Root cause analysis method and device for poor perception quality, storage medium and electronic equipment

Abstract

The invention provides a poor-perception-quality root cause analysis method, a poor-perception-quality root cause analysis device, a computer storage medium and electronic equipment, and relates to the technical field of computers. The method comprises the following steps: acquiring multi-modal data of a target object; a hypergraph structure for the target object is constructed according to the multi-modal data, the hypergraph structure comprises a plurality of initial nodes and a plurality of hyperedges used for connecting the initial nodes, and the initial nodes are used for representing a plurality of feature vectors obtained after feature extraction is carried out on the multi-modal data; the plurality of hyperedges are used for representing association relationships among the plurality of feature vectors respectively; and inputting the hypergraph structure of the target object into a pre-trained hypergraph neural network model to obtain a root cause analysis result of poor perceptual quality for the target object. The application range of root cause analysis on various types of environments can be widened, and the accuracy of root cause analysis results can be improved.

Description

Background Art

[0002] With the vigorous development of mobile communication networks, mobile communications have penetrated into every aspect of users' daily lives, providing users with a wide range of network services. Taking the application scenario of high-speed railway networks as an example, passengers have increasingly higher demands for the quality of high-speed railway communications, especially for the stability and clarity of audio and video services. Therefore, it is necessary to monitor the performance of wireless networks along the high-speed railway in real time, and accurately identify and locate the root causes of poor quality problems affecting audio and video network perception.

[0003] In related technical solutions, manual analysis or one-by-one matching methods with preset detection algorithms are usually used to locate the causes of poor audio and video quality in high-speed rail. However, the above methods are not only time-consuming, costly, and inefficient, but also difficult to cope with the massive test data and rapidly changing network requirements in complex application scenarios such as high-speed rail environments, resulting in low accuracy of root cause analysis.

[0004] Therefore, there is an urgent need for a method that can efficiently and accurately implement real-time root cause analysis of perceived quality issues. Summary of the invention

[0005] The present disclosure provides a root cause analysis method for poor perception quality, a root cause analysis device for poor perception quality, a computer storage medium and an electronic device, thereby improving the scope of application of root cause analysis for various types of environments and improving the accuracy of root cause analysis results.

[0006] In a first aspect, an embodiment of the present disclosure provides a root cause analysis method for perceptual poor quality, the method comprising: acquiring multimodal data of a target object; constructing a hypergraph structure for the target object based on the multimodal data, the hypergraph structure comprising multiple initial nodes and multiple hyperedges for connecting the initial nodes, the multiple initial nodes being used to represent multiple feature vectors obtained after feature extraction of the multimodal data, and the multiple hyperedges being used to respectively represent the correlation between the multiple feature vectors; inputting the hypergraph structure of the target object into a pre-trained hypergraph neural network model to obtain a root cause analysis result of the perceptual poor quality of the target object; wherein the hypergraph neural network model comprises multiple hypergraph convolutional network layers and a classifier model, the multiple hypergraph convolutional network layers being used to perform hypergraph convolution processing on the hypergraph structure of the target object to obtain a node feature matrix of the target node associated with the perceptual poor quality; the classifier model being used to determine the root cause analysis result of the perceptual poor quality based on the node feature matrix of the target node.

[0007] In an optional embodiment of the present disclosure, the multimodal data includes two or more of network operation data for the target object, external environment data associated with the target object, and associated object data corresponding to the target object.

[0008] In an optional embodiment of the present disclosure, the method also includes: obtaining initial multimodal data of the target object; performing rule verification on the initial multimodal data according to preset data rules to filter out intermediate multimodal data that does not comply with the preset data rules; inputting the intermediate multimodal data into a generative adversarial network model to obtain corrected target multimodal data, so as to construct a hypergraph structure for the target object based on the corrected target multimodal data.

[0009] In an optional embodiment of the present disclosure, the intermediate multimodal data is input into a generative adversarial network model to obtain corrected target multimodal data, including: inputting the intermediate multimodal data into the generative adversarial network model to obtain corrected first target multimodal data; inputting the first target multimodal data into a discriminator model to obtain an identification result; if the probability value corresponding to the identification result is greater than or equal to a probability threshold, constructing a hypergraph structure for the target object based on the corrected first target multimodal data; or, if the probability value corresponding to the identification result is less than the probability threshold, retraining the generative adversarial network model to obtain an adjusted generative adversarial network model, and inputting the intermediate multimodal data into the adjusted generative adversarial network model to obtain corrected second target multimodal data, so as to construct a hypergraph structure for the target object based on the corrected second target multimodal data; the probability value obtained by inputting the second target multimodal data into the discriminator model is greater than or equal to the probability threshold.

[0010] In an optional embodiment of the present disclosure, before constructing a hypergraph structure for a target object based on multimodal data, the method further includes: performing data preprocessing on the multimodal data, and performing feature extraction on the preprocessed multimodal data to construct a hypergraph structure for the target object; wherein the data preprocessing includes one or more of data cleaning processing, missing value filling, and data standardization processing on the multimodal data.

[0011] In an optional embodiment of the present disclosure, a hypergraph structure of a target object is input into a pre-trained hypergraph neural network model to obtain a root cause analysis result of poor perceived quality of the target object, including: determining an initial node set consisting of multiple initial nodes, an initial node feature matrix, a hyperedge set, a hyperedge feature matrix, and an association matrix of hyperedges to which each initial node belongs according to the hypergraph structure, and obtaining a node set associated with each hyperedge, a degree matrix of the initial node, a degree matrix of the hyperedge, and a weight matrix of each hyperedge; inputting the initial node set, the initial node feature matrix, the hyperedge set, the hyperedge feature matrix, and an association matrix of hyperedges to which each initial node belongs, as well as the node set associated with each hyperedge, the degree matrix of the initial node, the degree matrix of the hyperedge, and the weight matrix of each hyperedge into multiple hypergraph convolutional network layers in the hypergraph neural network model to obtain a node feature matrix of the target node associated with the poor perceived quality; inputting the node feature matrix of the target node into a classifier model to obtain a root cause analysis result of poor perceived quality of the target object.

[0012] In an optional embodiment of the present disclosure, a node feature matrix of the target node is input into a classifier model to obtain a root cause analysis result of the perceived poor quality, including: obtaining classifier parameters of the classifier model, the classifier parameters including weight parameters and bias parameters; based on the node feature matrix and classifier parameters of the target node, determining a predicted root cause analysis result of the perceived poor quality corresponding to the target node.

[0013] In an optional embodiment of the present disclosure, the target node is a plurality of nodes, and the node feature matrix of the target node is input into the classifier model to obtain the root cause analysis result of the perceived poor quality, including: obtaining the classifier parameters of the classifier model, the classifier parameters including weight parameters and bias parameters; constructing a classification function based on the node feature matrix and the classifier parameters of the target node to determine the initial predicted root cause analysis result of the perceived poor quality corresponding to the target node based on the classification function; performing weighted aggregation processing on the initial predicted root cause analysis result corresponding to each target node to obtain the target predicted root cause analysis result of the perceived poor quality corresponding to the target object.

[0014] In an optional embodiment of the present disclosure, an actual root cause analysis result of the target node corresponding to the perceived quality difference is obtained; and an error value between the actual root cause analysis result and the predicted root cause analysis result is determined according to a pre-constructed cross entropy loss function, so as to update the classifier parameters of the classifier model according to the error value.

[0015] In a second aspect, an embodiment of the present disclosure provides a root cause analysis device for perceptual poor quality, the device comprising: a data acquisition module, used to acquire multimodal data of a target object; a hypergraph structure construction module, used to construct a hypergraph structure for the target object based on the multimodal data, the hypergraph structure comprising a plurality of initial nodes and a plurality of hyperedges for connecting the initial nodes, the plurality of initial nodes being used to represent a plurality of feature vectors obtained after feature extraction of the multimodal data, and the plurality of hyperedges being used to respectively represent the correlation between the plurality of feature vectors; an input module, used to input the multimodal data into a pre-trained hypergraph neural network model, to obtain a root cause analysis result of the perceptual poor quality of the target object;

[0016] In an optional embodiment of the present disclosure, the multimodal data includes two or more of network operation data for the target object, external environment data associated with the target object, and associated object data corresponding to the target object.

[0017] In an optional embodiment of the present disclosure, the device also includes a rule verification module, wherein the data acquisition module is specifically used to acquire initial multimodal data of the target object; the rule verification module is used to perform rule verification on the initial multimodal data according to preset data rules to screen out intermediate multimodal data that does not comply with the preset data rules; the input module can also be used to input the intermediate multimodal data into the generative adversarial network model to obtain corrected target multimodal data, so as to construct a hypergraph structure for the target object based on the corrected target multimodal data.

