processed Internet access data may be grouped in chronological order according to a preset grouping rule. The preset grouping rule may include but is not limited to the following:

[0100] 1) Grouping with equal data quantity, that is, grouping with fixed XDR quantity, that is, the number of rows in each sublist is exactly the same, and zeros are added if the number is less than that;

[0101] 2) Grouping with equal time length, that is, grouping with fixed time length, that is, the number of rows in each sub-list may not be the same, but these XDR data are in the same time period and are placed in one group. The length of this time period can be set flexibly, usually set to 1 hour.

[0102] Next, the model with the Transformer structure will automatically extract short-term features within each group, and then combine these short-term features to form an evaluation result of the long-term experience.

[0103] Step 302: Encode the plurality of data groups by using an encoder in a model having a Transformer structure to obtain a first feature.

[0104] As mentioned earlier, Transformer is a deep learning model based on the self-attention mechanism, which is often used in natural language processing tasks, especially machine translation. Compared with traditional recurrent neural networks (RNN) and convolutional neural networks (CNN), Transformer uses a global self-attention mechanism, which can process input sequences in parallel and is not limited by the length of the input sequence. This gives it better parallel computing performance and can better capture long-distance dependencies in the sequence. In addition, the Transformer encoder uses a multi-layer self-attention mechanism, so that each word can pay attention to all other words in the input sequence and weight them according to their relevance. This self-attention mechanism can help the model understand the contextual information of the input sequence and provide better feature representation.

[0105] This specification takes into account the characteristics of user Internet data such as long time span and long-term correlation between data, which are similar to the data that can be processed by Transformer. Therefore, a model with a Transformer structure is introduced, and the encoder therein is used to encode the Internet data to obtain the first feature.

[0106] As mentioned above, when users use the Internet, they often have a poor experience when using a certain APP. When users open an APP, they will naturally think of the previous bad experience. Such experience accumulation may eventually produce a qualitative change. In other words, users have a certain attention, which is generated through long-term associations. This is the scenario that the attention mechanism tries to describe. The encoding-decoding structure can be understood as a process of information compression (encoding) and then decompression (decoding).

[0107] This specification only uses the encoder in the model with a Transformer structure, because the purpose of this specification is to extract the key features related to quality difference in the XDR data (i.e., the first feature) as efficiently and accurately as possible, which is equivalent to performing necessary (lossy) compression on the Internet data, retaining the valid information and eliminating redundant information.

[0108] Therefore, step 302 may specifically include: encoding the several data groups separately through an encoder in a model with a Transformer structure to obtain several short-term features for characterizing short-term network quality; and fusing the several short-term features to obtain a first feature for characterizing long-term network quality.

[0109] In step 302, the model with the Transformer structure is a model that has been trained and converged. During training, a supervised training method can be used to train the model with the Transformer structure according to a preset ratio of negative samples and positive samples, wherein the negative samples are the Internet access data of known users who are not of poor network quality, and the positive samples are the Internet access data of known users of poor network quality.

[0110] The model with a Transformer structure may include but is not limited to a neural network with a Transformer structure.

[0111] The "model with a Transformer structure" used in the method for determining users with poor network quality provided in the embodiments of this specification has been fully trained and tested offline, and is generally capable of dealing with the identification of users with poor quality in different offices or regions. If there are situations where the outbound call results of some offices are significantly deviated from the predicted results due to outbound call speech, network quality, user habits, etc., then the model with a Transformer structure can be fine-tuned based on these features (i.e., the second feature mentioned above) to apply to special situations.

[0112] Step 303: Determine the probability that the target user is a user with poor network quality based on the first feature and a preset classification model.

[0113] Among them, the preset classification model can be a binary classification model, such as softmax function, logistic regression, support vector machine (SVM), random forest, naive Bayes or neural network. Optionally, a binary classification model that can consider the correlation between features (such as a neural network) is selected as the preset classification model, because the user's Internet experience may be correlated at different times (such as when the user uses the Internet, he often has a poor experience whenever he uses a certain APP, so when the user opens a certain APP, he will naturally think of the previous bad experience, and such a sense of experience is constantly superimposed and may eventually produce a qualitative change effect).

[0114] Step 308, determining whether the probability exceeds a preset threshold; if yes, executing step 309, otherwise, executing step 310, or returning to executing step 301 to re-evaluate the network quality of the target user.

[0115] Step 309: determine the target user as a user with poor network quality.

