Positioning method and apparatus, electronic device, and computer storage medium

Abstract

Embodiments of the present specification provide an application positioning method and device, electronic equipment and computer storage medium, the method comprising: obtaining a semantic correlation degree between a first candidate application and a target positioning text based on the target positioning text and related text of the first candidate application, confirming a second candidate application in the first candidate application based on the semantic correlation degree, obtaining a third candidate application associated with the second candidate application from the first candidate application, obtaining a transaction correlation degree between fourth candidate applications based on a traffic call graph and a link relationship graph of the fourth candidate applications, wherein the fourth candidate applications include the second candidate application and the third candidate application, and obtaining a target application of the target positioning text from the fourth candidate applications based on the transaction correlation degree.

Description

Application positioning methods, devices, electronic devices and computer storage media Technical Field

[0001] This specification relates to the field of computer technology, and in particular to application positioning methods, devices, electronic devices and computer storage media. Background Technology

[0002] Rapidly responding to and resolving customer complaints is a crucial aspect of improving user experience. However, due to the large volume of services provided, the sheer number of functionalities involved, and the fact that there isn't a one-to-one correspondence between front-end service pages and back-end functionalities, pinpointing potential back-end failures based on customer complaints is an extremely challenging task.

[0003] Therefore, there is an urgent need to propose a method that can effectively pinpoint the root causes of customer complaints. Summary of the Invention

[0004] The main purpose of this specification is to provide an application positioning method, device, electronic device, and computer storage medium, aiming to provide an efficient and highly reliable application positioning method. The technical solution is as follows:

[0005] Firstly, embodiments of this specification provide an application positioning method, including:

[0006] Based on the target location text and the relevant text of the first candidate application, the semantic correlation between the first candidate application and the target location text is obtained;

[0007] Based on the semantic relevance, a second candidate application is identified from the first candidate applications;

[0008] Obtain a third candidate application associated with the second candidate application from the first candidate application;

[0009] Based on the traffic call graph and link relationship graph of the fourth candidate application, the transaction correlation degree between the fourth candidate applications is obtained. The fourth candidate application includes the second candidate application and the third candidate application.

[0010] The target application for the target location text is obtained from the fourth candidate application based on the transaction correlation.

[0011] Secondly, embodiments of this specification provide a training method for a transaction correlation model, including:

[0012] Obtain the traffic call graph and link relationship graph corresponding to the first candidate application;

[0013] The traffic call graph and the link relationship graph are input into a graph convolutional neural network to obtain the graph node embedding vector of the first candidate application;

[0014] The similarity loss value corresponding to the graph node embedding vector is calculated using a similarity loss function;

[0015] The network parameters of the graph convolutional neural network are optimized based on the similarity loss value, and a pre-trained transaction correlation model is obtained based on the optimized graph convolutional neural network.

[0016] Thirdly, embodiments of this specification provide an application positioning device, including:

[0017] The semantic relevance module is used to obtain the semantic relevance between the first candidate application and the target location text based on the target location text and the relevant text of the first candidate application.

[0018] The first confirmation module is used to confirm the second candidate application in the first candidate application based on the semantic relevance.

[0019] The second confirmation module is used to obtain a third candidate application associated with the second candidate application from the first candidate application;

[0020] The transaction correlation module is used to obtain the transaction correlation between the fourth candidate applications based on the traffic call graph and link relationship graph of the fourth candidate application.

[0021] The positioning module is used to obtain the target application of the target positioning text from the fourth candidate application based on the transaction correlation.

[0022] Fourthly, embodiments of this specification provide an application positioning device, including:

[0023] The acquisition module is used to acquire the traffic call graph and link relationship graph corresponding to the first candidate application;

[0024] The embedding vector calculation module is used to input the traffic call graph and the link relationship graph into the graph convolutional neural network to obtain the graph node embedding vector of the first candidate application;

[0025] The loss function calculation module is used to calculate the similarity loss value corresponding to the graph node embedding vector through the similarity loss function;

[0026] An optimization module is used to optimize the network parameters of the graph convolutional neural network based on the similarity loss value, and to obtain a pre-trained transaction association model based on the optimized graph convolutional neural network.

[0027] Fifthly, embodiments of this specification provide an electronic device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method described above.