[0018] In an optional embodiment of the present disclosure, the input module can also be used to input the intermediate multimodal data into the generative adversarial network model to obtain the corrected first target multimodal data; input the first target multimodal data into the discriminator model to obtain the identification result; if the probability value corresponding to the identification result is greater than or equal to the probability threshold, then construct a hypergraph structure for the target object based on the corrected first target multimodal data; or, if the probability value corresponding to the identification result is less than the probability threshold, then retrain the generative adversarial network model to obtain an adjusted generative adversarial network model, and input the intermediate multimodal data into the adjusted generative adversarial network model to obtain the corrected second target multimodal data, so as to construct a hypergraph structure for the target object based on the corrected second target multimodal data; the probability value obtained by inputting the second target multimodal data into the discriminator model is greater than or equal to the probability threshold.

[0019] In an optional embodiment of the present disclosure, the device also includes a preprocessing module, which is used to perform data preprocessing on the multimodal data and perform feature extraction on the preprocessed multimodal data to construct a hypergraph structure for the target object; wherein the data preprocessing includes one or more of data cleaning processing, missing value filling, and data standardization processing on the multimodal data.

[0020] In an optional embodiment of the present disclosure, an input module is specifically used to determine an initial node set composed of multiple initial nodes, an initial node feature matrix, a hyperedge set, a hyperedge feature matrix, and an association matrix of the hyperedges to which each initial node belongs according to a hypergraph structure, and to obtain a node set associated with each hyperedge, a degree matrix between the initial nodes, a degree matrix of the hyperedges, and a weight matrix of each hyperedge; the initial node set, the initial node feature matrix, the hyperedge set, the hyperedge feature matrix, and the association matrix of the hyperedges to which each initial node belongs, as well as the node set associated with each hyperedge, the degree matrix of the initial node, the degree matrix of the hyperedges, and the weight matrix of each hyperedge are input into multiple hypergraph convolutional network layers in a hypergraph neural network model to obtain a node feature matrix of a target node associated with poor perceptual quality; the node feature matrix of the target node is input into a classifier model to obtain a root cause analysis result of poor perceptual quality for the target object.

[0021] In some example embodiments of the present disclosure, the input module is specifically used to obtain classifier parameters of the classifier model, and the classifier parameters include weight parameters and bias parameters; based on the node feature matrix and classifier parameters of the target node, the predicted root cause analysis result of the perceived quality difference corresponding to the target node is determined.

[0022] In an optional embodiment of the present disclosure, the target node is a plurality of nodes, and the input module is specifically used to obtain classifier parameters of the classifier model, the classifier parameters including weight parameters and bias parameters; a classification function is constructed based on the node feature matrix and the classifier parameters of the target node to determine the initial predicted root cause analysis result of the perceptual quality difference corresponding to the target node based on the classification function; the initial predicted root cause analysis results corresponding to each target node are weighted and aggregated to obtain the target predicted root cause analysis result of the perceptual quality difference corresponding to the target object.

[0023] In an optional embodiment of the present disclosure, the device also includes a parameter updating module, which is used to obtain an actual root cause analysis result of the perceived quality difference corresponding to the target node; determine the error value between the actual root cause analysis result and the predicted root cause analysis result according to a pre-constructed cross entropy loss function, so as to update the classifier parameters of the classifier model according to the error value.

[0024] In a third aspect, an embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, which implements the above-mentioned root cause analysis method of perceptual quality difference when executed by a processor.

[0025] In a fourth aspect, an embodiment of the present disclosure provides an electronic device, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform the above-mentioned root cause analysis method of perceptual quality difference by executing the executable instructions.

[0026] In a fifth aspect, an embodiment of the present disclosure provides a computer program product, including a computer program, which is executed by a processor to implement the above-mentioned root cause analysis method of perceptual quality difference.

[0027] The technical solution disclosed in this disclosure has the following beneficial effects:

[0028] The root cause analysis method of the above-mentioned perceptual poor quality obtains multimodal data of the target object; constructs a hypergraph structure for the target object according to the multimodal data, wherein the hypergraph structure includes multiple initial nodes and multiple hyperedges for connecting the initial nodes, the multiple initial nodes are used to represent multiple feature vectors obtained after feature extraction of the multimodal data, and the multiple hyperedges are used to respectively represent the correlation between the multiple feature vectors; the hypergraph structure of the target object is input into a pre-trained hypergraph neural network model to obtain the root cause analysis result of the perceptual poor quality of the target object; wherein the hypergraph neural network model includes multiple hypergraph convolutional network layers and a classifier model, the multiple hypergraph convolutional network layers are used to perform hypergraph convolution processing on the hypergraph structure of the target object, and obtain the node feature matrix of the target node associated with the perceptual poor quality; the classifier model is used to determine the root cause analysis result of the perceptual poor quality according to the node feature matrix of the target node. On the one hand, by acquiring the multimodal data of the target object, this method can integrate multiple data sources and comprehensively reflect the network status, providing a rich analytical basis for the subsequent root cause analysis of the perceived quality difference, and constructing a hypergraph structure of the multimodal data of the target object through the above multimodal data, so as to connect multiple nodes through hyperedges, so as to fully express the multivariate relationship between each feature vector in the complex application scenario, effectively capture the correlation between multimodal data, and thus better adapt and express the complex system interaction in complex application scenarios such as high-speed rail operation, and improve the accuracy and speed of the root cause analysis of the user's perceived quality difference. On the other hand, the use of the hypergraph neural network model to perform the root cause analysis of the perceived quality difference of the target object can rely on the adaptive ability of the model for dynamic learning, so as to cope with the rapidly changing and complex and diverse fault modes in complex application scenarios such as high-speed rail operation, and can avoid possible problems such as missed detection and false detection when new faults occur, thereby improving the scope of application and validity of root cause analysis for various types of environments. On the other hand, through multimodal data processing, hypergraph neural network model training and root cause analysis, real-time root cause analysis of perceived quality issues in scenarios such as high-speed rail can be achieved efficiently and accurately, greatly shortening the troubleshooting time, thereby improving the efficiency of root cause analysis and processing, and providing operators with efficient technical tools and intelligent solutions.

[0029] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and are used together with the specification to explain the principles of the present disclosure. Obviously, the drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0031] Figure 1 A schematic diagram of a root cause analysis method for perceived poor quality shown in one of the related technical solutions of this exemplary embodiment is shown;

[0032] Figure 2 An architectural diagram showing a root cause analysis system for perceived poor quality according to the exemplary embodiment of the present invention is shown;

[0033] Figure 3 A flow chart showing a root cause analysis method of perceived poor quality according to the exemplary embodiment;

[0034] Figure 4 A schematic diagram of a data preprocessing process in this exemplary embodiment is shown;

[0035] Figure 5A Another schematic diagram of data preprocessing process in this exemplary embodiment is shown;

[0036] Figure 5B A flow chart showing another data preprocessing method in this exemplary embodiment;

[0037] Figure 6 A schematic diagram showing a model training process of one of the hypergraph neural network models in this exemplary embodiment;

[0038] Figure 7 A schematic diagram of the overall processing process of a root cause analysis of poor perceived quality according to the exemplary embodiment is shown;

[0039] Figure 8 A schematic diagram of the structure of a root cause analysis device for poor perceived quality in this exemplary embodiment is shown;

[0040] Fig. 9 A schematic structural diagram of an electronic device in this exemplary embodiment is shown. DETAILED DESCRIPTION

[0041] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, exemplary embodiments can be implemented in a variety of forms and should not be construed as being limited to the examples set forth herein; on the contrary, these embodiments are provided so that the present disclosure will be more comprehensive and complete, and the concepts of the exemplary embodiments are fully conveyed to those skilled in the art. The described features, structures, or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known technical solutions are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0042] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0043] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the steps. For example, some steps may be decomposed, while some steps may be combined or partially combined, so the actual execution order may change according to the actual situation.

[0044] In the relevant technical background, with the vigorous development of mobile communication networks, mobile communications have penetrated into all aspects of users' lives, thereby providing users with a wide range of network services. The present disclosure embodiment takes the application scenario of the high-speed railway network as an example. With the rapid development of my country's high-speed railway network, the "eight vertical and eight horizontal" high-speed railway network has gradually improved, and passengers have higher and higher demands on the communication quality of mobile terminals during the operation of high-speed railways, especially during long-distance travel, and have high expectations for the stability and clarity of services such as audio and video running on mobile terminals. Therefore, it is necessary to monitor the performance of wireless networks along the high-speed railway in real time, accurately identify and locate the root causes of user-perceived poor quality problems such as audio and video running on mobile terminals, so as to make subsequent improvements and upgrades. It should be explained that Root Cause Analysis (RCA) is a systematic process designed to identify the root causes of problems, failures or events.

[0045] In order to solve the problem of poor perceived quality of audio and video services in high-speed rail scenarios, related technical solutions often use manual analysis or one-by-one matching through preset detection algorithms to locate the cause of poor perceived quality in high-speed rail scenarios. Figure 1 The above-mentioned related technical solutions of matching one by one through the preset detection algorithm are exemplified.

