Data processing method and system applied to supply chain management platform

Abstract

The present application provides a data processing method and system for a supply chain management platform, which obtains crash feature data of a program crash event, performs cluster analysis on the crash feature data, generates crash feature clusters, and performs potential association configuration to generate potential association feature combinations and their corresponding potential association representations between combinations. Then, according to the potential association representations between combinations, the potential association feature combinations are aggregated to generate target aggregate feature data. Finally, based on the target aggregate feature data, the recurrence risk index of the crash event is predicted, and the risk level is determined based on the recurrence risk index and loaded into the risk management session. This can not only deeply analyze the linkage crash path and potential association of the crash event, but also accurately predict the recurrence risk of the crash event, providing data support for the stable operation and continuous improvement of the supply chain management platform.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a data processing method and system applied to a supply chain management platform. Background Art

[0002] During the operation of the supply chain management platform, program crash events are one of the inevitable problems. These program crash events may be caused by a variety of factors, such as insufficient system resources, code defects, network delays, etc., and often involve multiple different crash nodes. Traditional data processing methods usually only focus on the direct characteristics of crash events, such as crash time, crash nodes, etc., and lack analysis of the deep-level associations and causes behind the crash events. Therefore, for the supply chain management platform, how to efficiently process and analyze program crash events, accurately assess their recurrence risks, and take effective risk management measures has become an urgent problem to be solved. Summary of the invention

[0003] 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 data processing method applied to a supply chain management platform, the method comprising:

[0004] Obtain crash feature data corresponding to each of the X program crash events of the target supply chain management application, wherein the crash feature data represents linkage crash path data from a plurality of different crash nodes;

[0005] Traversing each of the program crash events, clustering the crash feature data, and generating Y crash feature clusters, where Y is an integer not less than 2;

[0006] Traversing each of the program crash events, performing potential association configuration on the Y crash feature clusters, generating Z potential association feature combinations and potential association representations between combinations corresponding to each of the potential association feature combinations, where Z is an integer not less than 1;

[0007] Traversing each of the program crash events, and according to the potential correlation representation between each of the combinations, aggregating Z of the potential correlation feature combinations to generate target aggregate feature data corresponding to the program crash event;

[0008] Predicting a recurrence risk index corresponding to each of the program crash events based on the target aggregation feature data corresponding to each of the program crash events;

[0009] The risk levels of the X program crash events are determined according to the recurrence risk index, and the X program crash events are loaded into the risk management session corresponding to the target supply chain management application according to the risk level.

[0010] In a possible implementation of the first aspect, traversing each of the program crash events, performing potential association configuration on the Y crash feature clusters, generating Z potential association feature combinations and potential association representations between combinations corresponding to each of the potential association feature combinations, includes:

[0011] Randomly determine two of the crash feature clusters from the Y crash feature clusters for pairing, to generate Z potential correlation feature combinations;

[0012] Perform potential correlation enhancement processing on each of the Z potential correlation feature combinations respectively, and generate a potential correlation representation between combinations corresponding to each of the potential correlation feature combinations.

[0013] In a possible implementation manner of the first aspect, the arbitrarily determining two of the crash feature clusters from the Y crash feature clusters for pairing to generate Z potential correlation feature combinations includes:

[0014] Encode each of the crash feature clusters to generate Y code vectors corresponding to the Y crash feature clusters;

[0015] Pair any two of the encoding vectors to generate Z potential association feature combinations.

[0016] In a possible implementation of the first aspect, performing potential correlation enhancement processing on each of the Z potential correlation feature combinations to generate a potential correlation representation between combinations corresponding to each of the potential correlation feature combinations includes:

[0017] Obtaining participation coefficients corresponding to two matching encoding vectors in each of the potential associated feature combinations;

[0018] Performing weighted calculation on the participation coefficient and the two matching encoding vectors in each potential association feature combination to generate a potential association representation between combinations corresponding to each potential association feature combination;

[0019] In a possible implementation of the first aspect, obtaining participation coefficients corresponding to two matching encoding vectors in each potential association feature combination includes:

[0020] Combining the two matching encoding vectors in each of the potential associated feature combinations to generate a target encoding vector;

[0021] Passing the target encoding vector through an attention network to generate the participation coefficient;

[0022] The step of performing weighted calculation on the participation coefficient and the two matching encoding vectors in each potential association feature combination to generate a potential association representation between combinations corresponding to each potential association feature combination includes:

[0023] The participation coefficient and the target coding vector are weightedly calculated to generate a potential correlation representation between combinations corresponding to each potential correlation feature combination.

[0024] In a possible implementation manner of the first aspect, traversing each of the program crash events, clustering the crash feature data, and generating Y crash feature clusters includes:

[0025] Acquire a path attribute corresponding to each linkage crash path data in the crash feature data, wherein the path attribute represents a relationship feature between a crash node element corresponding to each linkage crash path data and the program crash event;

[0026] The crash feature data is clustered according to the crash node elements and the path attributes to generate Y crash feature clusters.

[0027] In a possible implementation of the first aspect, traversing each of the program crash events, aggregating Z potential correlation feature combinations according to potential correlation representations between each of the combinations, and generating target aggregated feature data corresponding to the program crash event includes:

[0028] Each of the program crash events is traversed, and the potential correlation representations between the Z combinations corresponding to the Z potential correlation feature combinations are spliced ​​to generate target converged feature data corresponding to the program crash event.

[0029] In a possible implementation of the first aspect, predicting the recurrence risk index corresponding to each of the program crash events according to the target converged feature data corresponding to each of the program crash events includes:

[0030] The target aggregation feature data corresponding to each of the program crash events are mapped to recurrence risk features using a deep neural network of a recurrence risk prediction model to generate a target recurrence risk feature mapping vector corresponding to each of the program crash events;

[0031] The target recurrence risk feature mapping vector corresponding to each program crash event is mapped to a recurrence risk index corresponding to each program crash event using the softmax network of the recurrence risk prediction model.

[0032] In a possible implementation of the first aspect, the target aggregation feature data corresponding to each of the program crash events is mapped using a deep neural network of a recurrence risk prediction model to generate a target recurrence risk feature mapping vector corresponding to each of the program crash events, and before the training step of the recurrence risk prediction model includes:

[0033] Acquire template collapse feature data corresponding to the template collapse event, and cluster the template collapse feature data to generate Y template collapse feature clusters, wherein the template collapse feature data represents template linkage collapse path data from multiple different collapse nodes;

[0034] Performing potential association configuration on the Y template collapse feature clusters to generate Z template potential association feature combinations and a template potential association feature combination corresponding to each template potential association feature combination;

[0035] According to each of the template potential associated feature combinations, Z template potential associated feature combinations are aggregated to generate template aggregated feature data corresponding to the template collapse event;

[0036] Aggregating feature data of the template, performing recurrence risk feature mapping using a deep neural network of a recurrence risk prediction model, and generating a template recurrence risk feature mapping vector corresponding to the template collapse event;

[0037] The template recurrence risk feature mapping vector is used to generate a template recurrence risk index using the softmax network of the recurrence risk prediction model;

[0038] According to the template recurrence risk index and the recurrence risk supervised learning data, a training error parameter for the recurrence risk is calculated, and the recurrence risk prediction model is iteratively trained according to the training error parameter.