[0028] Sixthly, embodiments of this specification provide a computer storage medium storing a computer program, which, when executed by a processor, implements the steps of the method described above.

[0029] In a seventh aspect, embodiments of this specification provide a computer program product, comprising: a computer program that, when executed by a processor of an electronic device, enables the processor to at least implement the methods described in the first and second aspects.

[0030] In the embodiments of this specification, the semantic correlation between the first candidate application and the target location text is obtained based on the relevant text of the first candidate application. Based on the semantic correlation, a second candidate application is identified within the first candidate application. A third candidate application associated with the second candidate application is obtained from the first candidate application. The transaction correlation between the fourth candidate applications is obtained based on the traffic call graph and link relationship graph of the fourth candidate application. The fourth candidate application includes both the second and third candidate applications. Based on the transaction correlation, the target application of the target location text is obtained from the fourth candidate applications. This approach considers not only the semantic correlation between the target location text and the application text but also the transaction correlation between the application and the target location text, solving the problem of weak matching between the target location text and the application. Furthermore, by obtaining the transaction correlation between applications through the link relationship graph and traffic call graph, effective transaction connections are strengthened, allowing for better application investigation and improving the accuracy of application location for customer complaint root causes. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 is a schematic diagram of a scenario for an application positioning method provided in an embodiment of this specification;

[0033] Figure 2 is a flowchart illustrating an application positioning method provided in an embodiment of this specification;

[0034] Figure 3 is a flowchart illustrating an application positioning method provided in an embodiment of this specification;

[0035] Figure 4 is a flowchart illustrating an application positioning method provided in an embodiment of this specification;

[0036] Figure 5 is a flowchart illustrating an application positioning method provided in an embodiment of this specification;

[0037] Figure 6 is a flowchart illustrating an application positioning method provided in an embodiment of this specification;

[0038] Figure 7 is a flowchart illustrating a training method for a transaction correlation model provided in an embodiment of this specification;

[0039] Figure 8 is a flowchart illustrating a training method for a transaction correlation model provided in an embodiment of this specification;

[0040] Figure 9 is a flowchart illustrating a training method for a transaction correlation model provided in an embodiment of this specification.

[0041] Figure 10 is a schematic diagram of a training method for a transaction correlation model provided in an embodiment of this specification;

[0042] Figure 11 is a clustering diagram illustrating a training method for a transaction correlation model provided in an embodiment of this specification;

[0043] Figure 12 is a structural schematic diagram of an application positioning device provided in an embodiment of this specification;

[0044] Figure 13 is a structural schematic diagram of an application positioning device provided in an embodiment of this specification;

[0045] Figure 14 is a schematic diagram of the structure of an electronic device provided in an embodiment of this specification. Detailed Implementation

[0046] The technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this specification.

[0047] The application positioning device can be a terminal device such as a mobile phone, computer, tablet, smartwatch, or in-vehicle device, or it can be a module in the terminal device used to implement the application positioning method. The application positioning device can obtain the semantic correlation between the first candidate application and the target positioning text based on the target positioning text and the relevant text of the first candidate application. Based on the semantic correlation, the second candidate application is confirmed in the first candidate application. The third candidate application associated with the second candidate application is obtained from the first candidate application. Based on the traffic call graph and link relationship graph of the fourth candidate application, the transaction correlation between the fourth candidate applications is obtained. The fourth candidate application includes the second candidate application and the third candidate application. Based on the transaction correlation, the target application of the target positioning text is obtained from the fourth candidate application.

[0048] Correspondingly, the application location device can also train a transaction correlation model. The device can acquire the traffic call graph and link relationship graph corresponding to the first candidate application, input these graphs into a graph convolutional neural network to obtain the graph node embedding vectors of the first candidate application, calculate the similarity loss value corresponding to the graph node embedding vectors using a similarity loss function, optimize the network parameters of the graph convolutional neural network based on the similarity loss value, and obtain a pre-trained transaction correlation model based on the optimized graph convolutional neural network.

[0049] It should be noted that the application positioning device used for application positioning and the application positioning device used for training the transaction correlation model can be the same device or different devices.