[0046] Figure 1 A schematic diagram of a root cause analysis method for perceived quality difference shown in one of the related technical solutions of this exemplary embodiment is shown, referring to Figure 1 As shown in the figure, the method mainly obtains the key performance indicator data of multiple target sampling points of the problem high-speed railway section, and determines the problem type based on the obtained key performance indicator data (key performance indicator data such as Figure 1 The call voice drive test data shown in the figure) is matched with the corresponding target detection algorithm one by one, and finally the root cause of the problem is determined. Figure 1 A variety of root cause detection algorithms are provided. The call voice test data is input and the output results of each detection algorithm are used to determine whether the problem exists, thereby realizing the root cause analysis process of poor quality.

[0047] However, the above-mentioned related technical solutions mainly have the following technical problems:

[0048] 1) Figure 1 The solution shown mainly relies on predefined detection algorithms and the obtained key performance indicator data for analysis, resulting in poor adaptability to real-time and complex and changing network environments, and cannot provide comprehensive support for application scenarios or fields. For example, Figure 1 The root cause analysis method shown is difficult to cope with the massive test data and rapidly changing network requirements in the high-speed rail environment, providing comprehensive support for the planning, construction, optimization, perception improvement and evaluation of the high-speed rail network.

[0049] 2) The relevant technical solutions lack the ability to dynamically learn and adapt to new perception quality problems, making it difficult to cope with the rapidly changing and complex and diverse failure modes in high-speed rail scenarios. It is also very easy to cause technical problems such as missed detection and false detection when new failures occur, which in turn affects the accuracy of root cause analysis.

[0050] 3) The use of manual analysis or one-by-one matching through a preset detection algorithm results in a time-consuming, costly and inefficient root cause analysis process for perceived poor quality.

[0051] In view of the above problems, the exemplary embodiments of the present disclosure propose a root cause analysis method for perceived quality difference, which can be applied to any application scenario where the real-time root cause of user perceived quality difference needs to be analyzed and located, especially in application scenarios where the real-time root cause of user perceived quality difference in complex scenarios can be analyzed and located, such as real-time root cause analysis of perceived quality difference of users running audio and video applications on user terminals in application scenarios such as high-speed railway environments and subway operation environments. The method obtains multimodal data of the target object and inputs the multimodal data into a pre-trained hypergraph neural network model (Hypergraph Neural Networks, HGNN for short), thereby obtaining the root cause analysis result of perceived quality difference for the target object based on the hypergraph neural network model. On the one hand, unlike traditional graph neural networks (GNN for short), hypergraph neural networks can handle more complex relationships and structures, that is, the hypergraph structure constructed by the hypergraph neural network model through multiple nodes and multiple hyperedges for connecting each node can fully express the complex multivariate relationship in the application scenario, and thus is easy to apply to any complex application scenario. In addition, the hypergraph convolutional network layer improves the expressiveness of the model and the accuracy of the result analysis, so that the source of the problem can be located quickly and intelligently, improving the accuracy and speed of root cause analysis.

[0052] The following will be combined Figure 2 The root cause analysis method and device of the perceived quality difference proposed in the present disclosure are applied to Figure 1 An embodiment of the system architecture of the exemplary application environment is described.

[0053] Figure 2 FIG. 1 shows an architecture diagram of a root cause analysis system for perceptual quality degradation according to the exemplary embodiment; Figure 2 As shown, it is a schematic diagram of an application scenario provided by an embodiment of the present application. In this scenario, a terminal device 201 and a server 202 may be included.

[0054] The terminal device 201 can be, for example, a mobile phone, a tablet computer (PAD), a laptop computer, a desktop computer, a smart TV, a smart vehicle-mounted device, a smart wearable device, a smart TV, an aircraft, or any other device involving poor perceived quality. The terminal device 201 can be installed with a target application, and the target application can have functions such as root cause analysis of triggering poor perceived quality for users and displaying the actual root cause of poor perceived quality to users. The application involved in the embodiment of the present disclosure can be a software client or a client such as a web page or a small program. The server 202 is a server corresponding to the software or a web page, a small program, etc., and does not limit the specific type of the client.

[0055] Server 202 may be a background server of the target application, used to provide corresponding background services for it, such as root cause analysis service of poor user perceived quality, etc. It may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, i.e., content delivery networks (CDNs), as well as big data and artificial intelligence platforms, but is not limited thereto.

[0056] It should be noted that the root cause analysis method of the perceived poor quality in the embodiment of the present application can be executed by the terminal device 201 or the server 202 alone, or can be executed by the server 202 and the terminal device 201 together. When executed by the terminal device 201 or the server 202 alone, the model training and application process can be implemented by the terminal device 201 or the server 202 alone. For example, the model can be trained on the terminal device 201 by the collected sample data to adjust the model parameters of the model. The terminal device 201 can then use the trained model to generate the root cause analysis result, or the server 202 can also perform the above process. When executed by the server 202 and the terminal device 201 together, the server 202 can train the model and then deploy the trained model to the terminal device 201, and the terminal device 201 implements the root cause analysis process of the perceived poor quality, or part of the model training or application process can be implemented by the terminal device 201, and part of the process can be implemented by the server 202, and the two cooperate to implement the model training or application process. In actual application, specific configuration can be made according to the situation, and this application does not make any specific limitations here.

[0057] Both the server 202 and the terminal device 201 may include one or more processors, memories, and interactive I / O interfaces, etc. In addition, the server 202 may also be configured with a database, which may be used to store the model parameters obtained through training, etc. The memories of the server 202 and the terminal device 201 may also store program instructions required to be executed in the root cause analysis method of the perceived poor quality provided by the embodiment of the present disclosure, and these program instructions, when executed by the processor, may be used to implement the model training process or the root cause analysis process of the perceived poor quality provided by the embodiment of the present disclosure.

[0058] It should be noted that when the root cause analysis method for perceptual poor quality provided by the embodiment of the present disclosure is executed separately by the server 202 or the terminal device 201, the above-mentioned application scenario may also only include a single device of the server 202 or the terminal device 201, or the server 202 and the terminal device 201 may also be considered to be the same device. Of course, in actual application, when the root cause analysis method for perceptual poor quality provided by the embodiment of the present disclosure is jointly executed by the server 202 and the terminal device 201, the server 202 and the terminal device 201 may also be the same device, that is, the server 202 and the terminal device 201 may be different functional modules of the same device, or virtual devices virtualized from the same physical device.

[0059] In the embodiment of the present disclosure, the terminal device 201 and the server 202 can be directly or indirectly connected to each other through one or more networks 203. The network 203 can be a wired network or a wireless network. For example, the wireless network can be a mobile cellular network or a wireless fidelity (Wireless-Fidelity, WIFI) network. Of course, it can also be other possible networks, which are not limited in the embodiment of the present application. It should be noted that Figure 2 The figure is only an example. In fact, the number of terminal devices and servers is not limited and is not specifically limited in the embodiments of the present application.

[0060] For example, in an exemplary embodiment, a user may trigger a root cause analysis request through a terminal device 201. The server 202 may obtain multimodal data of a target object; construct a hypergraph structure for the target object according to the multimodal data, wherein the hypergraph structure includes multiple initial nodes and multiple hyperedges for connecting each initial node, wherein the multiple initial nodes are used to represent multiple feature vectors obtained after feature extraction of the multimodal data, and the multiple hyperedges are used to represent the association relationship between the multiple feature vectors; the hypergraph structure of the target object is input into a pre-trained hypergraph neural network model to obtain a root cause analysis result of the perceived poor quality of the target object; wherein the hypergraph neural network model includes multiple hypergraph convolutional network layers and a classifier model, wherein the multiple hypergraph convolutional network layers are used to perform hypergraph convolution processing on the hypergraph structure of the target object to obtain a node feature matrix of the target node associated with the perceived poor quality; and the classifier model is used to determine the root cause analysis result of the perceived poor quality according to the node feature matrix of the target node.

[0061] Below, in combination with the application scenarios described above, the method provided by the exemplary implementation of the present application is described with reference to the accompanying drawings. It should be noted that the above application scenarios are only shown to facilitate understanding of the spirit and principle of the present application, and the implementation of the present application is not limited in this respect. It should be noted that the following method can be executed by the above terminal device or server, or by the terminal device and the server together. Here, it is specifically shown by taking the execution of the terminal device or the server as an example.

[0062] See also Figure 3 As shown, Figure 3 A flowchart showing a root cause analysis method of perceived poor quality in this exemplary embodiment is shown. The method can be executed by a computer device, which can be Figure 2 The terminal device or server shown. Figure 3 The root cause analysis method of the perceived poor quality provided by the embodiment of the present disclosure includes the following steps 301 to 303:

[0063] Step 301: Acquire multimodal data of a target object.

[0064] Step 302: construct a hypergraph structure for the target object based on the multimodal data, wherein the hypergraph structure includes multiple initial nodes and multiple hyperedges for connecting the initial nodes, the multiple initial nodes are used to represent multiple feature vectors obtained after feature extraction of the multimodal data, and the multiple hyperedges are used to respectively represent the correlation relationship between the multiple feature vectors.

[0065] Step 303: Input the hypergraph structure of the target object into a pre-trained hypergraph neural network model to obtain a root cause analysis result of the perceived poor quality of the target object.