[0039] In a possible implementation manner of the first aspect, acquiring template collapse feature data corresponding to the template collapse event, and clustering the template collapse feature data to generate Y template collapse feature clusters includes:

[0040] Acquire a template path attribute corresponding to each template linkage crash path data in the template crash feature data, wherein the template path attribute represents a relationship feature between a crash node element corresponding to each linkage crash path data and the template crash event;

[0041] Clustering the template crash feature data according to the crash node element and the template path attribute to generate Y template crash feature clusters;

[0042] In a possible implementation of the first aspect, the clustering of the Y template collapse features is configured to be potentially associated to generate Z template potentially associated feature combinations and a template potentially associated feature combination corresponding to each of the template potentially associated feature combinations, including:

[0043] Encode each of the template collapse feature clusters to generate Y template encoding vectors corresponding to the Y template collapse feature clusters;

[0044] Pairing any two of the template encoding vectors to generate Z potential associated feature combinations of the templates;

[0045] Obtaining participation coefficients corresponding to two matching template encoding vectors in each combination of potential associated features of the template;

[0046] Performing weighted calculation on the participation coefficient and the two matching template encoding vectors in each template potential associated feature combination to generate a template potential associated feature combination corresponding to each template potential associated feature combination;

[0047] In a possible implementation of the first aspect, determining risk levels of the X program crash events according to the recurrence risk index, and loading the X program crash events into a risk management session corresponding to the target supply chain management application according to the risk levels, respectively, includes:

[0048] Performing risk level mapping on the X program crash events according to the recurrence risk index to generate risk levels of the X program crash events;

[0049] According to the risk levels of the X program crash events, the X program crash events are loaded into the risk management session in order of risk levels.

[0050] On the other hand, an embodiment of the present application also provides a data processing system applied to a supply chain management platform, 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.

[0051] Based on the above aspects, the embodiments of the present application significantly improve the management efficiency and accuracy of the supply chain management platform for program crash events. Specifically, first, the crash feature data of the program crash event in the target supply chain management application is obtained. These crash feature data describe in detail the linkage crash paths of multiple different crash nodes, providing a rich information basis for subsequent analysis. Next, the crash feature data is clustered to generate multiple crash feature clustering clusters to effectively identify different crash modes. Furthermore, the crash feature clustering clusters are potentially associated with the configuration, and potential associated feature combinations and potential associated representations between their combinations are generated, revealing the deep-level associations between program crash events.

[0052] By aggregating the potential correlation feature combinations, the target aggregation feature data of each program crash event is generated. These target aggregation feature data not only contain the direct features of the program crash event, but also incorporate the potential correlation and deep patterns behind the program crash event. Based on the target aggregation feature data, the recurrence risk index of each program crash event can be predicted, providing a quantitative basis for risk assessment. Finally, according to the risk level determined by the recurrence risk index, the crash event is loaded into the risk management session of the target supply chain management application, realizing efficient classification and management of crash events.

[0053] Therefore, by deeply analyzing and processing the crash feature data of program crash events, the ability to identify, analyze and assess risks of crash events has been significantly improved, providing strong support for the stable operation and continuous improvement of the supply chain management platform, which not only improves management efficiency but also reduces the potential losses caused by system crashes. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a schematic diagram of the execution flow of the data processing method applied to the supply chain management platform provided in the embodiment of the present application.

[0055] Figure 2 It is a schematic diagram of the hardware architecture of a data processing system applied to a supply chain management platform provided in an embodiment of the present application. DETAILED DESCRIPTION

[0056] The present application will be described in detail below with reference to the accompanying drawings. Figure 1 This is a flow chart of a data processing method applied to a supply chain management platform provided by an embodiment of the present application. The data processing method applied to the supply chain management platform is introduced in detail below.

[0057] Step S110 , obtaining crash feature data corresponding to each of the X program crash events of the target supply chain management application, wherein the crash feature data represents linkage crash path data from a plurality of different crash nodes.

[0058] In detail, the crash feature data refers to a series of crash-related detailed information captured and recorded when a program crash event occurs, describing the system status, nodes involved, exception type, crash path, etc. when the crash occurred, which is used for subsequent program debugging, problem location and cause analysis.

[0059] The linkage crash path data refers to a series of path information in which different nodes or services interact and influence each other in a program crash event, ultimately leading to the crash, revealing the complex relationships and potential causes behind the crash event.

[0060] That is, in this embodiment, the server first accesses the system log and monitoring data of the target supply chain management application. When a program crash event occurs in the target supply chain management application, the server can capture and record these program crash events in real time. For each program crash event, the server deeply analyzes and extracts feature data related to the crash. The crash feature data may include linkage information of multiple nodes (such as database services, application services, network services, etc.) that caused the crash, that is, which nodes had abnormalities before the crash event occurred, and the mutual influence and crash path between these abnormal nodes.

[0061] For example, the server may detect a memory overflow in an order processing service node, and at the same time find that the associated inventory management node also has a response timeout. The linkage anomaly of these two nodes eventually leads to the crash of the entire order processing process. The server records these linkage crash path data as part of the crash feature data.

[0062] Step S120 , traversing each of the program crash events, clustering the crash feature data, and generating Y crash feature clusters, where Y is an integer not less than 2.

[0063] In this embodiment, the server will then perform cluster analysis on the collected crash feature data. The purpose of clustering is to classify crash events with similar crash features into one category, so as to more accurately identify different types of crash patterns.

[0064] Specifically, the server first analyzes the path attributes in each crash feature data, such as the type of crash node, the type of anomaly, the time sequence of anomaly occurrence, etc. Then, clustering algorithms (such as K-means, DBSCAN, etc.) are used to cluster these features. After clustering, the server can obtain Y crash feature clusters, each of which represents a specific crash mode.

[0065] Step S130, traversing each of the program crash events, configuring the Y crash feature clusters for potential association, generating Z potential association feature combinations and potential association representations between combinations corresponding to each of the potential association feature combinations, where Z is an integer not less than 1.