[0050] Please refer to Figure 1, which is a schematic diagram of a scenario for an application positioning method provided in this specification. After the application positioning device obtains the target positioning text from the user device, it obtains the semantic correlation between the first candidate application and the target positioning text based on the target positioning text and the related text of the first candidate application. Based on the semantic correlation, it confirms the second candidate application in the first candidate application, obtains the third candidate application associated with the second candidate application from the first candidate application, obtains the transaction correlation between the fourth candidate applications based on the traffic call graph and link relationship graph of the fourth candidate application, and the fourth candidate application includes the second candidate application and the third candidate application. Based on the transaction correlation, it obtains the target application of the target positioning text from the fourth candidate application.

[0051] The application positioning method provided in this specification will be described in detail below with reference to specific embodiments.

[0052] Please refer to Figure 2, which is a flowchart illustrating an application positioning method provided in an embodiment of this specification. As shown in Figure 2, the method described in this embodiment may include the following steps S102-S110.

[0053] S102, based on the target location text and the relevant text of the first candidate application, obtain the semantic correlation degree between the first candidate application and the target location text;

[0054] S104, Based on the semantic relevance, confirm the second candidate application among the first candidate applications;

[0055] S106, Obtain a third candidate application associated with the second candidate application from the first candidate application;

[0056] S108, Based on the traffic call graph and link relationship graph of the fourth candidate application, the transaction correlation degree between the fourth candidate applications is obtained. The fourth candidate application includes the second candidate application and the third candidate application.

[0057] S110, based on the transaction correlation, obtain the target application of the target location text from the fourth candidate application.

[0058] The application location method described in this specification is applied to the location of applications associated with customer complaint text. Of course, the application location method provided in this specification can also be used to locate other types of text and applications, such as application review text and application functions. Since users may encounter some faults or errors when using application functions or services, users can provide feedback on the faults so that technicians can troubleshoot and repair them. Traditional root cause investigation of technical faults often relies on manual methods. After receiving a customer complaint, emergency personnel contact a large number of potentially related application managers and technicians, relying on experience to investigate possible fault paths. Current manual troubleshooting faces three challenges: First, for technical faults, anomaly interception before technical changes and monitoring after technical changes are already quite comprehensive, so faults that escalate to the customer complaint stage are highly concealed. It is difficult to troubleshoot using existing rules and experience. Second, the number of service scenarios provided is enormous, making it impractical for emergency personnel to master the terminology of so many service scenarios. Third, and most importantly, the semantics of front-end services and back-end functional applications in text are not necessarily perfectly matched. For example, the root cause application of a customer complaint about "failure to claim free benefits with a black card" is the merchant's access capability. The relationship between the customer complaint and the root cause application is a weak text match; therefore, relying solely on text matching is insufficient for location, and further consideration of the complex call relationships between applications is necessary. Thus, the embodiments in this specification provide a highly accurate application location method.

[0059] The following will provide a detailed explanation of each step:

[0060] S102, based on the target location text and the relevant text of the first candidate application, obtain the semantic correlation degree between the first candidate application and the target location text;

[0061] In the embodiments of this specification, the target location text refers to the text that needs to be associated with the application for location. For example, it can be extracted based on user complaint information. For instance, if the complaint information includes the user's complaint content and complaint time, the complaint content can be extracted as the target location text. The first candidate application refers to the matched application, which can be a specific application (APP) or an access capability. The related text of the first candidate application refers to the descriptive text of the first candidate application, such as the capabilities it can achieve and the application name.

[0062] Since both the target text and related text are text, the semantic relevance between the two texts can be calculated. Optionally, the application-related multi-class text data is first cleaned to obtain related text. Then, the application keywords are extracted from the related text using the TF-IDF algorithm (term frequency–inverse document frequency). Semantic relevance is then calculated based on the application keywords and the target text. For example, the application keywords and the target text can be mapped to the same low-dimensional space, and their similarity can be calculated as the semantic relevance. Optionally, the calculation of text semantic relevance can be implemented using a neural network model, such as semantic relevance modeling based on BERT.

[0063] S104, Based on the semantic relevance, confirm the second candidate application among the first candidate applications;

[0064] In the embodiments of this specification, after obtaining the semantic relevance, the semantic relevance between the first candidate application and the target location text is sorted, and the applications with the highest semantic relevance are selected, for example, the second candidate applications with the highest semantic relevance are selected for further matching and filtering. It should be noted that the number of second candidate applications can be set according to actual needs.