[0066] The hypergraph neural network model includes multiple hypergraph convolutional network layers and a classifier model. The multiple hypergraph convolutional network layers are used to perform hypergraph convolution processing on the hypergraph structure of the target object to obtain the node feature matrix of the target node associated with the perceived quality difference. The classifier model is used to determine the root cause analysis result of the perceived quality difference based on the node feature matrix of the target node. The root cause analysis result refers to the fundamental factor that causes the perceived quality difference of the target object.

[0067] pass Figure 3In the technical solution provided by the illustrated embodiment, on the one hand, the method can integrate multiple data sources by acquiring multimodal data of the target object, comprehensively reflect the network status, and provide rich analysis basis for the subsequent root cause analysis of the perceived quality difference, and construct a hypergraph structure of the multimodal data of the target object through the above multimodal data, so as to connect multiple nodes through hyperedges, so as to fully express the multivariate relationship between each feature vector in the complex application scenario, effectively capture the correlation between the multimodal data, and thus better adapt and express the complex system interaction in complex application scenarios such as high-speed rail operation, and improve the accuracy and speed of the root cause analysis of the user's perceived quality difference. On the other hand, the use of a hypergraph neural network model to perform root cause analysis of the perceived quality difference of the target object can rely on the adaptive ability of the model for dynamic learning, so as to cope with the rapidly changing and complex and diverse fault modes in complex application scenarios such as high-speed rail operation, and can avoid possible problems such as missed detection and false detection when new faults occur, thereby improving the scope of application and validity period of root cause analysis for various types of environments. On the other hand, through multimodal data processing, hypergraph neural network model training and root cause analysis, real-time root cause analysis of perceived quality issues in scenarios such as high-speed rail can be achieved efficiently and accurately, greatly shortening the troubleshooting time, thereby improving the efficiency of root cause analysis and processing, and providing operators with efficient technical tools and intelligent solutions.

[0068] The following will be combined with specific embodiments to Figure 3 The specific implementation methods of each step in the embodiment shown are described in detail:

[0069] In step 301, multimodal data of a target object is acquired.

[0070] Among them, the target object can be a device used to carry user perception in an application scenario. For example, the target object can be a user terminal device that runs audio and video applications and carries audio and video perception. The target object can also be an application that requires network configuration in the user terminal device, such as audio applications, video applications, shopping applications, voice / video call applications, social applications, etc. However, the target application usually has poor perceived quality in changing network applications, such as video playback failure, picture sending failure, etc., which in turn affects the user experience. Therefore, network operators need to monitor the wireless network performance along the high-speed rail in real time, accurately identify and locate the root causes of poor user perception quality problems such as audio and video running on mobile terminals, so as to improve and enhance them.

[0071] In an optional embodiment of the present disclosure, the multimodal data of the target object includes two or more of network operation data for the target object, external environment data associated with the target object, and associated object data corresponding to the target object.

[0072] Among them, the associated object data is the data corresponding to the object with which the target object has an associated relationship. For example, taking the high-speed rail operation scenario as an example, the target object can be a mobile terminal in the high-speed rail operation scenario or various applications that require network configuration running on the mobile terminal, and the associated object can be a high-speed rail, and the above-mentioned mobile terminal or various applications that require network configuration running on the mobile terminal are carried on the high-speed rail object. It should be understood that in this exemplary application scenario, the target object can be a network device on a high-speed rail train (such as a call device, a television, etc. in the train) or an application running in the network device.

[0073] In one embodiment, the network operation data of the target object may include: route test data (such as the strength, quality, speed, etc. of the wireless signal), base station configuration data, historical alarm data, etc. The network operation data also includes data used to demonstrate the perceived quality of the application, such as data used to characterize the audio and video quality, including voice connection rate, voice drop rate, video playback success rate, picture sending success rate, etc.

[0074] The associated object data corresponding to the target object may be the relevant data of the associated object. For example, if the associated object in the high-speed rail operation scene is a high-speed rail, the associated object data may be train operation data; and the external environment data associated with the target object may be geographic information data, operation environment data, etc. It should be explained that the multimodal data corresponding to the acquired target object may be flexibly adjusted according to the target object and the scene in which the target object is located.

[0075] Exemplarily, the server can obtain the multimodal data of the target object from the configured database, or can obtain the corresponding multimodal data from the target object in real time. For obtaining the multimodal data of the target object from the configured database, the data in the database is updated in real time so that the server can obtain the latest multimodal data.

[0076] In this embodiment, considering that in complex application scenarios, the network environment of the target object is affected by many factors, the multimodal data generated by combining network operation data, external operating environment, related object data, etc. can more accurately reflect the actual situation, thereby improving the robustness of the system and the accuracy of the root cause analysis results of poor perception quality.

[0077] Furthermore, after obtaining the multimodal data of the target object in step 301, it is not directly input into the pre-trained hypergraph neural network model, but needs to be preprocessed to ensure the accuracy and standardization of the data, thereby avoiding system delays and misjudgments.

[0078] In an optional embodiment of the present disclosure, initial multimodal data of the target object is obtained; the initial multimodal data is rule-checked according to preset data rules to filter out intermediate multimodal data that does not conform to the preset data rules; the intermediate multimodal data is input into a generative adversarial network model to obtain corrected target multimodal data, so as to construct a hypergraph structure for the target object based on the corrected target multimodal data.

[0079] In this embodiment, the generative adversarial network model is used to regenerate input data that does not conform to the preset data rules, thereby outputting data that conforms to the preset data rules. It can be based on a large amount of data that does not conform to the preset data rules and data that conforms to the preset data rules as positive and negative samples for model training.

[0080] For example, the initial multimodal data obtained usually does not fall within the preset physical and logical range values ​​due to human input, abnormal test equipment or other reasons, especially for the road test data in the network operation data, there is a large amount of non-standard data. The above non-standard multimodal data seriously affects the training process of the hypergraph neural network model and the accuracy of the root cause analysis results obtained by the hypergraph neural network model.

[0081] In response to the above technical problems, the embodiments provided by the present disclosure can perform rule verification on the initial multimodal data of the target object through preset data rules, thereby screening out intermediate multimodal data that does not conform to the preset data rules, and input the intermediate multimodal data into the generative adversarial network model for regeneration, so as to correct the data that does not conform to the rules of the initial multimodal data.

[0082] In one embodiment, the above-mentioned preset data rules may include, but are not limited to: data type verification rules, data range verification rules, and data format verification rules.

[0083] For ease of understanding, the following will be combined Figure 4 , the data preprocessing process such as rule verification and data regeneration in the above embodiment is exemplified.

[0084] Figure 4 A schematic diagram of a data preprocessing process in this exemplary embodiment is shown, referring to Figure 4As shown, the initial multimodal data of the target object is input into the preset data rules for rule verification to obtain the verification result. If the initial multimodal data of the target object are all successful and pass the rule verification of the above preset data rules, the initial multimodal data of the target object can be directly input into the pre-trained hypergraph neural network model to obtain the corresponding root cause analysis result. On the contrary, if part or all of the initial multimodal data of the target object fails to pass the rule verification, the intermediate multimodal data that does not conform to the preset data rules is screened out, and the intermediate multimodal data is input into the generative adversarial network model (also referred to as the GAN network model) for data regeneration, thereby obtaining the target multimodal data that conforms to the preset data rules.

[0085] When executing the above embodiment, due to the high error probability of network operation data, in order to save computing resources, in an optional embodiment, only the network operation data in the multimodal data can be subjected to rule verification, that is, the network operation data in the multimodal data is input into the preset data rule for rule verification. If the rule verification (rule detection) is passed, the network operation data and other data (for example, external environment data associated with the target object, target object corresponding associated object data) are directly combined to construct a hypergraph structure for the target object. On the contrary, if the network data does not pass the rule verification, the network operation data is used as one of the input parameters of the GAN network model, and the other modal data in the initial multimodal data except the network operation data is also used as another input parameter of the GAN network model, so as to generate the target multimodal data that meets the specification requirements through the GAN network model, and then construct the hypergraph structure for the target object in the subsequent steps.

[0086] In order to further improve the accuracy of data, so as to improve the accuracy of the root cause analysis results through the hypergraph neural network model. Figure 4 A discriminator model is added on the basis shown in the figure to verify the accuracy of the target multimodal data output by the GAN network model through the feedback results of the discriminator model, so as to achieve continuous optimization of the GAN network model.

[0087] Figure 5A A schematic diagram of a data preprocessing process in this exemplary embodiment is shown schematically, see Figure 5A As shown, in Figure 4 Based on the data preprocessing shown in the figure, a discriminator model is added after the GAN network model. The specific implementation process can be as follows Figure 5B As shown:

[0088] Figure 5B A schematic diagram of a data preprocessing process in this exemplary embodiment is shown schematically. Figure 5BIn another optional embodiment of the present disclosure, the above steps input the intermediate multimodal data into the generative adversarial network model to obtain the corrected target multimodal data, so as to input the corrected target multimodal data into the hypergraph neural network model, including the following steps 501-505:

[0089] Step 501: input the intermediate multimodal data into a generative adversarial network model to obtain corrected first target multimodal data.