[0066] The potential correlation feature combination is the result generated when analyzing the potential correlation between the crash feature clusters, representing the possible internal connection and mutual influence between different clusters. The potential correlation representation between combinations is a way to quantitatively describe the potential correlation feature combination, which can represent the correlation strength and nature between different clusters through numerical values, vectors or other complex structures, providing an important basis for subsequent crash prediction and problem analysis.

[0067] That is, in this embodiment, after obtaining the crash feature clustering clusters, the server can further explore the potential correlations between these crash feature clustering clusters. This is because in actual system crashes, it is often not caused by a single node or a single type of problem, but the result of multiple nodes or multiple problems that are interconnected and influence each other.

[0068] To find these potential associations, the server can encode each crash feature cluster and generate a corresponding encoding vector. Then, the server can try to pair different encoding vectors to explore possible associations between them. This pairing can be based on similarity or some rules or models.

[0069] After the pairing is completed, the server can obtain Z potential correlation feature combinations, each of which represents the possible correlation between two or more crash feature clusters. At the same time, the server will also calculate the potential correlation representation between each potential correlation feature combination. This potential correlation representation between combinations can be a value, a vector, or a more complex structure to quantitatively describe the strength and nature of this correlation.

[0070] Step S140, traverse each of the program crash events, and according to the potential correlation representation between each combination, aggregate the Z potential correlation feature combinations to generate target aggregate feature data corresponding to the program crash event, and predict the recurrence risk index corresponding to each of the program crash events according to the target aggregate feature data corresponding to each of the program crash events.

[0071] The recurrence risk index is an indicator that quantitatively assesses the possibility of a program crash event occurring again in the future. It is calculated based on target aggregation feature data through a pre-trained risk prediction model and is used to guide subsequent risk management and resource allocation.

[0072] In this embodiment, after determining the potential correlation feature combination, the server can start a deeper analysis of each program crash event. The server can traverse each program crash event and aggregate Z potential correlation feature combinations based on the potential correlation representation between the combinations calculated previously. This process can be understood as integrating the information scattered in different crash nodes and crash modes to form a more comprehensive and in-depth description of the crash event.

[0073] After the aggregation is completed, the server can obtain the target aggregation feature data corresponding to each program crash event. These target aggregation feature data not only include the direct features of the crash event (such as crash node, exception type, etc.), but also include the potential associations and deep patterns behind the crash event.

[0074] Next, the server can use these target converged feature data to predict the recurrence risk index, which is usually done through a pre-trained recurrence risk prediction model. The recurrence risk prediction model analyzes the dimensions and associations in the target converged feature data and then gives a numerical recurrence risk index. The higher the recurrence risk index, the greater the possibility that the program crash event will occur again in the future.

[0075] Step S150 , determining risk levels of the X program crash events according to the recurrence risk index, and loading the X program crash events into the risk management session corresponding to the target supply chain management application according to the risk levels.

[0076] In this embodiment, the risk management session is a user interface or functional module in the target supply chain management application for centrally displaying and processing risk events, allowing administrators or developers to view detailed information, risk levels, and processing status of crash events, and take corresponding risk response measures.

[0077] That is, in the end, the server can determine the risk level of each program crash event based on the predicted recurrence risk index. The risk level can be high, medium, or low, or it can be more subdivided into five or ten levels. The risk level classification standard is usually set according to actual business needs and system tolerance.

[0078] After determining the risk level, the server can feed this information back to the risk management system of the target supply chain management application. The risk management system will decide how to handle these program crash events based on these risk levels. For example, for high-risk events, an alarm can be triggered immediately and the administrator can be notified for emergency processing; for low-risk events, you can choose to record them in the background and wait for regular maintenance to handle them.

[0079] In this way, the server can not only help the target supply chain management application detect and record program crash events in real time, but also deeply analyze the inherent correlation and recurrence risks of these program crash events, providing strong data support for the stable operation and continuous improvement of the system.

[0080] Based on the above steps, the embodiment of the present application significantly improves the management efficiency and accuracy of the supply chain management platform for program crash events. Specifically, first, the crash feature data of the program crash event in the target supply chain management application is obtained. These crash feature data describe in detail the linkage crash paths of multiple different crash nodes, providing a rich information basis for subsequent analysis. Next, the crash feature data is clustered to generate multiple crash feature clustering clusters to effectively identify different crash modes. Furthermore, the crash feature clustering clusters are potentially associated with the configuration, and potential associated feature combinations and potential associated representations between their combinations are generated, revealing the deep-level associations between program crash events.

[0081] By aggregating the potential correlation feature combinations, the target aggregation feature data of each program crash event is generated. These target aggregation feature data not only contain the direct features of the program crash event, but also incorporate the potential correlation and deep patterns behind the program crash event. Based on the target aggregation feature data, the recurrence risk index of each program crash event can be predicted, providing a quantitative basis for risk assessment. Finally, according to the risk level determined by the recurrence risk index, the crash event is loaded into the risk management session of the target supply chain management application, realizing efficient classification and management of crash events.

[0082] Therefore, by deeply analyzing and processing the crash feature data of program crash events, the ability to identify, analyze and assess risks of crash events has been significantly improved, providing strong support for the stable operation and continuous improvement of the supply chain management platform, which not only improves management efficiency but also reduces the potential losses caused by system crashes.

[0083] In a possible implementation, step S130 includes:

[0084] Step S131 : randomly determining two of the crash feature clusters from the Y crash feature clusters for pairing, and generating Z potential correlation feature combinations.

[0085] In this embodiment, the server has extracted Y crash feature clusters from multiple program crash events through cluster analysis, and these crash feature clusters reflect different types of crash modes and associated crash nodes. Now, the server needs to explore the potential associations between these clusters.

[0086] The server starts by randomly selecting two pairs from the Y crash signature clusters. For example, assuming there are 5 crash signature clusters (A, B, C, D, E), the server may first select A and B to pair, then A and C, and so on, until all possible pairs have been considered.

[0087] Each pairing generates a new potential correlation feature combination, which represents the possible intrinsic connection between different crash modes. For example, pairing A and B generates a potential correlation feature combination AB.

[0088] Step S132: Perform potential correlation enhancement processing on each of the Z potential correlation feature combinations, and generate a potential correlation representation between combinations corresponding to each of the Z potential correlation feature combinations.

[0089] After generating potential correlation feature combinations, the server needs to further process these potential correlation feature combinations to strengthen the potential correlation between them and generate potential correlation representations between combinations corresponding to each combination.

