Personalized content recommendation method and system realized by utilizing artificial intelligence

Abstract

The invention provides a personalized content recommendation method and system realized by utilizing artificial intelligence, and the method comprises the steps: firstly recognizing to-be-mined application program operation data, and generating operation trajectory data of different abnormal knowledge categories through graph convolution processing; and then, by utilizing a cross-field matching technology, generating a corresponding cross-field matching vector according to the knowledge field of the abnormal knowledge category so as to accurately identify a key graph path. Then, in combination with the key graph path and the category moving trajectory data, abnormal root causes are estimated through intelligent analysis, and detailed abnormal root cause data are generated; and finally, based on the abnormal root cause data, personalized repair suggestions can be recommended, so that a user can be helped to quickly and effectively solve the abnormal problem of the application program, the intelligence and the efficiency of abnormal processing are remarkably improved, and the method has important significance for improving the user experience and the system stability.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to a personalized content recommendation method and system using artificial intelligence. Background Art

[0002] With the rapid development of information technology, various applications are playing an increasingly important role in our daily life and work. However, various anomalies will inevitably occur during the operation of applications, such as system crashes and data errors. These problems not only affect the user experience, but may also cause significant losses to the company. Therefore, how to effectively monitor, analyze and handle application anomalies has become an urgent problem to be solved.

[0003] Traditional exception handling methods mainly rely on manual analysis and empirical judgment, which is inefficient and difficult to deal with complex and changing exception situations. In recent years, the rapid development of artificial intelligence technology has provided new ideas for exception handling. By using machine learning, deep learning and other technologies, artificial intelligence can automatically analyze the operation data of applications, discover abnormal patterns, and provide targeted solutions.

[0004] Among the existing personalized content recommendation methods, artificial intelligence has been widely used in user behavior analysis and interest mining. However, in the field of application exception handling, the application of artificial intelligence is relatively rare. Most of the existing exception handling methods are based on fixed rules and templates, which are difficult to adapt to the complex and changing actual situation. Therefore, developing an artificial intelligence-based personalized content recommendation method for application exception handling has important practical significance and application value. Summary of the invention

[0005] In view of the above-mentioned problems, in combination with the first aspect of the present application, an embodiment of the present application provides a personalized content recommendation method implemented by artificial intelligence, the method comprising: Determine the application running data to be mined, and perform graph convolution processing on the application running data to generate category running trajectory data of multiple abnormal knowledge categories; According to the category running trajectory data of the multiple abnormal knowledge categories, cross-field matching is performed according to the category knowledge fields of the multiple abnormal knowledge categories respectively, to generate cross-field matching vectors corresponding to the category running trajectory data of the multiple abnormal knowledge categories respectively; Determine at least one key graph path for the application program running data based on the cross-field matching vectors respectively corresponding to the category running trajectory data of the plurality of abnormal knowledge categories; Estimating abnormal root causes according to the at least one key graph path and the category operation trajectory data of the plurality of abnormal knowledge categories, and generating abnormal root cause data for the application operation data; Personalized repair suggestion content is recommended based on the abnormal root cause data of the application running data.

[0006] In a possible implementation manner of the first aspect, the multiple abnormal knowledge categories include a logic error category and an environment error category; the category operation trajectory data of the multiple abnormal knowledge categories include logic-related operation trajectory data and environment-related operation trajectory data; The step of estimating the abnormal root cause based on the at least one key graph path and the category operation trajectory data of the plurality of abnormal knowledge categories to generate abnormal root cause data for the application operation data includes: Integrating the environment-related running trajectory data and the at least one key graph path to generate integrated trajectory data; Using a trajectory analysis model to perform pattern recognition on a logic-type operation vector sequence corresponding to the logic-related operation trajectory data, and generate a pattern recognition vector diagram; An abnormal root cause decision network is used to estimate the abnormal root cause based on the integrated trajectory data and the pattern recognition vector map to generate abnormal root cause data for the application running data.

[0007] For example, in a possible implementation of the first aspect, the training steps of the trajectory analysis model and the abnormal root cause decision network include: A temporary trajectory analysis model is generated by performing parameter learning based on a first template logic class operation vector sequence of the first template application operation data and template representation information of the first template application operation data; the first template logic class operation vector sequence is generated by performing graph convolution processing on logic-related operation trajectory data of the first template application operation data; Obtaining a second template logic class operation vector sequence, template environment related operation trajectory data, and template abnormality root cause data of the second template application operation data; the second template logic class operation vector sequence is generated by performing graph convolution processing on the logic related operation trajectory data of the second template application operation data; Using the temporary trajectory analysis model, pattern recognition is performed on the second template logic class operation vector sequence to generate a second pattern recognition vector diagram; Using the initialized basic abnormal root cause decision network to estimate the abnormal root cause according to the template environment related running trajectory data and the second pattern recognition vector diagram, to generate an abnormal root cause estimation result for the second template application running data; Based on the abnormal root cause estimation result and the template abnormal root cause data, iterative joint parameter optimization is performed on the temporary trajectory analysis model and the basic abnormal root cause decision network to generate a trajectory analysis model and an abnormal root cause decision network with completed parameter optimization.

[0008] For example, in a possible implementation of the first aspect, the step of performing parameter learning based on a first template logic class operation vector sequence of the first template application operation data and template representation information of the first template application operation data to generate a temporary trajectory analysis model includes: Acquire a first template logic class running vector sequence of first template application running data and template representation information for the first template application running data; Performing pattern recognition on the first template logic class operation vector sequence using the initialized basic trajectory analysis model to generate a first pattern recognition vector graph; Using the initialized basic abnormal root cause decision network to embed the first pattern recognition vector graph to generate embedding representation information for the first template application running data; Iteratively optimize the parameters of the basic trajectory analysis model according to the embedded representation information and the template representation information to generate a temporary trajectory analysis model.

[0009] For example, in a possible implementation of the first aspect, the determining at least one key graph path for the application running data based on the cross-field matching vectors respectively corresponding to the category running trajectory data of the multiple abnormal knowledge categories includes: Summarizing the cross-field matching vectors corresponding to the category running trajectory data of the multiple abnormal knowledge categories, and generating a summary matching vector result; At least one critical graph path for the application execution data is determined based on the summarized matching vector results.

[0010] On the other hand, an embodiment of the present application also provides a personalized content recommendation system implemented using artificial intelligence, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0011] Based on the above aspects, the embodiments of the present application can significantly improve the efficiency and accuracy of application exception handling. Through graph convolution processing, key information in application operation data can be automatically mined, and category operation trajectory data of multiple abnormal knowledge categories can be generated, providing a data basis for subsequent analysis. The application of cross-field matching technology can accurately match historical operation trajectory data, generate corresponding cross-field matching vectors, and provide strong support for the determination of key graph paths. Based on cross-field matching vectors, key graph paths closely related to the occurrence of exceptions can be accurately identified, which directly reflects the context and process of the occurrence of exceptions, and provides important clues for abnormal root cause estimation. Through abnormal root cause estimation, detailed abnormal root cause data can be generated. These data not only include abnormal type and cause description, but also involve information such as the scope of impact, which provides a scientific basis for subsequent repair suggestions. Finally, personalized repair suggestion content recommendation is performed based on abnormal root cause data. These suggestions are highly targeted and operational, and can help users quickly and effectively solve abnormal problems in applications and improve system performance and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a schematic diagram of the execution flow of the personalized content recommendation method implemented by artificial intelligence provided in an embodiment of the present application.

[0013] Figure 2 It is a schematic diagram of the hardware architecture of a personalized content recommendation system implemented using artificial intelligence provided in an embodiment of the present application. DETAILED DESCRIPTION

[0014] The present application will be described in detail below with reference to the accompanying drawings. Figure 1 This is a flow chart of a personalized content recommendation method using artificial intelligence provided by an embodiment of the present application. The personalized content recommendation method using artificial intelligence is introduced in detail below.

[0015] Step S110, determining the application running data to be mined, and performing graph convolution processing on the application running data to generate category running trajectory data of multiple abnormal knowledge categories.

[0016] In detail, the application operation data to be mined refers to the data generated by the application that needs to be analyzed and processed during its actual operation, including various operation information of the application, such as user behavior, system status, abnormal events, etc. For example, in the backend management system of an e-commerce platform, the application operation data to be mined may include log data of business processes such as user order placement, payment, inventory update, etc., as well as system monitoring data (such as CPU usage, memory occupancy, etc.).