[0116] It should be noted that 1) the data preprocessing or standardization methods provided in the embodiments of this specification are universal, but there may be different combination optimization methods on different types of data sets. Other effective data preprocessing and standardization methods should be regarded as alternatives to the embodiments of this specification; 2) Although the neural network with a Transformer structure is used in the examples of the embodiments of this specification, it can also be replaced by other simplified networks or optimized versions of networks with a Transformer structure; 3) In addition to the two grouping methods used in the embodiments of this specification, there are still other possible grouping methods, which should be regarded as the coverage of this application.

[0117] A method for determining users with poor network quality proposed in an embodiment of this specification can use a model with a Transformer structure to automatically extract features that can characterize the poor network quality of the user from the Internet access data of the target user, and based on this and a preset classification model, obtain the probability that the target user is a user with poor network quality. Therefore, it is possible to quickly and effectively analyze the user's Internet access data, thereby efficiently identifying users with poor network quality.

[0118] In addition, the method for determining users with poor network quality proposed in the embodiments of this specification can also achieve at least one of the following beneficial effects by automatically extracting features that can characterize poor user network quality from target user's Internet access data using a model with a Transformer structure:

[0119] 1) High prediction accuracy: This specification proposes a method for determining users with poor network quality, which can achieve an accuracy of 70% at a new start point.

[0120] 2) Strong versatility: This specification proposes a method for determining users with poor network quality, which introduces a deep learning algorithm - a model with a Transformer structure into the identification of poor quality users. The deep learning model can automatically learn a variety of features, and also uses verified effective statistical features as data enhancement of the model input data, making the model more effective.

[0121] 3) Inference of ultra-long time series with poor quality features: Thanks to the appropriate XDR data grouping operation, the equivalent time series length is the number of groups, which is much smaller than the time series length of the original data. This greatly improves the efficiency of time series processing algorithms.

[0122] 4) Easy deployment: The code is relatively independent and can be easily deployed on the server.

[0123] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0124] Figure 6 This is a schematic diagram of the structure of an electronic device according to an embodiment of this specification. Figure 6 At the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and a memory. The memory may include a memory, such as a high-speed random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage. Of course, the electronic device may also include hardware required for other services.

[0125] The processor, network interface and memory can be interconnected through an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 6 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0126] The memory is used to store the program. Specifically, the program may include a program code, and the program code includes a computer operation instruction. The memory may include a memory and a non-volatile memory, and provides instructions and data to the processor.

[0127] The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, forming a determination device for users with poor network quality at the logical level. The processor executes the program stored in the memory and is specifically used to perform the following operations:

[0128] Collect the target user’s online data;

[0129] Extracting a first feature from the Internet access data by using a model having a Transformer structure, wherein the first feature is a feature used to characterize poor user network quality;

[0130] The probability that the target user is a user with poor network quality is determined according to the first feature and a preset classification model.

[0131] The above is as in this manual Figure 6 The method performed by the device for determining users with poor network quality disclosed in the illustrated embodiment can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of this specification can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of this specification can be directly embodied as being executed by a hardware decoding processor, or can be executed by a combination of hardware and software modules in a decoding processor. The software module can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0132] The electronic device may also perform Figure 3 or Figure 4 The method and device for determining the user with poor network quality are shown in Figure 3 or Figure 4 The functions in the illustrated embodiments will not be described in detail in this specification.

[0133] Of course, in addition to software implementation, the electronic device of this specification does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0134] The embodiment of the present specification also provides a computer-readable storage medium, which stores one or more programs, wherein the one or more programs include instructions, which, when executed by a portable electronic device including a plurality of application programs, enable the portable electronic device to execute Figure 3 or Figure 4 The method of the embodiment shown is specifically used to perform the following operations:

[0135] Collect the target user’s online data;

[0136] Extracting a first feature from the Internet access data by using a model having a Transformer structure, wherein the first feature is a feature used to characterize poor user network quality;

[0137] The probability that the target user is a user with poor network quality is determined according to the first feature and a preset classification model.

[0138] Figure 7 This is a schematic diagram of the structure of a device 700 for determining users with poor network quality according to an embodiment of this specification. Figure 7 In a software implementation, the device 700 for determining users with poor network quality may include: a data collection module 701 , a feature extraction module 702 and a first determination module 703 .

[0139] The data collection module 701 is used to collect the Internet access data of the target user.

[0140] The feature extraction module 702 is used to extract a first feature from the Internet access data by using a model with a Transformer structure, wherein the first feature is a feature used to characterize poor user network quality.