[0090] First, the server can encode each crash feature cluster and convert it into a numerical vector form, which is helpful for subsequent mathematical calculations and association analysis. For each potential association feature combination (such as AB), the server can calculate the participation coefficients of the two encoded vectors, which reflect the relative importance of each cluster in the potential association. Using the calculated participation coefficients, the server performs a weighted calculation on the two encoded vectors. This process aims to generate a new vector that can represent the potential association between the two crash feature clusters, that is, the potential association representation between the combinations.

[0091] Finally, the server can obtain Z potential correlation feature combinations and the potential correlation representations between combinations corresponding to each combination. These potential correlation representations between combinations not only contain the characteristics of a single crash mode, but also reveal the deep connections between different modes.

[0092] In the actual application of supply chain management, this potential correlation analysis can help the operation and maintenance team more accurately identify the complex chain of factors that lead to system crashes. For example, an event that seems to be caused by a database service crash may actually be related to a preceding network service anomaly or a memory leak in the application layer. Through in-depth analysis of the server and potential correlation configuration, these hidden connections can be revealed, providing strong support for fundamental problem solving.

[0093] In a possible implementation, step S131 includes:

[0094] Step S1311, encoding each of the crash feature clusters to generate Y code vectors corresponding to the Y crash feature clusters.

[0095] In this step, the server will encode each crash feature cluster and convert it into a digital form, namely a coding vector, which can be used for further analysis.

[0096] Assume that the server has five crash feature clusters, labeled C1, C2, C3, C4, and C5. For each crash feature cluster, the server can perform encoding operations.

[0097] For C1, the server analyzes its internal crash characteristics, such as exception type, occurrence frequency, memory usage, etc., and converts this information into a numerical encoding vector V1, which may contain multiple dimensions, each dimension representing a specific crash feature.

[0098] Similarly, the server performs the same encoding operation on C2, C3, C4, and C5 to generate corresponding encoding vectors V2, V3, V4, and V5.

[0099] Finally, the server obtains 5 encoding vectors, each of which represents the detailed feature information of the corresponding crash feature cluster.

[0100] Step S1312: Pair any two of the encoding vectors to generate Z potential correlation feature combinations.

[0101] In this step, the server pairs the encoding vectors generated in the previous step to explore the possible correlations between different crash feature clusters.

[0102] For example, continuing with the 5 encoding vectors (V1, V2, V3, V4, V5) in the previous step, the server will now perform a pairing operation.

[0103] The server first pairs V1 with V2 to form a potential correlation feature combination P1, which represents the possible correlation between the two crash feature clusters C1 and C2.

[0104] Next, the server pairs V1 with V3 to form a potential correlation feature combination P2. Similarly, this combination reflects the potential correlation between C1 and C3.

[0105] The server continues this pairing process, pairing V1 with V4 and V5 to form P3 and P4; pairing V2 with V3, V4, and V5 to form P5, P6, and P7; and so on, until all possible pairings have been considered.

[0106] Ultimately, the server generates a list of multiple potential correlation feature combinations that will help further analyze the correlation between crash features and thus optimize system stability and performance.

[0107] Through the above two steps, the server can systematically analyze and explore the correlation between crash feature clusters, providing strong support for subsequent troubleshooting and system optimization.

[0108] In a possible implementation, step S132 includes:

[0109] Step S1321, obtaining the participation coefficients corresponding to the two matching encoding vectors in each of the potential associated feature combinations.

[0110] In this step, the server needs to determine the relative importance of the two encoding vectors in each potential correlation feature combination, which is usually achieved by obtaining their respective participation coefficients.

[0111] For example, the server has generated multiple potential correlation feature combinations, and now needs to process each potential correlation feature combination. Take one of the potential correlation feature combinations as an example, assuming that it consists of encoding vectors V1 and V2.

[0112] The server first accesses a predefined coefficient database or model, which may contain the participation coefficients assigned to each code vector based on historical data or expert knowledge.

[0113] For the encoding vector V1, the server retrieves its corresponding participation coefficient α from the database or model. The participation coefficient can be a value between 0 and 1, representing the weight of V1 in the potential association enhancement process. Similarly, for the encoding vector V2, the server retrieves its corresponding participation coefficient β. The sum of α and β is usually set to 1 to ensure the accuracy of the weighted calculation. The server records these two participation coefficients for use in the next step.

[0114] Step S1322, weighted calculation is performed on the participation coefficient and the two matching coding vectors in each combination of potential correlation features to generate a potential correlation representation between combinations corresponding to each combination of potential correlation features.

[0115] In this step, the server uses the participation coefficients obtained in the previous step to weight the code vectors to generate a new representation that reflects the potential association between the two code vectors.

[0116] Continuing with the example of the encoding vectors V1 and V2 in the previous step, the server will now perform a weighted calculation.

[0117] The server first obtains the numerical representation of V1 and V2, which can be multidimensional, reflecting different crash characteristics. For each feature value in V1, the server multiplies it by the participation coefficient α. Similarly, for each feature value in V2, it multiplies it by the participation coefficient β.

[0118] Next, the server combines the weighted feature values ​​of V1 and V2, which can be achieved by simple addition, averaging, or other more complex combinations, depending on the pre-defined algorithm or model.

[0119] Finally, the server obtains a new numerical representation, namely the potential correlation representation between combinations, which integrates the information of V1 and V2 and reflects the potential correlation between them. This new representation can be used for subsequent tasks such as crash analysis, fault prediction or system optimization.

[0120] Through the above two steps, the server can systematically strengthen and process the correlations in the combination of potential correlation features, so as to more accurately understand and predict the crash behavior of the system.

[0121] In a possible implementation, step S1321 includes:

[0122] Step S1321-1, combining the two matching coding vectors in each combination of potential associated features to generate a target coding vector.

[0123] Step S1321-1, pass the target coding vector through the attention network to generate the participation coefficient.

[0124] Step S1322 includes: performing weighted calculation on the participation coefficient and the target coding vector to generate a potential correlation representation between combinations corresponding to each combination of potential correlation features.

[0125] In this step, the server needs to combine the two encoding vectors in each potential associated feature combination to form a new encoding vector, namely the target encoding vector.

[0126] For example, assume that the server has generated multiple potential correlation feature combinations and is now processing one of the combinations consisting of encoding vector A and encoding vector B.

[0127] The server first obtains the coding vector A and coding vector B, which respectively represent the feature information of two crash feature clusters. Next, the server can combine the two coding vectors. There are many ways to combine, for example, the two coding vectors can be directly spliced ​​together, or they can be fused into a new coding vector through a certain function (such as average, weighted average, etc.). In this scenario, it is assumed that the server chooses the direct splicing method to splice coding vector A and coding vector B together to form a new, longer coding vector, namely the target coding vector C. The server saves this target coding vector C for the next step of processing.