[0090] Among them, the intermediate multimodal data is the data that does not meet the preset data rules screened out based on the above embodiment. The intermediate multimodal data is input into the generative adversarial network model to obtain the corrected first target multimodal data.

[0091] Step 502: Input the first target multimodal data into the discriminator model to obtain a discrimination result.

[0092] Among them, the discriminator model is used to distinguish whether the input intermediate multimodal data is data directly obtained from the actual network or data generated based on the model. The output result of the discriminator model is a probability value. The higher the probability value, the higher the probability that the data input to the discriminator model is real data; conversely, the lower the probability value, the higher the probability that the data input to the discriminator model is data generated by the model, that is, the lower the probability that the data is directly obtained from the actual network.

[0093] Step 503: Determine whether the probability value corresponding to the identification result is greater than or equal to the probability threshold.

[0094] If the probability value corresponding to the identification result is greater than or equal to the probability threshold, step 504 is executed to construct a hypergraph structure for the target object based on the corrected first target multimodal data.

[0095] If the probability value corresponding to the identification result is less than the probability threshold, then step 505 is executed to retrain the generative adversarial network model, and the intermediate multimodal data is input into the adjusted generative adversarial network model to obtain the corrected second target multimodal data. Finally, step 506 is executed to construct a hypergraph structure for the target object based on the corrected second target multimodal data.

[0096] The probability value obtained by inputting the second target multimodal data into the discriminator model is greater than or equal to the probability threshold. Retraining the generative adversarial network model can be adjusting the model parameters of the GAN network model, the weight values ​​of each model parameter, etc. based on the sample data.

[0097] Exemplarily, since the discriminator model is used to distinguish whether the input intermediate multimodal data is data directly obtained from the actual network or data generated based on the model. The discriminator model is added to the back of the GAN network model to identify the first target multimodal data output by the GAN network model. If the probability value corresponding to the identification result is greater than or equal to the probability threshold, it indicates that the data input to the discriminator model is the real acquired data, that is, the first target multimodal data generated by the GAN network model is closer to the real acquired data, thereby improving the reliability and accuracy of the first target multimodal data generated based on the GAN network model, so as to generate the reliability and accuracy of the subsequent root cause analysis results.

[0098] On the contrary, if the probability value corresponding to the identification result is less than the probability threshold, the data input to the discriminator model is model-generated data, indicating that the first target multimodal data generated by the GAN network model is far from the real data, and the first target multimodal data generated based on the GAN network model is unreliable. The GAN network model needs to be retrained to adjust the model parameters of the GAN network model, the weights of the model parameters, etc., so that the generated second target multimodal data is closer to the real data.

[0099] In this embodiment, through the combination of the GAN network model and the discriminator model, the output back propagation loss of the discriminator model on the data generated by the GAN network model can be used to adjust the model parameter weights of the GAN network model (also referred to as the generator) so that the generated data is closer to the real data, and the target model data generated based on the GAN network model can be made more accurate. It can be seen that through the discriminator model in this embodiment, the important features of the generated data are determined again to ensure that the generated data with outliers eliminated has applicability, and then a hypergraph structure for the target object can be constructed so as to continue to be input into the hypergraph neural network model for root cause analysis.

[0100] In step 302, a hypergraph structure for the target object is constructed based on the multimodal data. The hypergraph structure includes multiple initial nodes and multiple hyperedges for connecting the initial nodes. The multiple initial nodes are used to represent multiple feature vectors obtained after feature extraction of the multimodal data, and the multiple hyperedges are used to respectively represent the correlation relationship between the multiple feature vectors.

[0101] Among them, a hyperedge can connect two or more than two initial nodes. The initial nodes are all feature vectors obtained after feature extraction of multimodal data.

[0102] Exemplarily, after obtaining the multimodal data of the target object in any of the above embodiments, a hypergraph structure for the target object can be constructed based on the multimodal data. The hypergraph structure is composed of multiple initial nodes and multiple hyperedges for connecting the initial nodes. The initial nodes are multiple feature vectors obtained after feature extraction of the multimodal data, and the hyperedges represent the correlation relationship between the multiple feature vectors. In the hypergraph structure, a hyperedge is allowed to connect multiple initial nodes, not just two initial nodes, so the multivariate relationship of data in complex application scenarios can be fully expressed.

[0103] In an optional embodiment of the present disclosure, before executing step 302, data preprocessing may be performed on the multimodal data, and feature extraction may be performed on the preprocessed multimodal data to construct a hypergraph structure for the target object.

[0104] Among them, data preprocessing includes one or more of data cleaning, missing value filling, and data standardization processing on multimodal data.

[0105] Exemplarily, before constructing the hypergraph structure for the target object according to the multimodal data in step 302, the input multimodal data may be preprocessed in advance, such as cleaning the input multimodal data, filling in default values, standardizing the data format, etc. For example, Figure 6 Then, feature extraction is performed on the multimodal data after data preprocessing to obtain multiple feature vectors, so that the multiple feature vectors are used as initial nodes and the correlation between the multiple feature vectors is used as a hyperedge to construct a hypergraph structure for the target object. For example, Figure 6 Part ② is shown in the figure. For example, taking the high-speed rail operating environment as an example, the initial node can represent the user equipment, base station and alarm event, and the hyperedge can be the connection between the user equipment and the base station, the connection between the base station and the alarm, the connection between the user equipment and the alarm event, etc. Based on the above nodes and hyperedges, a hypergraph structure for the target object can be constructed. For details, please refer to Figure 6 Part ③ shown.

[0106] In this embodiment, pre-processing of the multimodal data such as data cleaning, missing value filling, and data standardization can further ensure the reliability of the data.

[0107] In step 303, the hypergraph structure of the target object is input into a pre-trained hypergraph neural network model to obtain a root cause analysis result of the perceived poor quality of the target object.

[0108] Among them, the hypergraph neural network model includes multiple hypergraph convolutional network layers and a classifier model. The multiple hypergraph convolutional network layers are used to perform hypergraph convolution processing on the hypergraph structure of the target object to obtain the node feature matrix of the target node associated with the perceived quality difference; the classifier model is used to determine the root cause analysis result of the perceived quality difference based on the node feature matrix of the target node.

[0109] The target node is a node associated with the perceived quality difference selected from the initial nodes; the classifier model is a fully connected layer; the root cause analysis result may include the root cause category and the probability value corresponding to the root cause category.

[0110] This part can refer to Figure 6 As shown in Part ④, it can be seen that the hypergraph neural network model contains multiple hypergraph convolutional network layers and classifier models. Based on the multiple hypergraph convolutional network layers and classifier models in the hypergraph neural network model, the following technical features can be achieved:

[0111] In an optional embodiment of the present disclosure, when executing the above step 303, multiple initial nodes are determined to form an initial node set, an initial node feature matrix, a hyperedge set, a hyperedge feature matrix, and an association matrix of the hyperedges to which each initial node belongs, according to the hypergraph structure, and the node set associated with each hyperedge, the degree matrix between the initial nodes, the degree matrix of the hyperedges, and the weight matrix of each hyperedge are obtained; the initial node set, the initial node feature matrix, the hyperedge set, the hyperedge feature matrix, and the association matrix of the hyperedges to which each initial node belongs, as well as the node set associated with each hyperedge, the degree matrix between the initial nodes, the degree matrix of the hyperedges, and the weight matrix of each hyperedge are input into multiple hypergraph convolutional network layers in the hypergraph neural network model to obtain a node feature matrix of the target node associated with the perceptual quality difference; the node feature matrix of the target node is input into the classifier model to obtain a root cause analysis result of the perceptual quality difference of the target object.

[0112] Among them, based on the above hypergraph structure, the initial node set, the initial node feature matrix, the hyperedge set and the hyperedge feature matrix can be determined. Taking the above high-speed rail operation environment as an example, assuming that the user equipment initial node V_UE, the base station node V_BS, and the alarm node V_Alarm are created, then the initial node set V = V_UE∪V_BS∪V_Alarm. Correspondingly, the hyperedge set: assuming that a drive test hyperedge ε_DriveTest is created, which connects the corresponding V_UE, V_BS and V_Alarm, and an alarm hyperedge ε_Alarm is created, which connects the corresponding V_BS and V_Alarm, then the constructed hyperedge set is ε = ε_DriveTest∪ε_Alarm.

[0113] The initial node feature matrix X can be defined as: Where F0 is the first feature dimension of the initial node. For example, F0=16 means that the feature dimension of each initial node is 16. The hyperedge feature matrix E can be defined as: Among them, F e is the characteristic dimension of the hyperedge; and the hyperedge incidence matrix H can be defined as: H∈R |v|×|ε| , which is a sparse matrix used to represent the association relationship between hyperedges and nodes. Among them, H ve =1 means that an initial node v belongs to the hyperedge e; and H ve = 0 indicates that an initial node v does not belong to the hyperedge e. The node set V associated with each hyperedge e , which is used to represent the set of all nodes associated with a specific hyperedge e, and is expressed as: V e ={v∈V|H ve =1}.