[0141] The model with a Transformer structure may include but is not limited to a neural network with a Transformer structure.

[0142] The first determination module 703 is used to determine the probability that the target user is a user with poor network quality according to the first feature and a preset classification model.

[0143] Among them, the preset classification model can be a binary classification model, such as a softmax function, logistic regression (LogisticRegression), support vector machine (SVM), random forest (Random Forest), naive Bayes (Naive Bayes) or neural network (Neural Network) and other models.

[0144] The apparatus 700 for determining users with poor network quality provided in the embodiment of this specification can also execute Figure 3 method, and implement Figure 3 The functions of the embodiments shown in the figure are the same and the same technical effects are achieved, so the embodiments of this specification will not be described in detail here.

[0145] Figure 8 FIG. 7 is a schematic diagram of a structure of a device 700 for determining users with poor network quality according to another embodiment of the present specification. Figure 8 In a software implementation, the device 700 for determining users with poor network quality may include: a data acquisition module 701, a first judgment module 704, a model adjustment module 705, a data preprocessing module 706, a data grouping module 707, a feature extraction module 702, a first determination module 703, a second judgment module 708 and a second determination module 709.

[0146] The data collection module 701 is used to collect the Internet access data of the target user.

[0147] The first judgment module 704 is used to judge whether the current model with the Transformer structure is applicable; if so, the data preprocessing module 706 is triggered, otherwise, the model adjustment module 705 is triggered.

[0148] The model adjustment module 705 is used to adjust the current model with the Transformer structure.

[0149] Specifically, the second feature of the target area users can be obtained first, wherein the target area is the area where the target users are located, and the second feature is a common feature of the target area users using the target network; the second feature is used as the optimization target, and the parameters of the model with the Transformer structure are updated.

[0150] The second feature may include but is not limited to one or more of the following:

[0151] User preferences;

[0152] User preferences;

[0153] User evaluation, also known as outbound feedback.

[0154] The data preprocessing module 706 is used to preprocess the Internet access data.

[0155] This part is the pre-step for the model with Transformer structure to perform intelligent reasoning to obtain the first feature, and it is also an important part of optimizing data quality. In actual production environments, the collected online data usually has information redundancy or is affected by outliers. Therefore, it is necessary to effectively preprocess or standardize the collected online data.

[0156] As described above, the pretreatment may include but is not limited to at least one of the following (see above for details):

[0157] 1) Eliminate Internet access data corresponding to unnecessary background traffic;

[0158] 2) Eliminate abnormal peaks;

[0159] 3) Determine statistical characteristics.

[0160] The data grouping module 707 is used to group the pre-processed Internet access data to obtain a plurality of data groups.

[0161] For the above preprocessed or standardized Internet data, in order to better use the model with the Transformer structure to extract the quality difference feature (ie, the first feature), the embodiment of this specification further performs grouping processing.

[0162] like Figure 5A The preprocessed Internet access data - XDR data is a two-dimensional list. The row indicators in the list mark the order of the XDR data in terms of generation time, and the column indicators in the list are various network indicator dimensions.

[0163] like Figure 5B As shown, the grouping described in the embodiments of this specification refers to dividing the preprocessed two-dimensional list into a number of continuous sub-lists in form, and then arranging these sub-lists in sequence on the timeline. The columns contained in these sub-lists are usually consistent, but the rows are inconsistent. This grouping is to achieve the idea of ​​segmented short-term experience. For the XDR data within each group, it can be considered that they have a short-term impact on the user's experience, and all the groups combined together form a long-term user experience.

[0164] Specifically, the pre-processed Internet access data may be grouped in chronological order according to a preset grouping rule. The preset grouping rule may include but is not limited to the following:

[0165] 1) Grouping with equal data quantity, that is, grouping with fixed XDR quantity, that is, the number of rows in each sublist is exactly the same, and zeros are added if the number is less than that;

[0166] 2) Grouping with equal time length, that is, grouping with fixed time length, that is, the number of rows in each sub-list may not be the same, but these XDR data are in the same time period and are placed in one group. The length of this time period can be set flexibly, usually set to 1 hour.

[0167] The feature extraction module 702 is used to encode the plurality of data groups to obtain a first feature through an encoder in a model with a Transformer structure.

[0168] The model with a Transformer structure may include but is not limited to a neural network with a Transformer structure.

[0169] Specifically, the feature extraction module 702 can encode the several data groups separately through the encoder in the model with the Transformer structure to obtain several short-term features for characterizing the short-term network quality; and fuse the several short-term features to obtain the first feature for characterizing the long-term network quality.