[0128] Next, the server needs to use the attention network to process the target encoding vector and generate the corresponding participation coefficient.

[0129] Continuing the above scenario, the server inputs the target code vector C into a pre-trained attention network, which is designed to dynamically generate a set of participation coefficients based on the characteristics of the input vector. The attention network processes the target code vector C through internal calculation processes (such as self-attention mechanism, fully connected layer, etc.) and outputs two participation coefficients: coefficient α and coefficient β, which represent the importance of code vector A and code vector B in subsequent weighted calculations.

[0130] The server records the two participation coefficients α and β for the next step of weighted calculation.

[0131] Therefore, the server will use the participation coefficient generated in the previous step to perform weighted calculation on the target encoding vector to generate a new representation, namely the potential association representation between the combinations.

[0132] Continuing the above scenario: the server again obtains the target code vector C, and the previously generated participation coefficients α and β.

[0133] Next, the server can perform a weighted calculation on the target code vector C. Specifically, it multiplies each part of vector C (which may correspond to different parts of the original two code vectors) by coefficients α and β respectively. If the target code vector C is directly concatenated from the original code vectors A and B, the first part (corresponding to the part of code vector A) will be multiplied by coefficient α, and the second part (corresponding to the part of code vector B) will be multiplied by coefficient β.

[0134] Through this weighted calculation method, the server obtains a new vector representation, namely, the combined potential correlation representation, which fuses the information of the original two encoding vectors and is weighted according to their importance (determined by the participation coefficient generated by the attention network). This representation can more effectively reflect the potential correlation between the two collapsed feature clusters.

[0135] Finally, the server can save the potential correlation representation between the combinations for subsequent tasks such as crash analysis and system optimization.

[0136] In a possible implementation, step S120 may include:

[0137] Step S121 : obtaining a path attribute corresponding to each linkage crash path data in the crash feature data, wherein the path attribute represents a relationship feature between a crash node element corresponding to each linkage crash path data and the program crash event.

[0138] In this embodiment, the server starts to process a large amount of program crash event data, and the program crash event data includes various information when the program crashes during operation.

[0139] The server first reads the stored crash feature data, which records the relevant information of each crash event in detail, such as the time when the crash occurred, the system status at the time of the crash, the code modules involved, etc. Among this information, the server pays special attention to the linkage crash path data, which describes the changes in the program execution path from the start of the program to the crash.

[0140] The server further analyzes these linked crash path data and extracts the path attributes of each path. These path attributes may include the length of the path, the function call sequence involved, the execution frequency of specific functions or modules, etc. These attributes can characterize the relationship characteristics between the crash node elements corresponding to each linked crash path data and the program crash event.

[0141] Step S122: clustering the crash feature data according to the crash node elements and the path attributes to generate Y crash feature clusters.

[0142] Continuing with the above scenario, after determining the crash node elements and path attributes, the server will then perform cluster analysis on these data. The purpose of clustering is to group crash events with similar characteristics into one category for further analysis and processing.

[0143] The server may select a suitable clustering algorithm, such as K-means, hierarchical clustering, or DBSCAN, which can automatically group data based on their similarities.

[0144] During the clustering process, the server can consider the comprehensive information of crash node elements and path attributes. For example, if some crash events involve the same function call sequence or crash on similar program execution paths, they are likely to be clustered together.

[0145] After processing by the clustering algorithm, the server will eventually generate Y crash feature clusters. Each cluster contains a group of crash events with similar crash features.

[0146] These clusters are crucial for subsequent program debugging and optimization. By analyzing the crash events in each cluster, developers can more easily locate potential problems in the program and fix them.

[0147] In summary, by obtaining the path attributes of the linked crash path data and clustering the crash feature data based on these attributes and crash node elements, the server can effectively group similar crash events, thereby helping developers solve program crash problems more efficiently.

[0148] In a possible implementation, step S140 includes: traversing each of the program crash events, concatenating Z potential correlation representations between the combinations corresponding to the Z potential correlation feature combinations, and generating target converged feature data corresponding to the program crash event.

[0149] In this embodiment, for each program crash event, the server determines the potential correlation feature combination associated therewith, and each potential correlation feature combination has a corresponding potential correlation representation between combinations, which are generated through previous weighted calculations, attention network processing or other methods.

[0150] The server now concatenates the potential association representations between these combinations. The concatenation process is similar to connecting multiple data blocks in a certain order to form a longer data sequence. In this scenario, the server can concatenate the potential association representations between Z combinations in a predetermined order (such as sorting by the importance of the feature combination, the frequency of occurrence, etc.).

[0151] After the splicing is completed, the server obtains a long data sequence containing all potential associated feature combination information, which is the target converged feature data corresponding to the current crash event.

[0152] By splicing the potential correlation representations between Z combinations, the server generates the target converged feature data of the current crash event, which not only contains the detailed information of the crash event, but also incorporates the information of the potential correlation feature combinations related to it. The server can store this target converged feature data for subsequent analysis and processing. For example, this data can be used to train crash prediction models, optimize program performance, or perform troubleshooting.

[0153] Therefore, by traversing each program crash event and combining the potential associated features related to it, the server can generate target converged feature data containing rich information, providing strong support for subsequent crash event analysis and processing.

[0154] In a possible implementation, step S140 further includes:

[0155] Step S141, the target aggregation feature data corresponding to each of the program crash events are mapped with a deep neural network of a recurrence risk prediction model to generate a target recurrence risk feature mapping vector corresponding to each of the program crash events.

[0156] In this embodiment, the server has generated target converged feature data corresponding to each program crash event, and these target converged feature data integrate various potential associated features related to the crash event. Now, the server will use a pre-trained recurrence risk prediction model to evaluate the recurrence risk of these crash events.

[0157] The server first loads the recurrence risk prediction model, which contains a deep neural network specifically used to extract features related to recurrence risk from complex feature data.

[0158] For each target aggregated feature data of a program crash event, the server inputs it into a deep neural network, which performs nonlinear transformations on the data through multiple layers of neuron connections and activation functions, thereby capturing deeper feature relationships.

[0159] After the deep neural network processes the data, it will output a recurrence risk feature mapping vector. The recurrence risk feature mapping vector is learned by the neural network based on the input data and can represent the feature representation of the recurrence risk of the crash event.

[0160] Step S142, mapping the target recurrence risk feature vector corresponding to each of the program crash events, using the softmax network of the recurrence risk prediction model, to generate a recurrence risk index corresponding to each of the program crash events.

[0161] Continuing with the above scenario: The server then inputs the recurrence risk feature map vector corresponding to each program crash event into the softmax network of the recurrence risk prediction model. The softmax network is the classification layer in the model, which can convert the feature map vector into a probability distribution.