[0017] Graph convolution processing is a deep learning technology for graph-structured data. It extracts and learns the structural information of the graph by applying convolution operations on the nodes and edges of the graph. In graph convolution processing, each node updates its representation by aggregating the information of its neighboring nodes to capture the global structural features of the graph. For example, in the backend management system of an e-commerce platform, graph convolution processing can be used to convert complex business process data into a graph structure, where nodes represent individual steps or components in the business process and edges represent dependencies between steps or components. Through graph convolution processing, the server can extract key features of the business process, such as frequently occurring abnormal steps, key dependencies, etc.

[0018] The exception knowledge category is a label or category for classifying exceptions that occur during the operation of the application. It is usually divided based on factors such as the nature, source or impact of the exception, so as to facilitate targeted analysis and processing of the exception. For example, in the backend management system of an e-commerce platform, the exception knowledge category may include a logic error category and an environmental error category. The logic error category focuses on errors in business processes or business rules, such as duplicate order submissions, mismatched payment amounts, etc.; while the environmental error category focuses on anomalies caused by external environmental factors such as system resources and network conditions, such as excessive CPU usage and network delays.

[0019] The category operation trajectory data refers to the operation data related to the exception generated during the operation of the application under a specific exception knowledge category, which describes the context, process, impact, etc. of the exception. For example, in the backend management system of an e-commerce platform, the category operation trajectory data under the logical error category may include log data related to the order processing process, recording the execution of steps such as order submission and payment, and the exceptions that occurred; while the category operation trajectory data under the environmental error category may include system monitoring data, recording the system resource usage, network delay, etc. when the exception occurred.

[0020] That is, in this embodiment, the server first determines a batch of application operation data to be mined, for example, these application operation data come from the background management system of an e-commerce platform. The server performs graph convolution processing on these application operation data, converting complex operation data into a graph structure, in which nodes represent data entities and edges represent the relationship between data. During the processing, the server divides the data into two categories: "logical error category" and "environmental error category" according to the characteristics of the data. For the logical error category, the server mainly focuses on the logical flow and business rules of the data; for the environmental error category, it focuses on external environmental factors such as system resources and network conditions. Through graph convolution processing, the server generates category operation trajectory data for these two major abnormal knowledge categories.

[0021] Step S120 , performing cross-field matching according to the category running trajectory data of the multiple abnormal knowledge categories and the category knowledge fields of the multiple abnormal knowledge categories respectively, to generate cross-field matching vectors corresponding to the category running trajectory data of the multiple abnormal knowledge categories respectively.

[0022] In detail, cross-field matching is a data analysis technology that compares and matches information between different fields or attributes to discover the association or similarity between them. In exception analysis, cross-field matching can be used to compare the current abnormal operation trajectory data with historical data to find similar or related historical abnormalities. For example, in the backend management system of an e-commerce platform, the server can use cross-field matching technology to compare the operation trajectory data under the current logical error category with the historical logical error data in the cross-field matching knowledge base. By comparing the information of multiple fields such as timestamps, event sequences, data values, etc., the server can find the historical abnormalities that are most similar to the current abnormality and extract the relevant historical association attributes as a reference.

[0023] The cross-field matching vector is a data representation generated during the cross-field matching process, which usually contains the similarities and differences between the current operation trajectory data and the historical data in multiple dimensions. The cross-field matching vector can be used for subsequent data analysis and abnormal root cause estimation. For example, in the backend management system of an e-commerce platform, the server generates a cross-field matching vector of logically related operation trajectory data through cross-field matching. The cross-field matching vector not only contains the similarity information between the current logical error and the historical logical error in logical characteristics (such as similarity score), but may also contain the specific causes of the logical error, related system configuration and other key information. This information will serve as an important basis for subsequent abnormal root cause estimation.

[0024] In this embodiment, the server then performs cross-field matching based on the generated category operation trajectory data. For example, a cross-field matching knowledge base is first determined. This cross-field matching knowledge base contains the historical operation trajectory data of the historical application operation data under the categories of logical errors and environmental errors, as well as the historical association attributes corresponding to these historical operation trajectory data. The server compares the current category operation trajectory data with the historical operation trajectory data, and finds the historical operation trajectory data that is most similar to the current data through feature matching and pattern recognition. Based on these matching results, the server generates corresponding cross-field matching vectors for the current category operation trajectory data. These cross-field matching vectors reflect the similarities and differences between the current operation trajectory data and the historical operation trajectory data in multiple dimensions.

[0025] Step S130 : determining at least one key graph path for the application program running data based on the cross-field matching vectors respectively corresponding to the category running trajectory data of the plurality of abnormal knowledge categories.

[0026] In detail, the key graph path refers to a sequence of nodes and edges in the application operation graph structure that are closely related to the occurrence of exceptions and have a significant impact. It reflects the context and process of the exception occurrence and is an important clue for the root cause analysis of the exception. For example, in the backend management system of the e-commerce platform, the server identifies the key graph path by analyzing the cross-field matching vector. For example, in the logical error category, the server found that a specific order processing flow frequently had errors, and some steps in this process were closely related to the system resource occupancy under the environmental error category. Therefore, the server determines the path composed of the key steps in this order processing flow and the related system resource occupancy as the key graph path, which is the focus of analysis for subsequent abnormal root cause estimation.

[0027] Specifically, in this embodiment, after obtaining the cross-field matching vector, the server begins to determine the key graph path. For example, by analyzing these cross-field matching vectors, data paths that show significant characteristics under multiple abnormal knowledge categories are identified. For example, in the logical error category, a specific business process frequently fails; in the environmental error category, the same business process is also affected by network delays. The server determines these data paths that are significant in multiple categories as key graph paths, which are likely to be the root cause of the application exception.

[0028] Step S140 , estimating abnormal root causes based on the at least one key graph path and the category operation trajectory data of the plurality of abnormal knowledge categories, and generating abnormal root cause data for the application operation data.

[0029] In this embodiment, after determining the key graph path, the server begins to estimate the root cause of the exception. In detail, the key graph path and the category operation trajectory data are combined, and in-depth analysis is performed through complex algorithm models, such as machine learning or deep learning models. In this process, the server not only considers the direct association between the data, but also mines the implicit relationship between the data. Finally, the server generates abnormal root cause data for the application operation data, which describes in detail the specific causes of the abnormality and its scope of influence. In detail, the abnormal root cause data refers to the result generated by the abnormal root cause estimation process, which describes the root cause of the abnormality and its scope of influence. The abnormal root cause data usually includes information such as the abnormal type, cause description, and scope of influence, so as to facilitate subsequent repair suggestions and problem solving. Continuing with the above example, the server generates abnormal root cause data through the abnormal root cause estimation process, which may include abnormal type (such as logical error), cause description (such as payment interface logic error leading to repeated order submission), scope of influence (such as involving multiple orders and users), etc., which will serve as the basis for subsequent repair suggestions and problem solving.

[0030] Step S150: Recommend personalized repair suggestion content based on the abnormal root cause data of the application running data.

[0031] Finally, based on the generated abnormal root cause data, the server starts to recommend personalized repair suggestions. For example, the nature and impact of the abnormal root cause data are first analyzed, and then the most suitable repair solution is matched from the predefined repair suggestion library based on this information. For example, if the abnormality is caused by network latency, the server can recommend solutions such as optimizing network configuration or increasing cache. These repair suggestions are not only targeted at the current abnormal problem, but also take into account the overall performance and stability of the application, aiming to help users quickly and effectively solve problems and improve system performance.

[0032] That is, personalized repair suggestion content recommendation refers to the process of providing targeted repair suggestions for the anomalies of the operation data of a specific application based on the anomaly root cause data, which specifically involves in-depth analysis of the anomaly root cause data and matching the most suitable repair solution from the predefined repair suggestion library. Personalized repair suggestions are designed to help users solve problems quickly and effectively and improve system performance. For example, in the backend management system of an e-commerce platform, the server generates personalized repair suggestions based on the anomaly root cause data. Assuming that the anomaly root cause data indicates that the logical error of the payment interface causes repeated order submission, the server can recommend solutions to fix the logical error from the predefined repair suggestion library, such as modifying the code logic of the payment interface, adding exception handling, etc. These repair suggestions will target the current anomaly problem and take into account the overall performance and stability of the application, aiming to help users solve the problem quickly. At the same time, the server can also optimize and adjust according to user feedback and the actual operation of the system to provide more accurate and effective personalized repair suggestions.