[0170] The first determination module 703 is used to determine the probability that the target user is a user with poor network quality according to the first feature and a preset classification model.

[0171] Among them, the preset classification model can be a binary classification model, such as a softmax function, logistic regression (LogisticRegression), support vector machine (SVM), random forest (Random Forest), naive Bayes (Naive Bayes) or neural network (Neural Network) and other models.

[0172] The second judgment module 708 is used to determine whether the probability exceeds a preset threshold; if yes, the second determination module is triggered, otherwise, the data collection module 701 is triggered to re-evaluate the network quality of the target user.

[0173] The second determining module 709 is used to determine the target user as a user with poor network quality.

[0174] The apparatus 700 for determining users with poor network quality provided in the embodiment of this specification can also execute Figure 4 method, and implement Figure 4 The functions of the embodiments shown in the figure are the same and the same technical effects are achieved, so the embodiments of this specification will not be described in detail here.

[0175] In short, the above description is only a preferred embodiment of this specification and is not intended to limit the protection scope of this specification. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this specification shall be included in the protection scope of this specification.

[0176] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0177] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0178] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0179] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

Claims

1. A method for determining a user with poor network quality, comprising: Collect the target user’s online data; Extracting a first feature from the Internet access data by using a model having a Transformer structure, wherein the first feature is a feature used to characterize poor user network quality; The probability that the target user is a user with poor network quality is determined according to the first feature and a preset classification model.

2. According to the method of claim 1, extracting the first feature from the Internet access data by using a model having a Transformer structure comprises: The Internet access data is encoded by an encoder in a model with a Transformer structure to obtain a first feature.

3. According to the method of claim 2, before encoding the Internet access data to obtain the first feature through the encoder in the model with the Transformer structure, the method further comprises: Preprocess the Internet access data.

4. The method according to claim 3, wherein: The preprocessing of the online data includes at least one of the following: Eliminate Internet access data corresponding to unnecessary background traffic; Eliminate abnormal peaks; Determine statistical characteristics.

5. The method according to claim 3, wherein: The encoding of the online data to obtain the first feature by an encoder in a model having a Transformer structure includes: The preprocessed Internet access data is encoded by an encoder in a model with a Transformer structure to obtain a first feature.

6. According to the method of claim 3, before encoding the Internet access data to obtain the first feature through the encoder in the model with the Transformer structure, the method further comprises: The pre-processed Internet access data is grouped to obtain a plurality of data groups.

7. The method according to claim 6, wherein: The grouping of the pre-processed online data includes: The pre-processed Internet access data is grouped in chronological order according to preset grouping rules.

8. The method according to claim 7, wherein: The preset grouping rule includes one of the following: Group by equal amount of data; Group by time length.

9. The method according to any one of claims 6 to 8, wherein: The encoding of the online data to obtain the first feature by an encoder in a model having a Transformer structure includes: The plurality of data groups are encoded by an encoder in a model having a Transformer structure to obtain a first feature.

10. The method according to claim 9, wherein: The encoding of the plurality of data groups to obtain the first feature by an encoder in a model having a Transformer structure includes: By using an encoder in a model with a Transformer structure, the plurality of data packets are respectively encoded to obtain a plurality of short-term features for characterizing short-term network quality; The plurality of short-term features are fused to obtain a first feature for characterizing long-term network quality.

11. The method according to any one of claims 1 to 8 and 10, further comprising: determining whether the probability exceeds a preset threshold; When the probability exceeds a preset threshold, the target user is determined as a user with poor network quality.

12. The method according to any one of claims 1 to 8 and 10, further comprising: Acquire a second feature of users in a target area, wherein the target area is an area where the target user is located, and the second feature is a common feature of users in the target area using the target network; Taking the second feature as an optimization target, the parameters of the model with the Transformer structure are updated.

13. The method according to claim 12, wherein: The second feature includes one or more of the following: User preferences; User preferences; User reviews.

14. The method according to any one of claims 1 to 8, 10 and 13, wherein: The model with the Transformer structure includes a neural network with the Transformer structure.

15. The method according to any one of claims 1 to 8, 10 and 13, wherein: The network used by the target user includes a home broadband network.

16. The method according to claim 15, The online data includes user online detail record XDR data.

17. An electronic device comprising: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the method according to any one of claims 1 to 16.

18. A computer-readable storage medium, when instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the method according to any one of claims 1 to 16.

Images