[0162] In the softmax network, the output of each neuron represents the probability of a category. In this scenario, the categories may represent different recurrence risk levels. Through the calculation of the softmax function, the network will output the probability value of each risk level. Based on these probability values, the server determines the recurrence risk index of the current crash event. The recurrence risk index can be a weighted score that combines the probabilities of each risk level, or directly the risk level corresponding to the highest probability.

[0163] Finally, the server saves the recurrence risk index of each program crash event, which can be used for subsequent crash event management, priority sorting, or as input to the early warning system.

[0164] Therefore, by utilizing the deep neural network and softmax network of the recurrence risk prediction model, the server can accurately predict the recurrence risk index of each program crash event, thereby providing valuable decision support for software development and maintenance teams.

[0165] In a possible implementation, before step S141, the recurrence risk prediction model training step includes:

[0166] Step S1011, acquiring template collapse feature data corresponding to a template collapse event, and clustering the template collapse feature data to generate Y template collapse feature clusters, wherein the template collapse feature data represents template linkage collapse path data from a plurality of different collapse nodes.

[0167] In this embodiment, the server first collects a large amount of template crash event data. These template crash event data are template linkage crash path data collected from multiple different crash nodes, which can fully reflect the crash characteristics of the program in different situations.

[0168] The server preprocesses these template crash feature data and extracts key feature information, such as the system status at the time of the crash, memory usage, called function stack, etc. Next, the server uses clustering algorithms (such as K-means, hierarchical clustering, etc.) to cluster the template crash feature data. The purpose of clustering is to classify template crash events with similar crash features into one category to form Y template crash feature clusters.

[0169] Step S1012, performing potential association configuration on the Y template collapse feature clusters to generate Z template potential association feature combinations and a template potential association feature combination corresponding to each template potential association feature combination.

[0170] After obtaining Y template crash feature clusters, the server further analyzes the potential associations between these template crash feature clusters. By configuring the algorithm, the server identifies Z template potential association feature combinations, which reflect the possible associations between different clusters.

[0171] The server aggregates the relevant feature data based on the potential associated feature combinations of these templates. This process is similar to fusing the useful information in different clusters together to form more comprehensive and representative feature data, namely, template aggregated feature data.

[0172] Step S1013, according to each of the template potential associated feature combinations, Z template potential associated feature combinations are aggregated to generate template aggregated feature data corresponding to the template collapse event.

[0173] In this embodiment, the server inputs the template aggregation feature data into the deep neural network of the recurrence risk prediction model, which has a strong feature extraction capability and can capture the deep relationship in the data. Through the processing of the deep neural network, the server obtains the template recurrence risk feature mapping vector corresponding to the template collapse event, which is automatically learned by the neural network based on the input data and can effectively characterize the recurrence risk of the collapse event.

[0174] Step S1014, aggregating feature data of the template, performing recurrence risk feature mapping using a deep neural network of a recurrence risk prediction model, and generating a template recurrence risk feature mapping vector corresponding to the template collapse event.

[0175] Step S1015, mapping the template recurrence risk feature vector using the softmax network of the recurrence risk prediction model to generate a template recurrence risk index.

[0176] Step S1016, calculating a training error parameter for recurrence risk based on the template recurrence risk index and the recurrence risk supervised learning data, and iteratively training the recurrence risk prediction model based on the training error parameter.

[0177] Next, the server inputs the template recurrence risk feature mapping vector into the softmax network of the recurrence risk prediction model. The softmax network can convert the template recurrence risk feature mapping vector into a specific recurrence risk index, which intuitively reflects the recurrence possibility of the template collapse event.

[0178] Next, the server obtains the recurrence risk supervised learning data, which contains known crash events and their corresponding true recurrence risk index. By comparing the difference between the template recurrence risk index and the true risk index, the server calculates the training error parameter for the recurrence risk, which reflects the accuracy of the model's prediction results. Based on this training error parameter, the server iteratively trains the recurrence risk prediction model. In each iteration, the model adjusts its internal weights and parameters according to the error parameter to more accurately predict the recurrence risk of crash events.

[0179] Through multiple iterations of training, the performance of the recurrence risk prediction model will gradually improve, and it will eventually be able to more accurately assess the recurrence risk of program crash events.

[0180] In a possible implementation, step S1011 includes:

[0181] Step S1011 - 1 , obtaining a template path attribute corresponding to each template-linked crash path data in the template crash feature data, wherein the template path attribute represents a relationship feature between a crash node element corresponding to each of the linked crash path data and the template crash event.

[0182] Step S1011 - 2 , clustering the template crash feature data according to the crash node elements and the template path attributes, and generating Y template crash feature clusters.

[0183] In this embodiment, the server starts to process template crash events. These template crash events contain rich template crash feature data. These template crash feature data record the detailed situation when the program crashes under a specific template.

[0184] The server first deeply analyzes these template crash feature data and extracts the template path attributes corresponding to each template linkage crash path data. These template path attributes reflect the relationship characteristics between the crash node elements and the template crash events, which are the key to understanding the causes and patterns of the crash.

[0185] Template path attributes may include, but are not limited to: the function sequence called when the crash occurs, memory usage, system resource usage, etc. The server extracts these key information from the original crash data through specific data extraction technology.

[0186] After obtaining the template path attributes, the server then uses these template path attributes and crash node elements to perform cluster analysis on the template crash feature data. The purpose of clustering is to classify crash events with similar crash features and path attributes into one category.

[0187] The server selects a suitable clustering algorithm (such as K-means, spectral clustering, etc.) and sets the corresponding parameters. During the clustering process, the algorithm divides the data into Y different clusters based on the similarity of the template path attributes and the collapsed node elements.

[0188] As clustering proceeds, the server gradually forms Y template crash feature clusters. Each template crash feature cluster represents a set of template crash events with similar crash features and path attributes.

[0189] After clustering is completed, the server can conduct in-depth analysis of each template crash feature cluster to understand the crash patterns and causes behind them. This information will provide strong support for subsequent program optimization, error troubleshooting and prevention.

[0190] Therefore, by obtaining template path attributes and performing cluster analysis based on these attributes, the server can effectively classify template crash events with similar crash characteristics and path attributes together, thereby forming a more representative template crash feature clustering cluster, which provides an important data foundation and analysis basis for subsequent recurrence risk prediction and program optimization.

[0191] In a possible implementation, step S1012 includes:

[0192] Step S1012-1, encoding each of the template collapse feature clusters to generate Y template encoding vectors corresponding to the Y template collapse feature clusters.