[0033] Based on the above steps, the embodiment of the present application can significantly improve the efficiency and accuracy of application exception handling. Through graph convolution processing, it is possible to automatically mine key information in the application operation data, generate category operation trajectory data of multiple abnormal knowledge categories, and provide a data basis for subsequent analysis. The application of cross-field matching technology can accurately match historical operation trajectory data, generate corresponding cross-field matching vectors, and provide strong support for the determination of key graph paths. Based on cross-field matching vectors, it is possible to accurately identify key graph paths closely related to the occurrence of exceptions, directly reflect the context and process of the occurrence of exceptions, and provide important clues for abnormal root cause estimation. Through abnormal root cause estimation, detailed abnormal root cause data can be generated. These data not only include abnormal type and cause description, but also involve information such as the scope of impact, which provides a scientific basis for subsequent repair suggestions. Finally, personalized repair suggestion content recommendation is performed based on abnormal root cause data. These suggestions are highly targeted and operational, and can help users quickly and effectively solve abnormal problems in applications and improve system performance and stability.

[0034] In a possible implementation, step S120 may include: Step S121, determining a cross-field matching knowledge base, wherein the cross-field matching knowledge base includes historical running track data of historical application running data under the multiple abnormal knowledge categories and historical association attributes corresponding to each of the historical running track data.

[0035] In this embodiment, the server first determines a cross-field matching knowledge base, which is pre-built and continuously updated. The cross-field matching knowledge base contains historical running track data of historical application running data under multiple abnormal knowledge categories, and the historical associated attributes corresponding to each of these historical running track data. These historical running track data may come from past monitoring records, error logs, or data generated by simulation tests.

[0036] For example, in the backend management system of an e-commerce platform, historical operation trajectory data may record the execution of business processes such as user order placement, payment, inventory update, and various exceptions that occur in these processes. Historical association attributes can include the specific cause of the exception, the system status when the exception occurred, and related configuration parameters.

[0037] Step S122, based on the comparison results between the category operation trajectory data of the multiple abnormal knowledge categories and the historical operation trajectory data of the historical application operation data under the multiple abnormal knowledge categories, cross-field matching vectors corresponding to the category operation trajectory data of the multiple abnormal knowledge categories are matched and generated from the cross-field matching knowledge base.

[0038] Next, the server starts to compare the current category running trajectory data with the historical running trajectory data in the knowledge base. The comparison process involves feature matching and pattern recognition, aiming to find the historical running trajectory data that is most similar to the current data.

[0039] Specifically, the server can compare the current logic-related running trajectory data and environment-related running trajectory data with the logic-related historical running trajectory data and environment-related historical running trajectory data in the knowledge base. During the comparison process, the server can focus on multiple dimensions of the data, such as timestamps, event sequences, data values, etc., to determine the similarities and differences between them.

[0040] For example, if the current logic-related operation trajectory data shows that a logic error occurred in a certain order processing process, the server can compare this trajectory with all the logic-related historical operation trajectory data in the knowledge base to find those historical operation trajectory data that also had logic errors in the order processing process. At the same time, the server will also compare these logic-related operation trajectory data with the environment-related historical operation trajectory data to check whether there is any historical operation trajectory data similar to the current environmental conditions.

[0041] Based on the comparison results, the server can match and generate cross-field matching vectors corresponding to each category of operation trajectory data from the cross-field matching knowledge base. These cross-field matching vectors not only reflect the similarity between the current data and the historical operation trajectory data in specific dimensions, but may also contain key information such as the specific causes of the anomaly and related system status.

[0042] Through this process, the server can make full use of the knowledge and experience in the historical operation trajectory data to provide more accurate and comprehensive analysis support for the current application operation data.

[0043] In a possible implementation, step S122 includes: Step S1221 , respectively comparing each category of the category running trajectory data of the multiple abnormal knowledge categories with the historical running trajectory data of the historical application running data under the multiple abnormal knowledge categories, and generating a comparison result for each category of the running trajectory data.

[0044] In this embodiment, the server begins to compare the category operation trajectory data of multiple abnormal knowledge categories of the current application (such as logic-related operation trajectory data and environment-related operation trajectory data) with the historical application operation data stored in the cross-field matching knowledge base under the historical operation trajectory data of the corresponding abnormal knowledge category.

[0045] For example, suppose the current application is an online payment system, and the server monitors a set of category operation trajectory data, including payment process abnormal data under the logic error category (such as repeated order submission, payment amount mismatch, etc.) and system resource shortage data under the environment error category (such as high CPU usage, memory overflow, etc.).

[0046] The server first compares these category operation trajectory data (including payment process abnormality data and system resource shortage data) with the historical payment system operation trajectory data stored in the cross-field matching knowledge base. When comparing, the server can consider multiple factors, such as timestamps, event sequences, data values, etc., to determine the similarity between the current category operation trajectory data and the historical operation trajectory data.

[0047] Step S1222: From the comparison results of each category of running track data, determine the key comparison results that each category of running track data matches.

[0048] In this embodiment, after the comparison is completed, the server can analyze the comparison results of each category of running trajectory data to determine the key comparison results that are most consistent with the current category of running trajectory data. These key comparison results usually represent the historical abnormal situations that are closest to the current abnormality.

[0049] For example, in the comparison results of payment process abnormal data, the server may find several historical operation track data, which also have the problem of duplicate order submission, and the system status, user behavior and other factors when these historical problems occurred are highly similar to the current situation. The server can determine these historical operation track data as the key comparison results of payment process abnormal data.

[0050] Similarly, in the comparison results of the insufficient system resource data, the server may also find several historical operation trajectory data that reflect similar system resource usage, and the consequences of these historical problems are consistent with the current situation. The server can determine these historical operation trajectory data as the key comparison results of the insufficient system resource data.

[0051] Step S1223 , based on the historical running trajectory data corresponding to the key comparison results corresponding to each category of running trajectory data, respectively matching and generating cross-field matching vectors corresponding to each category of running trajectory data from the cross-field matching knowledge base.

[0052] In this embodiment, based on the key comparison results of each category of operation trajectory data, the server can match and generate corresponding cross-field matching vectors from the cross-field matching knowledge base. These cross-field matching vectors not only contain the similarity information between the current category operation trajectory data and the historical operation trajectory data, but may also contain key information such as the specific cause of the abnormality, related system status, etc.

[0053] Specifically, for the key comparison results of payment process abnormal data, the server can extract historical correlation attributes associated with these historical operation trajectory data from the knowledge base (such as system configuration and user operation records when historical problems occurred), and integrate these attributes into a cross-field matching vector, which will be used as a feature representation of the payment process abnormal data for subsequent analysis and abnormal root cause estimation.

[0054] Similarly, for the key comparison results of the insufficient system resource data, the server will also extract the corresponding historical association attributes from the knowledge base and generate a cross-field matching vector as the feature representation of the insufficient system resource data.

[0055] Through these cross-field matching vectors, the server can more accurately understand the abnormal characteristics of the current application running data and provide strong support for subsequent abnormal root cause estimation and repair suggestions.

[0056] In a possible implementation, the multiple abnormal knowledge categories include a logic error category and an environment error category. The category operation trajectory data of the multiple abnormal knowledge categories include logic-related operation trajectory data and environment-related operation trajectory data of the application operation data. The cross-field matching knowledge base includes logic-class historical operation trajectory data of historical application operation data under the logic error category and logic-class historical association attributes corresponding to the logic-class historical operation trajectory data, as well as environment-class historical operation trajectory data of the historical application operation data under the environment error category and environment-class historical association attributes corresponding to the environment-class historical operation trajectory data.

[0057] Step S122 may further include: Step S1224 , performing graph convolution processing respectively on the logic-related running trajectory data and the environment-related running trajectory data to generate a logic-related running vector sequence and an environment-related running vector sequence.

[0058] In this embodiment, in this scenario, it is assumed that an application of an e-commerce platform has anomalies during operation, and the server needs to analyze these anomalies and generate corresponding repair suggestions. The abnormal knowledge category is divided into a logic error category and an environment error category, and the category operation trajectory data includes logic-related operation trajectory data and environment-related operation trajectory data.

[0059] The server first performs graph convolution processing on the logic-related running trajectory data and environment-related running trajectory data in the application running data to be mined. Graph convolution processing is a machine learning technology used to process graph structured data. It can convert complex running data into vector representation, which is convenient for subsequent analysis and comparison.

[0060] For example, the logic-related operation trajectory data may include the execution of business processes such as user order placement, payment, and order cancellation. The server can process this data using a graph convolution processing model, extract key feature information, and encode this information into a logic-like operation vector sequence that can reflect the main features and patterns of the logic-related operation trajectory data.

[0061] Similarly, environment-related running trajectory data can include environmental factors such as system resource usage and network latency. The server will also perform graph convolution on these data to generate an environment-related running vector sequence, which can reflect the main characteristics and status of the environment-related running trajectory data.