[0114] Taking the two-layer hypergraph convolutional network layer as an example, for each initial node v in the initial node set, the core of the first layer of hypergraph convolutional network layer is as follows:

[0115]

[0116] In formula (1), is the feature representation of the initial node v in the first layer of the hypergraph convolutional network, that is, the new node feature matrix output by the first layer; deg(v) is the degree matrix of the initial node v, which is used to represent the number of hyperedges associated with the initial node v; deg(e) is the degree matrix of the hyperedge e, which is used to represent the number of initial nodes associated with the hyperedge e; ε v is the set of hyperedges associated with the initial node v; V e is the set of all initial nodes that make up the hyperedge e; is the weight matrix of hyperedge e; σ is the activation function; X u is the initial node feature matrix of the initial node set u, and F0 is the feature dimension of the initial node v.

[0117] The above formula (1) is executed for each initial node in the initial node set u, and the initial node feature matrix of the target object in the first layer of the hypergraph convolutional network layer can be obtained: It can be seen that after the first layer of hypergraph convolutional network, the feature dimension corresponding to the new initial node feature matrix is ​​updated from F0 to F1. For example, it is transformed from 16 dimensions to 32 dimensions.

[0118] Continue to pass the initial node feature matrix H output by the first layer of hypergraph convolutional network layer (1) Input to the second layer of hypergraph convolutional network layer and calculate based on the following formula (2):

[0119]

[0120] In formula (2), is the feature representation of the initial node v in the second layer of the hypergraph convolutional network, that is, the new initial node feature matrix output by the second layer; deg(v) is the degree matrix of the initial node v, which is used to represent the number of hyperedges associated with the initial node v; deg(e) is the degree matrix of the hyperedge e, which is used to represent the number of initial nodes contained in the hyperedge e; ε v is the set of hyperedges associated with the initial node v; V e is the set of all initial nodes that make up the hyperedge e, that is, the set of adjacent initial nodes mentioned above; is the weight matrix of hyperedge e; σ is the activation function; It is the feature representation of the initial node set u in the first layer of the hypergraph convolutional network.

[0121] The above formula (1) is executed for each initial node in the initial node set u, and the initial node feature matrix of the target object in the first layer of the hypergraph convolutional network layer can be obtained: It can be seen that after the first layer of hypergraph convolutional network, the feature dimension corresponding to the new initial node feature matrix is ​​updated from F1 to F2. For example, it is transformed from 32 dimensions to 5 dimensions.

[0122] It should be noted that in the second hypergraph convolutional network layer, the third feature dimension F2 is smaller than the second feature dimension F1 and smaller than the first feature dimension F0, thereby realizing the compression process of the feature dimension.

[0123] Through the above multiple hypergraph convolutional network layers, all feature vectors related to the target object (i.e., initial nodes) can be aggregated to the target feature vector (i.e., target node) related to the perceived quality difference, so that only the target node related to the perceived quality difference can participate in the classifier model in the future, so as to realize the analysis of the root cause analysis results. This process can reduce the amount of calculation and save computing resources.

[0124] After obtaining the node feature matrix of the target node as described above, the node feature matrix of the target node can be input into the classifier model to obtain the root cause analysis result of the poor perceived quality of the target object.

[0125] In order to facilitate the description of the specific process of inputting the node feature matrix of the target node into the classifier model to obtain the root cause analysis result of the perceived poor quality of the target object, the following will be explained in conjunction with a specific embodiment:

[0126] Embodiment 1:

[0127] In an optional embodiment of the present disclosure, classifier parameters of the classifier model are obtained; a classification function is constructed based on the node feature matrix and the classifier parameters of the target node to determine a predicted root cause analysis result of the perceived quality difference corresponding to the target node based on the classification function.

[0128] Among them, the classifier parameters include weight parameters and bias parameters.

[0129] Exemplarily, the classifier model is used to map the features output by the hypergraph convolutional layer into root cause category probabilities, i.e., the node feature matrix of the target node is used as the input of the classifier model, so that the output result obtained is the predicted root cause analysis result of the perceptual quality difference corresponding to the target node, which is a classification function.

[0130] Based on the weight parameters and bias parameters of the classifier model, the predicted root cause analysis result of the target node corresponding to the perceived poor quality can be obtained, as shown in the following formula (3):

[0131]

[0132] In formula (3), is the probability distribution of the target node v belonging to c root cause categories. The softmax() function plays a vital role in multi-classification problems. It can convert any real value into a probability distribution so that each element ranges between (0,1) and the sum of all elements is 1; W c is the weight parameter of the classifier; b c is the bias parameter of the classifier; is the node feature matrix of the target node based on the output of multiple hypergraph convolutional network layers.

[0133] The number of nodes of the target node can be one or more. When the number of nodes of the target node is multiple, the corresponding

[0134] For example, for each target node (such as a base station, an alarm event), the corresponding root cause category and the probability of each root cause category are output. For example, the target nodes include user equipment node A and base station node B. The root cause analysis result corresponding to the poor perceived quality of user equipment node A is: Indicates insufficient signal coverage (80%) and network congestion (20%); the root cause analysis results corresponding to poor perceived quality at base station node B are: Indicates network congestion (70%) and hardware failure (30%).

[0135] It should be understood that the root cause category is a fault cause related to the target node. If the probability corresponding to a certain root cause category is 0, it indicates that the perceived quality difference of the current target node is irrelevant to the root cause category.

[0136] Embodiment 2:

[0137] Based on the above embodiment 1, an implementation method is also provided:

[0138] In another optional embodiment of the present disclosure, when there are multiple target nodes, classifier parameters of the classifier model are obtained, and the classifier parameters include weight parameters and bias parameters; a classification function is constructed based on the node feature matrix of the target node and the classifier parameters to determine the initial predicted root cause analysis result of the perceptual quality difference corresponding to the target node based on the classification function; the initial predicted root cause analysis result corresponding to each target node is weighted and aggregated to obtain the target predicted root cause analysis result of the perceptual quality difference corresponding to the target object.

[0139] For example, after obtaining the initial predicted root cause analysis results (e.g., root cause categories and probability values ​​corresponding to the root cause categories) corresponding to each target node in the above formula (3), the classification probabilities of all target nodes can be weighted and aggregated based on the following formula (4), thereby calculating the probability P of the root cause category c of the global (i.e., for the target object) root cause: c :

[0140]

[0141] In formula (4), P c is the probability distribution of the target object corresponding to all target nodes corresponding to the root cause category c; V core is the set of all target nodes; w v is the weight of node v; Represents the classification probability of node v in category c.

[0142] By polymerizing V core The classification probability of all target nodes in the network is calculated to generate the overall root cause analysis results. For example, the root cause analysis results of the poor audio and video quality of mobile terminals on the current high-speed rail line are: network congestion (60%), base station failure (25%), and environmental interference (15%).

[0143] Furthermore, when the root cause analysis results are obtained using the classifier model, in order to improve the accuracy, the classifier model needs to be continuously updated and optimized.

[0144] In an optional embodiment of the present disclosure, an actual root cause analysis result of the target node corresponding to the perceived quality difference is obtained; and an error value between the actual root cause analysis result and the predicted root cause analysis result is determined according to a pre-constructed cross entropy loss function, so as to update the classifier parameters of the classifier model according to the error value.

[0145] Exemplarily, the error value between the actual root cause analysis result and the predicted root cause analysis result can be calculated so that when the error value is greater than an error threshold, the classifier parameters of the classifier model are adjusted; when the error value is less than or equal to the error threshold, there is no need to adjust the classifier parameters of the classifier model.

[0146] That is, after the root cause analysis results are obtained in (4) above, the cross entropy loss function (also known as the Adam optimization loss function) can be used to update the classifier parameters of the classifier model. Specifically, the error value between the actual root cause analysis results and the predicted root cause analysis results can be determined by the cross entropy loss function:

[0147]

[0148] In formula (3), y vc is the actual root cause analysis result of node v in category c. Represents the predicted root cause analysis result of node v in category c predicted by the model, where c is the number of categories, C is the total number of categories, and N is the total number of target nodes.

[0149] Through the above embodiments, on the one hand, the method can integrate multiple data sources by acquiring multimodal data of the target object, comprehensively reflect the network status, and provide rich analysis basis for the subsequent root cause analysis of perceived quality difference, and construct a hypergraph structure of the multimodal data of the target object through the above multimodal data, so as to connect multiple nodes through hyperedges, so as to fully express the multivariate relationship between each feature vector in the complex application scenario, effectively capture the correlation between multimodal data, and thus better adapt and express complex system interactions in complex application scenarios such as high-speed rail operation, and improve the accuracy and speed of root cause analysis of user perceived quality difference. On the other hand, the use of a hypergraph neural network model to perform root cause analysis of perceived quality difference of the target object can rely on the adaptive ability of the model for dynamic learning, so as to cope with the rapidly changing and complex and diverse fault modes in complex application scenarios such as high-speed rail operation, and can avoid possible problems such as missed detection and false detection when new faults occur, thereby improving the scope of application and validity period of root cause analysis of various types of environments. On the other hand, through multimodal data processing, hypergraph neural network model training and root cause analysis, real-time root cause analysis of perceived quality issues in scenarios such as high-speed rail can be achieved efficiently and accurately, greatly shortening the troubleshooting time, thereby improving the efficiency of root cause analysis and processing, and providing operators with efficient technical tools and intelligent solutions.