[0193] In this embodiment, after completing the clustering of the template crash feature data, the server obtains Y template crash feature clusters. In order to perform subsequent data processing and analysis, the server needs to encode these template crash feature clusters.

[0194] The server selects a suitable encoding method, for example, it can use One-Hot Encoding or Word Embedding.

[0195] For each template collapse feature cluster, the server generates a unique template encoding vector according to the selected encoding method, and the template encoding vector can effectively represent the features and attributes of the cluster.

[0196] After encoding, the server obtains Y template encoding vectors corresponding to Y template collapse feature clusters, and these template encoding vectors will be used for subsequent potential association configuration.

[0197] Step S1012-2: Pair any two of the template encoding vectors to generate Z potential associated feature combinations of the template.

[0198] Continuing with the above scenario, in order to explore the potential associations between different template collapse feature clusters, the server needs to pair different template encoding vectors. The server arbitrarily selects two different template encoding vectors from Y template encoding vectors for pairing. This pairing is combinatorial, that is, the order is not considered. In this way, the server generates Z template potential association feature combinations. Each template potential association feature combination contains two template encoding vectors from different clusters.

[0199] Step S1012-3, obtaining participation coefficients corresponding to two matching template encoding vectors in each combination of potential associated features of the template.

[0200] Before performing weighted calculation, the server needs to determine the weight of each template encoding vector in the potential associated feature combination, that is, the participation coefficient.

[0201] The server assigns a participation coefficient to each template encoding vector according to a certain strategy or algorithm (such as based on the size, density or importance of the clustering clusters, etc.), and the participation coefficient reflects the importance of the vector in the potential associated feature combination. For each template potential associated feature combination, the server obtains the participation coefficients of the two matching template encoding vectors.

[0202] Step S1012-4, performing weighted calculation on the participation coefficient and the two matching template encoding vectors in each template potential associated feature combination, to generate a template potential associated feature combination corresponding to each template potential associated feature combination.

[0203] For each template potential association feature combination, the server multiplies the two template encoding vectors by their respective participation coefficients and then adds the results. Through this weighted calculation method, the server generates a template potential association feature combination vector corresponding to each template potential association feature combination, which integrates feature information from two different clustering clusters and takes into account their respective importance. Finally, the server obtains Z weighted template potential association feature combination vectors, which will be used for subsequent recurrence risk prediction model training and analysis.

[0204] In a possible implementation, step S150 includes:

[0205] Step S151 , mapping the X program crash events to risk levels according to the recurrence risk index to generate risk levels for the X program crash events.

[0206] Step S152: according to the risk levels of the X program crash events, the X program crash events are loaded into the risk management session in order of risk levels.

[0207] In this embodiment, after completing the calculation of the recurrence risk index, the server needs to evaluate the risk level of each program crash event.

[0208] The server first sets a series of risk level thresholds, for example, five risk levels may be set: very low, low, medium, high, and very high, and a corresponding recurrence risk index range is set for each level.

[0209] Next, the server traverses X program crash events and checks the recurrence risk index of each event. For each program crash event, the server compares its recurrence risk index with the preset risk level threshold to determine the risk level of the event. For example, if the recurrence risk index of a crash event falls within the "high" risk range, the event is mapped to a "high" risk level. After this step, the server successfully maps the corresponding risk levels for each of the X program crash events.

[0210] After determining the risk level of each program crash event, the server needs to load this information into the risk management session of the supply chain management application for further analysis and processing.

[0211] The server first sorts the X program crash events according to the risk level. Events with high risk levels are ranked first, and events with low risk levels are ranked last. Next, the server opens the risk management session interface of the target supply chain management application, which can be a platform specifically used to display and handle risk events. The server loads the information of the program crash events into the risk management session one by one in the sorted order, which includes the description of the crash event, the time of occurrence, the recurrence risk index, and the risk level. After loading, the risk management session interface will display a list of program crash events sorted by risk level, so that supply chain managers can see at a glance which crash events have the highest risk, and thus give priority to these high-risk events.

[0212] Through the above steps, the server not only assesses the risk level of the program crash event, but also effectively presents this information in the risk management session of the supply chain management application, providing strong support for subsequent risk response and program optimization.

[0213] Figure 2The hardware structure of the data processing system 100 applied to the supply chain management platform for implementing the data processing method applied to the supply chain management platform provided in the embodiment of the present application is shown as follows: Figure 2 As shown, the data processing system 100 applied to the supply chain management platform may include a processor 110 , a machine-readable storage medium 120 , a bus 130 , and a communication unit 140 .

[0214] In one possible design, the data processing system 100 applied to the supply chain management platform can be a single server or a server group. The server group can be centralized or distributed (for example, the data processing system 100 applied to the supply chain management platform can be a distributed system). In some embodiments, the data processing system 100 applied to the supply chain management platform can be local or remote. For example, the data processing system 100 applied to the supply chain management platform can access information and / or data stored in the machine-readable storage medium 120 via a network. For another example, the data processing system 100 applied to the supply chain management platform can be directly connected to the machine-readable storage medium 120 to access the stored information and / or data. In some embodiments, the data processing system 100 applied to the supply chain management platform can be implemented on the data processing system applied to the supply chain management platform. By way of example only, the data processing system applied to the supply chain management platform may 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.

[0215] 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 used by the data processing system 100 applied to the supply chain management platform to execute or use to complete the exemplary methods described in this application.

[0216] During 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 data processing method applied to the supply chain management platform 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.

[0217] The specific implementation process of the processor 110 can refer to the various method embodiments executed by the data processing system 100 applied to the supply chain management platform mentioned above. The implementation principles and technical effects are similar and will not be repeated in this embodiment.

[0218] In addition, an embodiment of the present application further provides a readable storage medium, in which computer executable instructions are preset. When a processor executes the computer executable instructions, the data processing method applied to the supply chain management platform as above is implemented.

[0219] It should be noted that in order to simplify the description disclosed in the present application and thereby facilitate the understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of the present application, various features are sometimes grouped into one embodiment, drawing or description thereof.