[0062] Step S1225, performing logical feature comparison on the logic class operation vector sequence, the logic class historical operation trajectory data and the environment class historical operation trajectory data respectively, and generating a cross-field matching vector of the logic-related operation trajectory data from the logic class historical association attributes and the environment class historical association attributes based on the logical feature comparison result.

[0063] After generating the logic class running vector sequence, the server starts to compare it with the historical data in the cross-field matching knowledge base. The cross-field matching knowledge base contains the logic class historical running trajectory data under the logic error category and the corresponding logic class historical association attributes.

[0064] The server first compares the logical features of the logic class operation vector sequence with the logic class historical operation trajectory data. The logical feature comparison focuses on the logically related features such as business processes and event sequences in the data to determine the similarity between the current data and the historical data.

[0065] If the logic class operation vector sequence is highly similar to a certain logic class historical operation trajectory data in terms of logical characteristics, the server can further consider the logic class history association attributes corresponding to the historical operation trajectory data, which may include the specific causes of the logic error, related system configurations, etc.

[0066] Based on the logical feature comparison results and logical class historical association attributes, the server can generate a cross-field matching vector of logically related running trajectory data. This vector not only contains the similarity information between the current data and the historical data in logical features, but may also contain the specific causes and related information that caused the logical error.

[0067] Step S1226, performing environmental feature comparison on the environment class operation vector sequence, the logic class historical operation trajectory data and the environment class historical operation trajectory data respectively, and generating a cross-field matching vector of the environment-related operation trajectory data from the logic class historical association attributes and the environment class historical association attributes based on the environmental feature comparison result.

[0068] Similar to the processing of logic-related operation trajectory data, the server then compares the environment-related operation vector sequence with the historical data in the cross-field matching knowledge base. This comparison will take into account both the logic-related historical operation trajectory data and the environment-related historical operation trajectory data.

[0069] First, the server compares the environmental characteristics of the environment-related operation vector sequence with the logic-related historical operation trajectory data. Although these data belong to different abnormal knowledge categories, sometimes there may be a correlation between logical errors and environmental factors. The environmental characteristic comparison focuses on the system resource usage, network latency and other environment-related characteristics in the data.

[0070] If the environment class operation vector sequence shows certain similarities with a certain logic class historical operation trajectory data in terms of environmental characteristics, the server can consider the logic class historical association attributes corresponding to this historical operation trajectory data. Although these attributes mainly focus on logical errors, they may also contain some information related to environmental factors.

[0071] Next, the server compares the environmental characteristics of the environment-related operation vector sequence with the environmental historical operation trajectory data. This comparison will pay more attention to the environment-related features to determine the similarity between the current data and the historical environmental data.

[0072] Based on the environmental feature comparison results and environmental historical association attributes, the server can generate a cross-field matching vector of environmental related running trajectory data, which also contains the similarity information between the current data and the historical data in environmental features, as well as the specific causes and related information of the environmental errors.

[0073] Through the above steps, the server can generate cross-field matching vectors corresponding to the logic-related operation trajectory data and the environment-related operation trajectory data, respectively, providing strong support for subsequent abnormal root cause estimation and repair suggestions.

[0074] In a possible implementation, step S1225 includes: Step S1225-1, performing a logical feature comparison on the logic class operation vector sequence and the logic class historical operation trajectory data to generate a first logical feature comparison result.

[0075] In this embodiment, in this scenario, the server is performing cross-field matching analysis on the logic-related operation trajectory data appearing in the backend management system of the e-commerce platform. These logic-related operation trajectory data may involve anomalies in logic processes such as order processing and payment processes.

[0076] The server will first compare the logic class operation vector sequence with the logic class historical operation trajectory data stored in the cross-field matching knowledge base. The logic feature comparison mainly focuses on logic-related features such as business processes and event sequences.

[0077] For example, the current logic class operation vector sequence may reflect that an order has an exception during the payment process. The server can compare this vector sequence with all the logic class historical operation trajectory data in the knowledge base to find historical data that also involves payment process exceptions.

[0078] Through the comparison, the server generates a first logic feature comparison result, which may include multiple similarity scores, each score corresponding to a logic class historical operation trajectory data, indicating the similarity between the current logic class operation vector sequence and the historical data.

[0079] Step S1225-2, based on the logic class historical operation trajectory data corresponding to the first logic feature comparison result having a feature matching relationship, determine the first logic error category associated attribute from the logic class historical associated attributes.

[0080] Next, the server may analyze the first logical feature comparison result to find out the historical running track data with feature matching relationship. The feature matching relationship may set a threshold based on the similarity score, and only the historical data with a score exceeding this threshold will be considered to be matching.

[0081] For example, the server may find three historical operation trajectory data, and their similarity scores all exceed the set threshold. These three historical data all record anomalies in the payment process, such as order amount mismatch, payment interface failure, etc.

[0082] For each matching historical operation track data, the server can extract relevant information from the logic class history association attributes, which may include the specific cause of the logic error, error code, related system configuration, etc. The server integrates this information into a first logic error category association attribute set.

[0083] Step S1225-3, performing a logical feature comparison on the logic class operation vector sequence and the environment class historical operation trajectory data to generate a second logical feature comparison result.

[0084] Although the logic class operation vector sequence mainly focuses on the anomalies in the logic process, sometimes the logic error may be affected by environmental factors. Therefore, the server will also compare the logic class operation vector sequence with the environmental class historical operation trajectory data for logical features.

[0085] The purpose of this comparison is to check whether there are environmental factors related to the current logic error. For example, the server can look for historical data where payment process anomalies also occurred in the past and system resources were tight or network latency was high.

[0086] Through the comparison, the server generates a second logical feature comparison result, which also includes multiple similarity scores, each score corresponding to an environmental historical operation trajectory data.

[0087] Step S1225-4, based on the environment class historical operation trajectory data corresponding to the second logical feature comparison result having a feature matching relationship, determine the first environment error category associated attribute from the environment class historical associated attributes.

[0088] Similar to step S1225-1, the server can analyze the second logical feature comparison result to find out the environmental historical operation trajectory data with a feature matching relationship. These historical data may record environmental factors related to the current logical error, such as insufficient system resources, network delays, etc.

[0089] For each matching environment class historical running track data, the server can extract relevant information from the environment class historical associated attributes, which may include environment status, resource occupancy, network delay indicators, etc. The server integrates this information into a first environment error category associated attribute set.

[0090] Step S1225 - 5 : generating a cross-field matching vector of the logic-related operation trajectory data based on the first logic error category association attribute and the first environment error category association attribute.

[0091] Finally, the server generates a cross-field matching vector of logic-related operation trajectory data based on the first logic error category association attribute and the first environmental error category association attribute. The vector not only includes information directly related to the current logic error (such as error code, system configuration, etc.), but may also include environmental factor information related to the logic error (such as system resource status, network delay, etc.).

[0092] This cross-field matching vector will serve as an important reference for subsequent abnormal root cause estimation and repair suggestions, helping the server to more accurately understand the characteristics and causes of the current logical error.

[0093] In a possible implementation, step S1226 includes: Step S1226 - 1 , performing an environmental feature comparison on the environmental class operation vector sequence and the logical class historical operation trajectory data to generate a first environmental feature comparison result.

[0094] In this embodiment, in this scenario, the server is performing cross-field matching analysis on the environment-related operation trajectory data appearing in the backend management system of the e-commerce platform. These environment-related operation trajectory data may involve anomalies caused by environmental factors such as system resource usage and network delay.

[0095] Although the environment class operation vector sequence mainly focuses on environmental factors, sometimes environmental factors may be associated with logical errors. Therefore, the server first compares the environment class operation vector sequence with the logic class historical operation trajectory data stored in the cross-field matching knowledge base for environmental features.

[0096] Environmental feature comparison focuses on environmental features such as system resource usage and network latency. For example, the current environmental class operation vector sequence may reflect that the system CPU usage is too high. The server can compare this vector sequence with all the logic class historical operation trajectory data in the knowledge base to find records with similar environmental factors when logic errors occurred in history.

[0097] Through the comparison, the server generates a first environmental feature comparison result, which may include multiple similarity scores, each score corresponding to a logical class historical operation trajectory data, indicating the similarity between the current environmental class operation vector sequence and the historical data in environmental features.

[0098] Step S1226-2: Based on the logic class historical operation trajectory data corresponding to the first environmental feature comparison result having a feature matching relationship, determine the second logic error category associated attribute from the logic class historical associated attributes.

[0099] Next, the server may analyze the first environmental feature comparison result to find out logic-related historical operation trajectory data having a feature matching relationship, and these logic-related historical operation trajectory data may record logic errors that occurred under similar environmental conditions.