[0150] The following will refer to Figure 7 The entire root cause analysis process of the root cause analysis method of the perceived poor quality according to the exemplary embodiment of the present disclosure is described in detail.

[0151] First, the initial multimodal data of the target object is obtained.

[0152] Exemplarily, the initial multimodal data not only covers network operation data such as drive test data, base station configuration data, and historical alarm data, but also combines external environment data associated with the target object and associated object data corresponding to the target object.

[0153] The multimodal data of the target object is preprocessed.

[0154] That is, the initial multimodal data of the target object is input into the preset data rules, and the initial multimodal data is checked according to the preset data rules to filter out the intermediate multimodal data that does not meet the preset data rules;

[0155] Inputting the intermediate multimodal data into the generative adversarial network model to obtain the corrected first target multimodal data; inputting the first target multimodal data into the discriminator model to obtain the identification result;

[0156] If the probability value corresponding to the identification result is greater than or equal to the probability threshold, a hypergraph structure for the target object is constructed based on the corrected first target multimodal data;

[0157] On the contrary, if the probability value corresponding to the identification result is less than the probability threshold, the generative adversarial network model is retrained to obtain an adjusted generative adversarial network model, and the intermediate multimodal data is input into the adjusted generative adversarial network model to obtain the corrected second target multimodal data, so as to construct a hypergraph structure for the target object based on the corrected second target multimodal data.

[0158] Among them, the probability value obtained by inputting the second target multimodal data into the discriminator model is greater than or equal to the probability threshold.

[0159] A hypergraph structure is constructed based on the rule-compliant multimodal data obtained in any of the above steps, so as to input the constructed hypergraph structure into the hypergraph neural network model.

[0160] After that, the data is preprocessed again, and the input hypergraph structure is subjected to hypergraph convolution processing through multiple hypergraph convolution network layers in the hypergraph neural network model, so as to obtain the node feature matrix of the target node associated with the perceived quality difference;

[0161] Then, the mapping between the node feature matrix of the target node and the root cause category probability is realized through the classifier model in the hypergraph neural network model, so as to obtain the root cause analysis results corresponding to each target node, or the overall root cause analysis results after weighted aggregation of each target node. And based on the pre-built cross entropy loss function, the error value between the actual root cause analysis result and the predicted root cause analysis result is determined, so as to update the classifier parameters of the classifier model according to the error value, thereby optimizing the classifier model to improve the accuracy of the root cause analysis results.

[0162] Finally, in an optional embodiment of the present disclosure, by Figure 7 The feedback loop, expert system and self-adjustment mechanism in the continuous learning module are used to continuously optimize and update the pre-trained hypergraph neural network model with the predicted root cause analysis results and the actual root cause analysis results, so that the model can adapt to new data and the changing network environment. For example, in the actual operation environment of the high-speed rail, the hypergraph neural network model will self-adjust according to the audio and video perception data monitored in real time and the feedback from expert tests, so as to capture new patterns in multimodal data by comparing the differences between the predicted results and the actual observed data, so as to timely adjust the model parameters and construct the hypergraph structure, so as to dynamically update the model according to the new data and fault patterns in the complex environment, so as to ensure the long-term effectiveness and adaptability of the model, thereby reducing the problem of missed detection.

[0163] In order to implement the above-mentioned root cause analysis method of perceptual poor quality, an apparatus for root cause analysis of perceptual poor quality is provided in one embodiment of the present disclosure. Figure 8 The schematic architecture diagram of the root cause analysis device of the perceived poor quality is schematically shown.

[0164] The root cause analysis device 800 of perceived poor quality includes a data acquisition module 801 , a hypergraph structure construction module 802 and an input module 803 .

[0165] Among them, the data acquisition module 801 is used to acquire the multimodal data of the target object; the hypergraph structure construction module 802 is used to construct a hypergraph structure for the target object according to the multimodal data, the hypergraph structure includes multiple initial nodes and multiple hyperedges for connecting the initial nodes, the multiple initial nodes are used to represent the multiple feature vectors obtained after feature extraction of the multimodal data, and the multiple hyperedges are used to respectively represent the correlation between the multiple feature vectors; the input module 803 is used to input the multimodal data into a pre-trained hypergraph neural network model to obtain the root cause analysis result of the perceptual quality difference of the target object; wherein the hypergraph neural network model includes multiple hypergraph convolutional network layers and a classifier model, the multiple hypergraph convolutional network layers are used to perform hypergraph convolution processing on the hypergraph structure of the target object to obtain the node feature matrix of the target node associated with the perceptual quality difference; the classifier model is used to determine the root cause analysis result of the perceptual quality difference according to the node feature matrix of the target node.

[0166] In an optional embodiment of the present disclosure, the multimodal data includes one or more of network operation data for the target object, external environment data associated with the target object, and associated object data corresponding to the target object.

[0167] In an optional embodiment of the present disclosure, the device also includes a rule verification module, wherein the data acquisition module 8010 is specifically used to obtain initial multimodal data of the target object; the rule verification module is used to perform rule verification on the initial multimodal data according to preset data rules to filter out intermediate multimodal data that does not comply with the preset data rules; the input module 803 can also be used to input the intermediate multimodal data into the generative adversarial network model to obtain corrected target multimodal data, so as to construct a hypergraph structure for the target object based on the corrected target multimodal data.

[0168] In an optional embodiment of the present disclosure, the input module 803 can also be used to input the intermediate multimodal data into the generative adversarial network model to obtain the corrected first target multimodal data; input the first target multimodal data into the discriminator model to obtain the identification result; if the probability value corresponding to the identification result is greater than or equal to the probability threshold, then construct a hypergraph structure for the target object based on the corrected first target multimodal data; or, if the probability value corresponding to the identification result is less than the probability threshold, then retrain the generative adversarial network model to obtain an adjusted generative adversarial network model, and input the intermediate multimodal data into the adjusted generative adversarial network model to obtain the corrected second target multimodal data, so as to construct a hypergraph structure for the target object based on the corrected second target multimodal data; the probability value obtained by inputting the second target multimodal data into the discriminator model is greater than or equal to the probability threshold.

[0169] In an optional embodiment of the present disclosure, the device also includes a preprocessing module, which is used to perform data preprocessing on the multimodal data and perform feature extraction on the preprocessed multimodal data to construct a hypergraph structure for the target object; wherein the data preprocessing includes one or more of data cleaning processing, missing value filling, and data standardization processing on the multimodal data.

[0170] In an optional embodiment of the present disclosure, an input module is specifically used to determine an initial node set composed of multiple initial nodes, an initial node feature matrix, a hyperedge set, a hyperedge feature matrix, and an association matrix of the hyperedges to which each initial node belongs according to a hypergraph structure, and to obtain a node set associated with each hyperedge, a degree matrix of the initial node, a degree matrix of the hyperedge, and a weight matrix of each hyperedge; the initial node set, the initial node feature matrix, the hyperedge set, the hyperedge feature matrix, and the association matrix of the hyperedges to which each initial node belongs, as well as the node set associated with each hyperedge, the degree matrix of each initial node, the degree matrix of the hyperedge, and the weight matrix of each hyperedge are input into multiple hypergraph convolutional network layers in a hypergraph neural network model to obtain a node feature matrix of a target node associated with poor perceptual quality; the node feature matrix of the target node is input into a classifier model to obtain a root cause analysis result of poor perceptual quality for the target object.

[0171] In some example embodiments of the present disclosure, the input module is specifically used to obtain classifier parameters of the classifier model, and the classifier parameters include weight parameters and bias parameters; based on the node feature matrix and classifier parameters of the target node, the predicted root cause analysis result of the perceived quality difference corresponding to the target node is determined.

[0172] In an optional embodiment of the present disclosure, the target node is a plurality of nodes, and the input module is specifically used to obtain classifier parameters of the classifier model, the classifier parameters including weight parameters and bias parameters; a classification function is constructed based on the node feature matrix and the classifier parameters of the target node to determine the initial predicted root cause analysis result of the perceptual quality difference corresponding to the target node based on the classification function; the initial predicted root cause analysis results corresponding to each target node are weighted and aggregated to obtain the target predicted root cause analysis result of the perceptual quality difference corresponding to the target object.

[0173] In an optional embodiment of the present disclosure, the device also includes a parameter updating module, which is used to obtain an actual root cause analysis result of the perceived quality difference corresponding to the target node; determine the error value between the actual root cause analysis result and the predicted root cause analysis result according to a pre-constructed cross entropy loss function, so as to update the classifier parameters of the classifier model according to the error value.