Claims

1. A data processing method applied to a supply chain management platform, characterized in that: The method comprises: Obtain crash feature data corresponding to each of the X program crash events of the target supply chain management application, wherein the crash feature data represents linkage crash path data from a plurality of different crash nodes; Traversing each of the program crash events, clustering the crash feature data, and generating Y crash feature clusters, where Y is an integer not less than 2; Traversing each of the program crash events, performing potential association configuration on the Y crash feature clusters, generating Z potential association feature combinations and potential association representations between combinations corresponding to each of the potential association feature combinations, where Z is an integer not less than 1; Traversing each of the program crash events, and according to the potential correlation representation between each of the combinations, aggregating Z of the potential correlation feature combinations to generate target aggregate feature data corresponding to the program crash event; Predicting a recurrence risk index corresponding to each of the program crash events based on the target aggregation feature data corresponding to each of the program crash events; Determining risk levels of the X program crash events according to the recurrence risk index, and loading the X program crash events into a risk management session corresponding to the target supply chain management application according to the risk level; The predicting the recurrence risk index corresponding to each program crash event according to the target aggregation feature data corresponding to each program crash event includes: The target aggregation feature data corresponding to each of the program crash events are mapped to recurrence risk features using a deep neural network of a recurrence risk prediction model to generate a target recurrence risk feature mapping vector corresponding to each of the program crash events; The target recurrence risk feature mapping vector corresponding to each program crash event is used to generate a recurrence risk index corresponding to each program crash event by using the softmax network of the recurrence risk prediction model; Before the target aggregation feature data corresponding to each program crash event is mapped using a deep neural network of a recurrence risk prediction model to generate a target recurrence risk feature mapping vector corresponding to each program crash event, the training step of the recurrence risk prediction model includes: Acquire template collapse feature data corresponding to the template collapse event, and cluster the template collapse feature data to generate Y template collapse feature clusters, wherein the template collapse feature data represents template linkage collapse path data from multiple different collapse nodes; Performing potential association configuration on the Y template collapse feature clusters to generate Z template potential association feature combinations and a weighted association feature combination corresponding to each template potential association feature combination; According to each of the template potential associated feature combinations, Z template potential associated feature combinations are aggregated to generate template aggregated feature data corresponding to the template collapse event; Aggregating feature data of the template, performing recurrence risk feature mapping using a deep neural network of a recurrence risk prediction model, and generating a template recurrence risk feature mapping vector corresponding to the template collapse event; The template recurrence risk feature mapping vector is used to generate a template recurrence risk index using the softmax network of the recurrence risk prediction model; Calculating a training error parameter for recurrence risk based on the template recurrence risk index and the recurrence risk supervised learning data, and iteratively training the recurrence risk prediction model based on the training error parameter; The acquiring of template collapse feature data corresponding to the template collapse event, and clustering of the template collapse feature data to generate Y template collapse feature clusters, includes: Acquire a template path attribute corresponding to each template linkage crash path data in the template crash feature data, wherein the template path attribute represents a relationship feature between a crash node element corresponding to each linkage crash path data and the template crash event; Clustering the template crash feature data according to the crash node element and the template path attribute to generate Y template crash feature clusters; The step of configuring the Y template collapse feature clusters for potential association to generate Z template potential association feature combinations and a weighted association feature combination corresponding to each template potential association feature combination includes: Encode each of the template collapse feature clusters to generate Y template encoding vectors corresponding to the Y template collapse feature clusters; Pairing any two of the template encoding vectors to generate Z potential associated feature combinations of the templates; Obtaining participation coefficients corresponding to two matching template encoding vectors in each combination of potential associated features of the template; Performing weighted calculation on the participation coefficient and the two matching template encoding vectors in each template potential associated feature combination to generate a weighted associated feature combination corresponding to each template potential associated feature combination; The step of determining the risk levels of the X program crash events according to the recurrence risk index, and loading the X program crash events into the risk management session corresponding to the target supply chain management application according to the risk levels, respectively, includes: Performing risk level mapping on the X program crash events according to the recurrence risk index to generate risk levels of the X program crash events; According to the risk levels of the X program crash events, the X program crash events are loaded into the risk management session in order of risk levels.

2. The data processing method applied to the supply chain management platform according to claim 1, characterized in that: The traversing each of the program crash events, performing potential correlation configuration on the Y crash feature clusters, generating Z potential correlation feature combinations and potential correlation representations between combinations corresponding to each of the potential correlation feature combinations, includes: Randomly determine two of the crash feature clusters from the Y crash feature clusters for pairing, to generate Z potential correlation feature combinations; Perform potential correlation enhancement processing on each of the Z potential correlation feature combinations respectively, and generate a potential correlation representation between combinations corresponding to each of the potential correlation feature combinations.

3. The data processing method applied to the supply chain management platform according to claim 2 is characterized in that: The step of arbitrarily determining two of the crash feature clusters from the Y crash feature clusters for pairing to generate Z potential correlation feature combinations includes: Encode each of the crash feature clusters to generate Y code vectors corresponding to the Y crash feature clusters; Pair any two of the encoding vectors to generate Z potential association feature combinations.

4. The data processing method applied to the supply chain management platform according to claim 3 is characterized in that: The step of performing a potential correlation strengthening process on each of the Z potential correlation feature combinations to generate a potential correlation representation between combinations corresponding to each potential correlation feature combination includes: Obtaining participation coefficients corresponding to two matching encoding vectors in each of the potential associated feature combinations; Performing weighted calculation on the participation coefficient and the two matching encoding vectors in each potential association feature combination to generate a potential association representation between combinations corresponding to each potential association feature combination; The step of obtaining the participation coefficients corresponding to the two matching encoding vectors in each of the potential association feature combinations includes: Combining the two matching encoding vectors in each of the potential associated feature combinations to generate a target encoding vector; Passing the target encoding vector through an attention network to generate the participation coefficient; The step of performing weighted calculation on the participation coefficient and the two matching encoding vectors in each potential association feature combination to generate a potential association representation between combinations corresponding to each potential association feature combination includes: The participation coefficient and the target coding vector are weightedly calculated to generate a potential correlation representation between combinations corresponding to each potential correlation feature combination.

5. The data processing method applied to the supply chain management platform according to claim 1, characterized in that: The traversing each of the program crash events, clustering the crash feature data, and generating Y crash feature clusters, includes: Acquire a path attribute corresponding to each linkage crash path data in the crash feature data, wherein the path attribute represents a relationship feature between a crash node element corresponding to each linkage crash path data and the program crash event; The crash feature data is clustered according to the crash node elements and the path attributes to generate Y crash feature clusters.

6. The data processing method applied to the supply chain management platform according to claim 1, characterized in that: The traversing each of the program crash events, aggregating Z potential correlation feature combinations according to the potential correlation representation between each of the combinations, and generating target aggregate feature data corresponding to the program crash event, includes: Each of the program crash events is traversed, and the potential correlation representations between the Z combinations corresponding to the Z potential correlation feature combinations are spliced ​​to generate target converged feature data corresponding to the program crash event.

7. A data processing system applied to a supply chain management platform, characterized in that: The data processing system applied to the supply chain management platform 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 data processing method applied to the supply chain management platform as described in any one of claims 1 to 6 above.

Images