[0100] For each matching logic class historical operation trajectory data, the server can extract relevant information from the logic class historical associated attributes, which may include the specific cause of the logic error, the error code, and the environmental status at the time, etc. The server integrates this information into a second logic error category associated attribute set.

[0101] Step S1226 - 3 , performing an environmental feature comparison on the environment-type operation vector sequence and the environment-type historical operation trajectory data to generate a second environmental feature comparison result.

[0102] The server then compares the environmental characteristics of the environmental operation vector sequence with the environmental historical operation trajectory data stored in the cross-field matching knowledge base. This comparison mainly focuses on the similarities between environmental factors.

[0103] For example, the current environment class operation vector sequence may reflect a high network delay. The server can compare this environment class operation vector sequence with all environment class historical operation trajectory data to find historical data that also records abnormal network delays.

[0104] Through the comparison, the server generates a second environmental feature comparison result, which also includes multiple similarity scores, each score corresponding to an environmental class historical operation trajectory data.

[0105] Step S1226-4: based on the environment class historical operation trajectory data corresponding to the second environment feature comparison result having a feature matching relationship, determine the second environment error category associated attribute from the environment class historical associated attributes.

[0106] The server may analyze the second environment feature comparison result to find out environment-related historical operation trajectory data having a feature matching relationship, and these environment-related historical operation trajectory data may record error conditions that are the same or similar to the current environment abnormality.

[0107] For each matching environment class historical operation trajectory data, the server can extract relevant information from the environment class historical associated attributes, which may include error type, error description, system resource status at that time, network status, etc. The server integrates this information into a second environment error category associated attribute set.

[0108] Step S1226-5: generating a cross-field matching vector of the environment-related running trajectory data based on the second logic error category association attribute and the second environment error category association attribute.

[0109] Finally, the server generates a cross-field matching vector of the environment-related operation trajectory data based on the second logical error category association attribute and the second environmental error category association attribute. The vector not only contains information directly related to the current environmental anomaly (such as error type, error description, etc.), but may also contain logical error information related to the environmental anomaly (such as logical error code, related system configuration, etc.).

[0110] This cross-field matching vector will serve as an important reference for subsequent abnormal root cause estimation and repair suggestions, helping the server to more comprehensively understand the characteristics and possible causes of the current environmental abnormalities. By integrating information on logical errors and environmental factors, the server can provide more accurate and targeted abnormal handling suggestions.

[0111] In a possible implementation, the step of performing graph convolution processing respectively according to the logic-related running trajectory data and the environment-related running trajectory data to generate a logic-type running vector sequence and an environment-type running vector sequence is implemented by a graph convolution processing model, and the training step of the graph convolution processing model includes: Step S101 : obtaining a training data combination, wherein the training data combination includes sample logic-related running trajectory data and sample environment-related running trajectory data for the sample logic-related running trajectory data.

[0112] Step S102, using the initialized basic graph convolution processing model to perform graph convolution processing on the sample logic-related operation trajectory data and the sample environment-related operation trajectory data respectively, to generate a sample logic class operation vector sequence of the sample logic-related operation trajectory data and a training data combination environment class operation vector sequence of the sample environment-related operation trajectory data.

[0113] Step S103, performing iterative parameter optimization on the basic graph convolution processing model according to the sample logic class running vector sequence and the training data combination environment class running vector sequence to generate the graph convolution processing model.

[0114] In this scenario, the server is building and training a graph convolution processing model, which will be used to convert the application's logic-related running trajectory data and environment-related running trajectory data into vector representations for subsequent analysis and comparison.

[0115] First, the server needs to obtain a set of training data to train the graph convolution processing model. This set of training data is a combination, including sample logic-related running trajectory data and the corresponding sample environment-related running trajectory data.

[0116] These sample data are usually derived from historical application operation data, and may include a variety of abnormal situations and normal operation conditions. For example, in the backend management system of an e-commerce platform, sample logic-related operation trajectory data may include the execution of business processes such as user order placement, payment, and order cancellation, while sample environment-related operation trajectory data may record environmental information such as system resource usage and network latency.

[0117] The server can retrieve this sample data from a data warehouse or log system and organize it into a format suitable for model training.

[0118] Next, the server uses an initialized basic graph convolution processing model to perform graph convolution processing on the sample data. Graph convolution processing is a machine learning technology used to process graph structured data, which can convert complex running data into vector representation.

[0119] For the sample logic-related running trajectory data, the server inputs it into the basic graph convolution processing model. The model extracts key features based on the graph structure (nodes and edges) of the data and generates a logic-class running vector sequence, which will be used as the feature representation of the sample logic-related running trajectory data.

[0120] Similarly, for the sample environment-related running trajectory data, the server will also input it into the model to generate an environment-class running vector sequence, which will serve as the feature representation of the sample environment-related running trajectory data.

[0121] After generating vector representations of the sample data, the server needs to use these representations to optimize the parameters of the graph convolution processing model. The goal of the optimization is to enable the model to more accurately extract key features from the running trajectory data and generate representative vector representations.

[0122] The server usually uses an iterative optimization method, such as the gradient descent algorithm. In each iteration, the server can calculate the loss function value (i.e., prediction error) of the model on the training data, and then update the model parameters according to the gradient of the loss function. This process is repeated many times until the performance of the model on the validation set reaches the preset standard or the number of iterations reaches the upper limit.

[0123] During the iterative optimization process, the server will also monitor the training status of the model, such as the changing trend of the loss function value, the convergence speed of the model, etc. If the model is overfitting or underfitting, the server can take corresponding measures to adjust the complexity of the model or add regularization terms.

[0124] After multiple rounds of iterative optimization, the server will eventually generate a trained graph convolution processing model that can accurately extract key features from logic-related running trajectory data and environment-related running trajectory data, and generate representative vector representations. These vector representations will serve as the basis for subsequent abnormal root cause estimation and repair suggestions.

[0125] The server can save the trained model to the model warehouse so that it can be loaded and used when needed. At the same time, the server will also record information such as the model training process and parameter configuration to facilitate subsequent model maintenance and updates.

[0126] In a possible implementation, step S102 includes: Step S1021: When the sample logic-related operation trajectory data included in the training data combination belongs to application operation data, a logic class operation vector sequence is extracted for at least one application operation fragment in the application operation data using the initialized basic graph convolution processing model, and a sample logic class operation vector sequence of the application operation data is generated based on the logic class operation vector sequence of the at least one application operation fragment.

[0127] Step S1022: extracting an environment-related operation vector sequence from the sample environment-related operation trajectory data included in the training data combination through the initialized basic graph convolution processing model, and generating an environment-related operation vector sequence for the training data combination of the sample environment-related operation trajectory data.

[0128] In this embodiment, in this scenario, the server is using the initialized basic graph convolution processing model to perform graph convolution processing on the sample logic-related running trajectory data and the sample environment-related running trajectory data in the training data combination to generate a corresponding vector sequence. The following is a detailed step description: The server first prepares a set of training data combinations, which include sample logic-related operation trajectory data and corresponding sample environment-related operation trajectory data. These data are derived from the historical operation data of the e-commerce platform backend management system and include a variety of different business logic and environmental factor combinations.

[0129] The server starts processing the sample logic-related operation trajectory data in the training data combination. These data usually describe the business logic processes involved in the operation of the application, such as user ordering, payment, inventory update, etc.

[0130] The operation of an application is a continuous process, and the server can divide these logically related operation track data into multiple application operation segments. Each segment represents a relatively independent business logic unit, such as a complete order processing process.

[0131] Next, the server uses the initialized basic graph convolution processing model to extract the logic class operation vector sequence for each application operation fragment. The basic graph convolution processing model analyzes the graph structure data (nodes and edges) in the fragment, extracts key business logic features, and encodes these features into a logic class operation vector, which can represent the main features of the application operation fragment at the logical level.

[0132] After completing the extraction of the logical class operation vectors of all fragments, the server can combine these vectors in a certain order to generate a sample logical class operation vector sequence of the entire application operation data. The sample logical class operation vector sequence will serve as the overall feature representation of the application at the logical level.

[0133] Similar to the processing of logic-related running trajectory data, the server then processes the sample environment-related running trajectory data in the training data combination. These sample environment-related running trajectory data describe the external environment status of the application during operation, such as system resource usage, network delay, etc.

[0134] Since the environment state is usually a continuously changing process, the server does not need to divide it into multiple segments. Instead, it directly uses the initialized basic graph convolution processing model to extract the environment class operation vector sequence from the entire environment-related operation trajectory data.

[0135] The basic graph convolution processing model analyzes the graph structure features in the environment data (such as the dependencies between different resources), extracts key environment state features, and encodes these features into an environment class operation vector, which can represent the main features of the environment state of the application during operation.