[0174] The root cause analysis device 800 for perceptual poor quality provided in the embodiment of the present disclosure can execute the technical solution of the root cause analysis method for perceptual poor quality in any of the above-mentioned embodiments, and its implementation principle and beneficial effects are similar to the implementation principle and beneficial effects of the root cause analysis method for perceptual poor quality. Please refer to the implementation principle and beneficial effects of the root cause analysis method for perceptual poor quality, which will not be repeated here.

[0175] In an exemplary embodiment of the present disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the above method of the present specification is stored. In some possible implementations, various aspects of the present invention may also be implemented in the form of a program product, which includes a program code, and when the program product is run on a terminal device, the program code is used to enable the terminal device to execute the steps according to various exemplary embodiments of the present invention described in the above "Exemplary Method" section of the present specification.

[0176] The program product for implementing the above method according to an embodiment of the present invention may adopt a portable compact disc read-only memory (CD-ROM) and include program code, and may be run on a terminal device, such as a personal computer. However, the program product of the present invention is not limited thereto, and in this document, a readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus or a device.

[0177] The program product may adopt any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0178] Computer readable signal media may include data signals propagated in baseband or as part of a carrier wave, in which readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Readable signal media may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0179] The program code contained on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, radio frequency (RF), etc., or any suitable combination of the foregoing.

[0180] Program code for performing the operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computing device (e.g., via the Internet using an Internet Service Provider).

[0181] In an exemplary embodiment of the present disclosure, an electronic device capable of implementing the above method is also provided.

[0182] It will be appreciated by those skilled in the art that various aspects of the present invention may be implemented as a system, method or program product. Therefore, various aspects of the present invention may be specifically implemented in the following forms, namely: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which may be collectively referred to herein as a "circuit", "module" or "system".

[0183] Refer to the following Fig. 9 An electronic device 900 according to this embodiment of the present invention will be described. Fig. 9 The electronic device 900 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0184] like Fig. 9As shown, the electronic device 900 is in the form of a general computing device. The components of the electronic device 900 may include, but are not limited to: the at least one processing unit 910, the at least one storage unit 920, a bus 930 connecting different system components (including the storage unit 920 and the processing unit 910), and a display unit 940.

[0185] The storage unit stores program codes, which can be executed by the processing unit 910, so that the processing unit 910 performs the steps of various exemplary embodiments of the present invention described in the above “Exemplary Method” section of this specification. For example, the processing unit 910 can perform the following steps: Figure 3 Steps 301 to 303 shown in FIG.

[0186] The storage unit 920 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 9201 and / or a cache storage unit 9202 , and may further include a read-only storage unit (ROM) 9203 .

[0187] The storage unit 920 may also include a program / utility 9204 having a set (at least one) of program modules 9205, such program modules 9205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0188] Bus 930 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0189] The electronic device 900 may also communicate with one or more external devices 1000 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 900, and / or communicate with any device that enables the electronic device 900 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 950. In addition, the electronic device 900 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 960. As shown, the network adapter 960 communicates with other modules of the electronic device 900 via a bus 930. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 900, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, disk arrays (Redundant Arrays of Independent Disks, RAID) systems, tape drives, and data backup storage systems.

[0190] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0191] In addition, the above-mentioned figures are only schematic illustrations of the processes included in the method according to an exemplary embodiment of the present invention, and are not intended to be limiting. It is easy to understand that the processes shown in the above-mentioned figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0192] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0193] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure. The specification and examples are to be considered as exemplary only, and the true scope and spirit of the present disclosure are indicated by the claims.

[0194] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A root cause analysis method for perceived poor quality, characterized in that: include: Obtain multimodal data of the target object; Constructing a hypergraph structure for the target object according to the multimodal data, wherein the hypergraph structure includes a plurality of initial nodes and a plurality of hyperedges for connecting the initial nodes, the plurality of initial nodes are used to represent a plurality of feature vectors obtained after feature extraction of the multimodal data, and the plurality of hyperedges are used to respectively represent associations between the plurality of feature vectors; Inputting the hypergraph structure of the target object into a pre-trained hypergraph neural network model to obtain a root cause analysis result of the perceived poor quality of the target object; wherein the hypergraph neural network model includes a plurality of hypergraph convolutional network layers and a classifier model, and the plurality of hypergraph convolutional network layers are used to perform hypergraph convolution processing on the hypergraph structure of the target object to obtain a node feature matrix of the target node associated with the perceived poor quality; The classifier model is used to determine a root cause analysis result of the perceived poor quality based on a node feature matrix of the target node.

2. The method according to claim 1, characterized in that The multimodal data includes two or more of network operation data for the target object, external environment data associated with the target object, and associated object data corresponding to the target object.

3. The method according to claim 1 or 2, characterized in that: The method further comprises: Acquiring initial multimodal data of the target object; Performing rule verification on the initial multimodal data according to preset data rules to filter out intermediate multimodal data that does not conform to the preset data rules; The intermediate multimodal data is input into a generative adversarial network model to obtain corrected target multimodal data, so as to construct a hypergraph structure for the target object according to the corrected target multimodal data.

4. The method according to claim 3, characterized in that: The step of inputting the intermediate multimodal data into a generative adversarial network model to obtain corrected target multimodal data comprises: Inputting the intermediate multimodal data into a generative adversarial network model to obtain corrected first target multimodal data; Inputting the first target multimodal data into a discriminator model to obtain a discriminant result; If the probability value corresponding to the identification result is greater than or equal to a probability threshold, constructing a hypergraph structure for the target object based on the corrected first target multimodal data; Alternatively, if the probability value corresponding to the identification result is less than the probability threshold, the generative adversarial network model is retrained to obtain an adjusted generative adversarial network model, and the intermediate multimodal data is input into the adjusted generative adversarial network model to obtain corrected second target multimodal data, so as to construct a hypergraph structure for the target object based on the corrected second target multimodal data; the probability value obtained by inputting the second target multimodal data into the discriminator model is greater than or equal to the probability threshold.

5. The method according to claim 1, characterized in that The step of inputting the hypergraph structure of the target object into a pre-trained hypergraph neural network model to obtain a root cause analysis result of the poor perceived quality of the target object includes: Determine, according to the hypergraph structure, an initial node set composed of the multiple initial nodes, an initial node feature matrix, a hyperedge set, a hyperedge feature matrix, and an association matrix of hyperedges to which each of the initial nodes belongs, and obtain a node set associated with each hyperedge, a degree matrix of the initial node, a degree matrix of the hyperedge, and a weight matrix of each hyperedge; Inputting the initial node set, the initial node feature matrix, the hyperedge set, the hyperedge feature matrix, the association matrix of the hyperedges to which each of the nodes belongs, the node set associated with each hyperedge, the degree matrix of the initial node and the degree matrix of the hyperedge, and the weight matrix of each hyperedge into multiple hypergraph convolutional network layers in the hypergraph neural network model to obtain a node feature matrix of the target node associated with the perceptual quality difference; The node feature matrix of the target node is input into the classifier model to obtain a root cause analysis result of the poor perceived quality of the target object.

6. The method according to claim 1 or 5, characterized in that: The step of inputting the node feature matrix of the target node into the classifier model to obtain the root cause analysis result of the perceived poor quality includes: Obtaining classifier parameters of the classifier model, wherein the classifier parameters include weight parameters and bias parameters; A classification function is constructed based on the node feature matrix and classifier parameters of the target node, so as to determine a predicted root cause analysis result of the perceived poor quality corresponding to the target node based on the classification function.

7. The method according to claim 6, characterized in that The method further comprises: Obtaining an actual root cause analysis result of the perceived poor quality corresponding to the target node; According to a pre-constructed cross entropy loss function, an error value between the actual root cause analysis result and the predicted root cause analysis result is determined, so as to update the classifier parameters of the classifier model according to the error value.

8. A root cause analysis device for perceived poor quality, characterized in that: The device comprises: A data acquisition module, used to acquire multimodal data of a target object; A hypergraph structure construction module, used to construct a hypergraph structure for the target object according to the multimodal data, wherein the hypergraph structure includes a plurality of initial nodes and a plurality of hyperedges for connecting the initial nodes, wherein the plurality of initial nodes are used to represent a plurality of feature vectors obtained after feature extraction of the multimodal data, and the plurality of hyperedges are used to respectively represent the association relationship between the plurality of feature vectors; An input module is used to input the hypergraph structure of the target object into a pre-trained hypergraph neural network model to obtain a root cause analysis result of the perceptual quality difference of the target object; wherein the hypergraph neural network model includes multiple hypergraph convolutional network layers and a classifier model, and the multiple hypergraph convolutional network layers are used to perform hypergraph convolution processing on the hypergraph structure of the target object to obtain a node feature matrix of the target node associated with the perceptual quality difference; the classifier model is used to determine the root cause analysis result of the perceptual quality difference based on the node feature matrix of the target node.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the root cause analysis method of perceived poor quality according to any one of claims 1 to 7 is implemented.

10. An electronic device, characterized in that: include: processor; as well as A memory, configured to store executable instructions of the processor; The processor is configured to execute the root cause analysis method of perceptual poor quality according to any one of claims 1 to 7 by executing the executable instructions.

Images