[0136] After the above processing, the server has generated a sample logic class operation vector sequence of the sample logic-related operation trajectory data and a training data combination environment class operation vector sequence of the sample environment-related operation trajectory data. These two vector sequences will be used as feature representations of the training data combination for subsequent iterative optimization of the graph convolution processing model.

[0137] The server can save these two vector sequences to the corresponding data storage for use in the model training process. At the same time, it will also record the relevant processing parameters and configuration information to facilitate subsequent model maintenance and debugging.

[0138] Through this process, the server can effectively convert complex application running data into easy-to-process vector representations, providing strong support for the subsequent training and optimization of graph convolution processing models.

[0139] In a possible implementation manner, the multiple abnormal knowledge categories include a logic error category and an environment error category. The category running track data of the multiple abnormal knowledge categories include logic-related running track data and environment-related running track data.

[0140] Step S140 may include: Step S141 : Integrate the environment-related running trajectory data and the at least one key graph path to generate integrated trajectory data.

[0141] In this scenario, the server is performing root cause analysis on anomalies that occur in the backend management system of an e-commerce platform. The anomaly knowledge categories are divided into logical error categories and environmental error categories, and the category operation trajectory data includes logic-related operation trajectory data and environment-related operation trajectory data. The server has determined at least one key graph path through the previous steps and is ready to perform anomaly root cause estimation.

[0142] The server first integrates the environment-related operation trajectory data with the previously determined key graph path to generate integrated trajectory data. The integration process aims to combine the key path at the logical level with the operation status at the environmental level to form a more comprehensive abnormality analysis perspective.

[0143] For example, the key graph path may reveal that a problem occurred at a specific link in the order processing process, while the environment-related operation trajectory data may show that the system CPU utilization was too high or the network latency was large at that time. The server can integrate these two parts of information to form an integrated trajectory data that includes the logical path and the environmental status.

[0144] Step S142: using a trajectory analysis model to perform pattern recognition on a logic-type operation vector sequence corresponding to the logic-related operation trajectory data, and generating a pattern recognition vector map.

[0145] Next, the server uses the trained trajectory analysis model to perform pattern recognition on the logic-related operation trajectory data corresponding to the logic-related operation trajectory sequence. The trajectory analysis model can identify abnormal patterns or regularities in the logic operation trajectory, providing important clues for subsequent abnormal root cause estimation.

[0146] In this example, the trajectory analysis model analyzes the logic class run vector sequence to identify possible abnormal patterns in the order processing process, such as repeated order submissions, payment amount mismatches, etc. The model encodes these abnormal patterns into a pattern recognition vector map, which intuitively shows the abnormal characteristics at the logic level.

[0147] Step S143, using an abnormal root cause decision network to estimate the abnormal root cause according to the integrated trajectory data and the pattern recognition vector map, and generating abnormal root cause data for the application running data.

[0148] After obtaining the integrated trajectory data and pattern recognition vector diagram, the server uses the abnormal root cause decision network to estimate the abnormal root cause. The abnormal root cause decision network is a complex machine learning model that can comprehensively consider the abnormal pattern at the logical level and the operating status at the environmental level to estimate the root cause of the abnormality.

[0149] In this example, the abnormal root cause decision network will receive the integrated trajectory data and pattern recognition vector map as input, and estimate the root cause of the abnormal order processing process through internal calculation and analysis. Possible causes include but are not limited to: performance bottlenecks caused by insufficient system resources, data transmission problems caused by network delays, business process design defects, etc.

[0150] The server will eventually generate root cause data for the application's running data. This data describes in detail the specific cause of the exception and its scope of impact. This data will serve as an important basis for subsequent repair suggestions, helping developers quickly locate and solve problems.

[0151] Through the above steps, the server can comprehensively utilize the critical paths at the logical level and the operating status information at the environmental level to conduct in-depth abnormal root cause analysis on the application operation data, providing strong support for the rapid resolution of problems.

[0152] For example, in one possible implementation, the training steps of the trajectory analysis model and the abnormal root cause decision network include: Step A110, performing parameter learning based on a first template logic class operation vector sequence of the first template application operation data and template representation information of the first template application operation data to generate a temporary trajectory analysis model. The first template logic class operation vector sequence is generated by performing graph convolution processing on the logic-related operation trajectory data of the first template application operation data.

[0153] In this scenario, the server is training the trajectory analysis model and the anomaly root cause decision network, which will be used for subsequent application anomaly root cause analysis. The following is a detailed description of the training steps: The server first obtains a set of first template application running data, which are pre-collected and representative application running data. This set of first template application running data contains logic-related running trajectory data, which describes the running trajectory of the application at the logical level.

[0154] The server uses the trained graph convolution processing model to perform graph convolution processing on the logic-related operation trajectory data of the first template application operation data. This process converts complex operation data into vector representation and generates a first template logic class operation vector sequence, which can capture the key features in the logic operation trajectory.

[0155] The server performs parameter learning based on the first template logic class running vector sequence and the corresponding template representation information (such as the metadata and configuration information of the application). Parameter learning is an iterative optimization process. The server continuously adjusts the parameters of the model so that the model can accurately perform pattern recognition on the logic class running vector sequence. After multiple rounds of iterations, the server generates a temporary trajectory analysis model.

[0156] Step A120, obtaining a second template logic class operation vector sequence, template environment related operation trajectory data and template abnormal root cause data of the second template application operation data. The second template logic class operation vector sequence is generated by performing graph convolution processing on the logic related operation trajectory data of the second template application operation data.

[0157] The server then obtains a set of second template application running data, which also includes logic-related running track data and environment-related running track data, as well as corresponding template abnormality root cause data. The template abnormality root cause data is the known root cause that causes the application abnormality.

[0158] Similarly, the server performs graph convolution processing on the logic-related running trajectory data of the second template application running data to generate a second template logic class running vector sequence.

[0159] Step A130: Use the temporary trajectory analysis model to perform pattern recognition on the second template logic class operation vector sequence to generate a second pattern recognition vector map.

[0160] The server inputs the second template logic class operation vector sequence into the temporary trajectory analysis model, and the model performs pattern recognition on the vector sequence to generate a second pattern recognition vector diagram, which captures abnormal patterns or regularities in the logic operation trajectory.

[0161] Step A140, using the initialized basic abnormal root cause decision network to estimate the abnormal root cause according to the template environment related running trajectory data and the second pattern recognition vector diagram, to generate an abnormal root cause estimation result for the second template application running data.

[0162] The server first initializes a basic anomaly root cause decision network, which is a complex machine learning model that can comprehensively consider the abnormal patterns at the logical level and the operating status at the environmental level to estimate the root cause of the anomaly.

[0163] The server takes the template environment related running trajectory data and the second pattern recognition vector diagram as input and passes them to the basic abnormal root cause decision network. The basic abnormal root cause decision network generates an abnormal root cause estimation result for the second template application running data through internal calculation and analysis.

[0164] Step A150, performing iterative joint parameter optimization on the temporary trajectory analysis model and the basic abnormal root cause decision network based on the abnormal root cause estimation result and the template abnormal root cause data, to generate a trajectory analysis model and an abnormal root cause decision network with completed parameter optimization.

[0165] The server compares the abnormal root cause estimation result with the known template abnormal root cause data and calculates the difference or error between the two.

[0166] Based on the comparison results, the server iteratively optimizes the parameters of the temporary trajectory analysis model and the basic abnormal root cause decision network. This process is a process of back propagation and gradient descent. The server continuously adjusts the parameters of the two models so that the models can more accurately identify abnormal patterns in the logical operation trajectory and estimate the root cause of the abnormality. After multiple rounds of iterations, the server generates a trajectory analysis model and an abnormal root cause decision network with optimized parameters.

[0167] Through the above steps, the server has trained a trajectory analysis model and an abnormal root cause decision network that can accurately identify abnormal patterns in the logic operation trajectory and estimate the root cause of the abnormality. These two models will serve as important tools for subsequent application abnormality root cause analysis to help developers quickly locate and solve problems.

[0168] For example, in a possible implementation, step A110 may include: Step A111: Acquire a first template logic class running vector sequence of a first template application running data and template representation information for the first template application running data.

[0169] In this scenario, the server is training a temporary trajectory analysis model based on the first template application running data, and the temporary trajectory analysis model will be used for subsequent logical running trajectory pattern recognition.

[0170] The first template application running data is predefined and representative application running data used for model training. This set of first template application running data contains logic-related running trajectory data, which describes the running trajectory of the application at the logical level.

[0171] The server uses the trained graph convolution processing model to perform graph convolution processing on the logic-related operation trajectory data of the first template application operation data. This process converts complex operation data into vector representation and generates a first template logic class operation vector sequence. The first template logic class operation vector sequence can capture the key features in the logic operation trajectory.

[0172] In addition to the logic class running vector sequence, the server also obtains template representation information for the first template application running data. The template representation information may include application metadata, configuration information, business logic description, etc., providing the model with contextual knowledge about the application.

[0173] Step A112: Use the initialized basic trajectory analysis model to perform pattern recognition on the first template logic class operation vector sequence to generate a first pattern recognition vector map.

[0174] The server loads an initialized basic trajectory analysis model, which is a machine learning model that already has basic functions but has not been fully trained for specific tasks. The server inputs the first template logic class operation vector sequence into the basic trajectory analysis model, and the basic trajectory analysis model performs pattern recognition on the first template logic class operation vector sequence, trying to identify the potential patterns or rules in the logic operation trajectory. The result of pattern recognition is a first pattern recognition vector diagram, which intuitively shows the key features or patterns in the logic operation trajectory.

[0175] Step A113: Use the initialized basic abnormal root cause decision network to embed the first pattern recognition vector map to generate embedded representation information for the first template application running data.

[0176] The server then loads an initialized basic abnormal root cause decision network, which is also a machine learning model that already has basic functions but has not been fully trained for specific tasks. The server inputs the first pattern recognition vector graph into the basic abnormal root cause decision network, and the basic abnormal root cause decision network embeds the first pattern recognition vector graph. Embedding representation is a process of mapping high-dimensional data to low-dimensional space. It can capture the key information in the vector graph and generate an embedded representation information for the first template application running data. The embedded representation information contains the key features of the logical running trajectory and is presented in a form that is easier to process and analyze.

[0177] Step A114: Iteratively optimize the parameters of the basic trajectory analysis model according to the embedded representation information and the template representation information to generate a temporary trajectory analysis model.

[0178] The server compares the generated embedding representation information with the known template representation information. The template representation information provides contextual knowledge about the running data of the first template application, while the embedding representation information is the result of the model capturing and representing the key features of the logical running trajectory. The server evaluates the performance of the trajectory analysis model by comparing the differences or similarities between the two.

[0179] Based on the comparison results, the server iteratively optimizes the parameters of the basic trajectory analysis model, which is a process of back propagation and gradient descent. The server first calculates the difference or loss between the embedded representation information and the template representation information, and then adjusts the parameters of the model according to the gradient of the loss function. Through multiple iterative optimizations, the server gradually adjusts the parameters of the model so that the model can more accurately identify the key features in the logical operation trajectory and generate embedded representation information that is more consistent with the template representation information. After multiple rounds of iterations, the server generates a temporary trajectory analysis model, which already has a certain ability to identify logical operation trajectory patterns and will continue to be optimized in subsequent training processes.

[0180] For example, in a possible implementation, step S130 may include: Step S131 , summarizing the cross-field matching vectors corresponding to the category running trajectory data of the plurality of abnormal knowledge categories, and generating a summary matching vector result.

[0181] Step S132: determining at least one key graph path for the application program running data based on the summary matching vector result.

[0182] In this embodiment, in this scenario, the server has completed the cross-field matching of the application running data and generated the corresponding cross-field matching vector for the category running trajectory data of each abnormal knowledge category. Next, the server will determine the key graph path in the application running data based on these cross-field matching vectors.

[0183] The server first summarizes the cross-field matching vectors corresponding to the category operation trajectory data of multiple abnormal knowledge categories. These cross-field matching vectors contain the similarity and difference information between the current operation trajectory data and the historical operation trajectory data in logical features and environmental features, as well as key attributes related to the abnormality.

[0184] For example, a server may have multiple cross-field matching vectors under the logical error category and the environmental error category. The vectors under the logical error category may reflect logical anomalies in the order processing process, such as duplicate order submissions or mismatched payment amounts, while the vectors under the environmental error category may reveal environmental problems such as excessive system resource usage or network latency.

[0185] The server groups these vectors according to the abnormal knowledge category and summarizes the vectors in the same category. The summarization process may involve weighted averaging or voting on each dimension of the vector to obtain a summary matching vector result that comprehensively reflects the abnormal characteristics of the category.

[0186] After obtaining the summary matching vector results, the server begins to determine the key graph paths in the application operation data based on these results. Key graph paths are those paths that are highly correlated with abnormal features and have a significant impact in the application operation graph structure.

[0187] The server first analyzes and summarizes the key information in the matching vector results, such as anomaly characteristics, impact range, and occurrence frequency. This information can help the server identify which paths are most relevant to the current anomaly.

[0188] Next, the server searches for paths that match the key information in the graph based on the application's operation graph structure. The operation graph structure is a graph consisting of nodes and edges, where nodes represent components or functions in the application and edges represent interactions or dependencies between components.

[0189] The server traverses the operation graph structure to find graph paths that contain abnormal feature nodes, connect key components, or cross important interaction paths. These paths may directly involve the logical process or environmental factors where the exception occurs, and have a significant impact on the operation status of the application.

[0190] Finally, the server selects at least one key graph path from the searched graph paths based on factors such as the importance of the path, the scope of impact, and the significance of the abnormal characteristics. These key graph paths will serve as an important basis for subsequent abnormal root cause estimation and repair suggestions.

[0191] Through this process, the server can comprehensively consider the category operation trajectory data and cross-field matching vector information of multiple abnormal knowledge categories, and determine the key graph path that is most relevant to the current abnormality and has a significant impact in the application operation graph structure. This will help the server locate the cause of the abnormality more accurately and provide targeted repair suggestions.

[0192] Figure 2 The hardware structure of the personalized content recommendation system 100 implemented by artificial intelligence for implementing the personalized content recommendation method implemented by artificial intelligence provided in the embodiment of the present application is shown as follows: Figure 2 As shown, the personalized content recommendation system 100 implemented using artificial intelligence may include a processor 110 , a machine-readable storage medium 120 , a bus 130 , and a communication unit 140 .

[0193] In one possible design, the personalized content recommendation system 100 implemented using artificial intelligence can be a single server or a server group. The server group can be centralized or distributed (for example, the personalized content recommendation system 100 implemented using artificial intelligence can be a distributed system). In some embodiments, the personalized content recommendation system 100 implemented using artificial intelligence can be local or remote. For example, the personalized content recommendation system 100 implemented using artificial intelligence can access information and / or data stored in a machine-readable storage medium 120 via a network. For another example, the personalized content recommendation system 100 implemented using artificial intelligence can be directly connected to the machine-readable storage medium 120 to access the stored information and / or data. In some embodiments, the personalized content recommendation system 100 implemented using artificial intelligence can be implemented on a personalized content recommendation system implemented using artificial intelligence. As an example only, the personalized content recommendation system implemented using artificial intelligence can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc. or any combination thereof.

[0194] The machine-readable storage medium 120 may store data and / or instructions. In some embodiments, the machine-readable storage medium 120 may store data obtained from an external terminal. In some embodiments, the machine-readable storage medium 120 may store data and / or instructions that the personalized content recommendation system 100 implemented using artificial intelligence uses to execute or use to complete the exemplary methods described in this application.

[0195] In the specific implementation process, one or more processors 110 execute computer executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the personalized content recommendation method implemented by artificial intelligence in the above method embodiment. The processor 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the sending and receiving actions of the communication unit 140.

[0196] The specific implementation process of the processor 110 can refer to the various method embodiments executed by the personalized content recommendation system 100 implemented by artificial intelligence mentioned above. The implementation principles and technical effects are similar and will not be repeated in this embodiment.

[0197] In addition, the embodiment of the present application further provides a readable storage medium, in which computer executable instructions are preset, and when the processor executes the computer executable instructions, the personalized content recommendation method implemented by artificial intelligence as described above is implemented. It should be noted that in order to simplify the description disclosed in this application and thus help understand one or more embodiments of the invention, in the above description of the embodiments of the present application, multiple features are sometimes merged into one embodiment, drawing or description thereof.

Claims

1. A personalized content recommendation method using artificial intelligence, characterized in that: The method comprises: Determine the application running data to be mined, and perform graph convolution processing on the application running data to generate category running trajectory data of multiple abnormal knowledge categories; According to the category running trajectory data of the multiple abnormal knowledge categories, cross-field matching is performed according to the category knowledge fields of the multiple abnormal knowledge categories respectively, to generate cross-field matching vectors corresponding to the category running trajectory data of the multiple abnormal knowledge categories respectively; Determine at least one key graph path for the application program running data based on the cross-field matching vectors respectively corresponding to the category running trajectory data of the plurality of abnormal knowledge categories; Estimating abnormal root causes according to the at least one key graph path and the category operation trajectory data of the plurality of abnormal knowledge categories, and generating abnormal root cause data for the application operation data; Personalized repair suggestion content is recommended based on the abnormal root cause data of the application running data.

2. The personalized content recommendation method using artificial intelligence according to claim 1, characterized in that: According to the category running trajectory data of the multiple abnormal knowledge categories, cross-field matching is performed according to the category knowledge fields of the multiple abnormal knowledge categories respectively, and cross-field matching vectors corresponding to the category running trajectory data of the multiple abnormal knowledge categories are generated, including: Determine a cross-field matching knowledge base; the cross-field matching knowledge base includes historical operation track data of historical application operation data under the multiple abnormal knowledge categories and historical association attributes corresponding to each of the historical operation track data; Based on the comparison results between the category operation trajectory data of the multiple abnormal knowledge categories and the historical operation trajectory data of the historical application operation data under the multiple abnormal knowledge categories, cross-field matching vectors corresponding to the category operation trajectory data of the multiple abnormal knowledge categories are matched and generated from the cross-field matching knowledge base.

3. The personalized content recommendation method using artificial intelligence according to claim 2, characterized in that: The cross-field matching vectors corresponding to the category running trajectory data of the multiple abnormal knowledge categories are generated from the cross-field matching knowledge base based on the comparison results between the category running trajectory data of the multiple abnormal knowledge categories and the historical running trajectory data of the historical application running data under the multiple abnormal knowledge categories, including: Comparing each category of the category running trajectory data of the multiple abnormal knowledge categories with the historical running trajectory data of the historical application running data under the multiple abnormal knowledge categories, respectively, to generate a comparison result of each category of the running trajectory data; From the comparison results of each category of running trajectory data, determine the key comparison results that each category of running trajectory data matches; Based on the historical running trajectory data corresponding to the key comparison results respectively corresponding to each category of running trajectory data, the cross-field matching vectors respectively corresponding to each category of running trajectory data are matched from the cross-field matching knowledge base.

4. The personalized content recommendation method using artificial intelligence according to claim 2, characterized in that: The multiple abnormal knowledge categories include a logic error category and an environment error category; the category operation track data of the multiple abnormal knowledge categories include logic-related operation track data and environment-related operation track data of the application operation data; the cross-field matching knowledge base includes logic-class historical operation track data of historical application operation data under the logic error category and logic-class historical associated attributes corresponding to the logic-class historical operation track data, as well as environment-class historical operation track data of the historical application operation data under the environment error category and environment-class historical associated attributes corresponding to the environment-class historical operation track data; The cross-field matching vectors corresponding to the category running trajectory data of the multiple abnormal knowledge categories are generated from the cross-field matching knowledge base based on the comparison results between the category running trajectory data of the multiple abnormal knowledge categories and the historical running trajectory data of the historical application running data under the multiple abnormal knowledge categories, including: Performing graph convolution processing on the logic-related running trajectory data and the environment-related running trajectory data to generate a logic-related running vector sequence and an environment-related running vector sequence; Performing logic feature comparison on the logic class operation vector sequence, the logic class historical operation trajectory data and the environment class historical operation trajectory data respectively, and generating a cross-field matching vector of the logic-related operation trajectory data from the logic class historical association attribute and the environment class historical association attribute based on the logic feature comparison result; The environment-related operation vector sequence is compared with the logic-related historical operation trajectory data and the environment-related historical operation trajectory data respectively for environmental features, and a cross-field matching vector of the environment-related operation trajectory data is generated from the logic-related historical association attributes and the environment-related historical association attributes based on the environmental feature comparison result.

5. The personalized content recommendation method using artificial intelligence according to claim 4, characterized in that: The step of performing logical feature comparison on the logic class operation vector sequence, the logic class historical operation trajectory data and the environment class historical operation trajectory data, and generating a cross-field matching vector of the logic-related operation trajectory data from the logic class historical association attributes and the environment class historical association attributes based on the logical feature comparison result includes: Performing a logical feature comparison on the logic class operation vector sequence and the logic class historical operation trajectory data to generate a first logical feature comparison result; Based on the logic class historical operation trajectory data corresponding to the first logic feature comparison result having a feature matching relationship, determining a first logic error category association attribute from the logic class historical association attributes; Performing a logical feature comparison on the logic class operation vector sequence and the environment class historical operation trajectory data to generate a second logical feature comparison result; Based on the environment class historical operation trajectory data corresponding to the second logical feature comparison result having a feature matching relationship, determining the first environment error category association attribute from the environment class historical association attributes; A cross-field matching vector of the logic-related running trajectory data is generated based on the first logic error category association attribute and the first environment error category association attribute.

6. The personalized content recommendation method using artificial intelligence according to claim 4, characterized in that: The step of comparing the environment-related operation vector sequence with the logic-related historical operation trajectory data and the environment-related historical operation trajectory data respectively, and generating a cross-field matching vector of the environment-related operation trajectory data from the logic-related historical association attributes and the environment-related historical association attributes based on the environment-related feature comparison result, includes: Performing an environmental feature comparison on the environmental class operation vector sequence and the logical class historical operation trajectory data to generate a first environmental feature comparison result; Based on the logic class historical operation trajectory data corresponding to the first environmental feature comparison result having a feature matching relationship, determining a second logic error category association attribute from the logic class historical association attribute; Performing an environmental feature comparison on the environment class operation vector sequence and the environment class historical operation trajectory data to generate a second environmental feature comparison result; Based on the environment class historical operation trajectory data corresponding to the second environment feature comparison result having a feature matching relationship, determining the second environment error category association attribute from the environment class historical association attributes; A cross-field matching vector of the environment-related running trajectory data is generated based on the second logical error category association attribute and the second environmental error category association attribute.

7. The personalized content recommendation method using artificial intelligence according to claim 4, characterized in that: The step of performing graph convolution processing respectively according to the logic-related running trajectory data and the environment-related running trajectory data to generate a logic-type running vector sequence and an environment-type running vector sequence is implemented by a graph convolution processing model, and the training step of the graph convolution processing model includes: Acquire a training data combination, the training data combination comprising sample logic-related running trajectory data and sample environment-related running trajectory data for the sample logic-related running trajectory data; Using the initialized basic graph convolution processing model, graph convolution processing is performed on the sample logic-related running trajectory data and the sample environment-related running trajectory data respectively to generate a sample logic-class running vector sequence of the sample logic-related running trajectory data and a training data combination environment-class running vector sequence of the sample environment-related running trajectory data; According to the sample logic class running vector sequence and the training data combination environment class running vector sequence, the basic graph convolution processing model is iteratively optimized to generate the graph convolution processing model.

8. The personalized content recommendation method using artificial intelligence according to claim 7, characterized in that: The method of using the initialized basic graph convolution processing model to perform graph convolution processing on the sample logic-related running trajectory data and the sample environment-related running trajectory data respectively to generate a sample logic-class running vector sequence of the sample logic-related running trajectory data and a training data combination environment-class running vector sequence of the sample environment-related running trajectory data includes: When the sample logic-related running trajectory data included in the training data combination belongs to application running data, a logic-class running vector sequence is extracted for at least one application running fragment in the application running data using an initialized basic graph convolution processing model, and a sample logic-class running vector sequence of the application running data is generated based on the logic-class running vector sequence of the at least one application running fragment; An environment-type operation vector sequence is extracted for the sample environment-related operation trajectory data included in the training data combination through the initialized basic graph convolution processing model, so as to generate an environment-type operation vector sequence for the training data combination of the sample environment-related operation trajectory data.

9. The personalized content recommendation method using artificial intelligence according to claim 1, characterized in that: The multiple abnormal knowledge categories include a logic error category and an environment error category; the category running track data of the multiple abnormal knowledge categories include logic-related running track data and environment-related running track data; The step of estimating the abnormal root cause based on the at least one key graph path and the category operation trajectory data of the plurality of abnormal knowledge categories to generate abnormal root cause data for the application operation data includes: Integrating the environment-related running trajectory data and the at least one key graph path to generate integrated trajectory data; Using a trajectory analysis model to perform pattern recognition on a logic-type operation vector sequence corresponding to the logic-related operation trajectory data, and generate a pattern recognition vector diagram; An abnormal root cause decision network is used to estimate the abnormal root cause based on the integrated trajectory data and the pattern recognition vector map to generate abnormal root cause data for the application running data.

10. A personalized content recommendation system using artificial intelligence, characterized in that: The personalized content recommendation system implemented using artificial intelligence includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the personalized content recommendation method implemented using artificial intelligence as described in any one of claims 1 to 9 above.

Images