[0065] S106, Obtain a third candidate application associated with the second candidate application from the first candidate application;

[0066] In the embodiments of this specification, after obtaining the second candidate application based on semantic relevance, it is also necessary to obtain a third candidate application related to the second candidate application. It is understood that because the semantic relevance between the application and the target location text may be weak, the matched second candidate application, although having some relevance to the target location text, may not be accurate, or there may be other possible applications that have not been found. To further improve the accuracy of application location, a third candidate application associated with the second candidate application is obtained. Optionally, the third candidate application is obtained based on the link relevance or transaction relevance between the second candidate application and other applications in the first candidate application.

[0067] S108, Based on the traffic call graph and link relationship graph of the fourth candidate application, the transaction correlation degree between the fourth candidate applications is obtained. The fourth candidate application includes the second candidate application and the third candidate application.

[0068] In the embodiments of this specification, the third and second candidate applications are used as the fourth candidate applications, and a traffic call graph and a link relationship graph between the fourth candidate applications are obtained. The traffic call graph refers to a topology graph that abstracts the traffic call relationships between applications. Nodes represent all backend applications (such as the first candidate application), and edges between nodes represent the traffic call relationships between different applications. If there is no traffic call between two applications, there is no edge between these two nodes in the traffic call graph; if there is a traffic call, there is an edge. The link relationship graph refers to a topology graph that abstracts the connections between applications on predefined transaction links. Nodes represent all backend applications (such as the first candidate application). If two applications are on the same transaction link, there is an edge connecting these two nodes in the link relationship graph; otherwise, there is no edge.

[0069] Optionally, a partial graph with the fourth candidate application as a node can be selected from the link relationship graph and traffic call graph of the first candidate application to obtain the traffic call graph and link relationship graph of the fourth candidate application.

[0070] After obtaining the traffic call graph and link relationship graph among the fourth candidate applications, the transaction correlation between the fourth candidate applications is confirmed based on the traffic call graph and link relationship graph. Specifically, each candidate application is treated as a node. If there is a correlation between the nodes, there will be an edge between the nodes. The correlation between each node can be obtained based on the node-edge relationship in the graph. For example, if application A has an edge connection with application B in both the link relationship graph and the traffic call graph, while application C only has an edge connection with application B in the traffic call graph, then the transaction correlation between application A and application B is higher.

[0071] S110, based on the transaction correlation, obtain the target application of the target location text from the fourth candidate application.

[0072] In the embodiments of this specification, after confirming the transaction correlation between the fourth candidate applications, some candidate applications can be selected as target applications based on the transaction correlation. In one optional implementation, candidate applications that have relationships with more applications are selected from the fourth candidate applications. It can be understood that if an application is associated with many other applications, it indicates that the application has a stronger transaction dependency with other applications, and the fault described in the target location text is likely related to that application. In another optional implementation, the transaction correlation between the second candidate application and other candidate applications can be confirmed, and candidate applications with a high transaction correlation with the second candidate application can be selected as target applications. The number of target applications can be set according to the actual situation. In the scenario of customer complaint text location, in order to meet the needs of investigation and control the workload, 50 candidate applications can generally be selected as target applications and fed back to relevant personnel. Relevant personnel can further screen the problems described in the target location text, and the target applications can be sorted, with the applications ranked higher being the applications more likely to have faults. Relevant personnel can screen according to the sorted order.

[0073] In the embodiments of this specification, the semantic correlation between the first candidate application and the target location text is obtained based on the relevant text of the first candidate application. Based on the semantic correlation, a second candidate application is identified within the first candidate application. A third candidate application associated with the second candidate application is obtained from the first candidate application. The transaction correlation between the fourth candidate applications is obtained based on the traffic call graph and link relationship graph of the fourth candidate application. The fourth candidate application includes both the second and third candidate applications. The target application of the target location text is obtained from the fourth candidate applications based on the transaction correlation. By calculating the semantic correlation between the target location text and the candidate applications, and calculating the transaction correlation between the candidate applications, the semantic correlation between the application and the target location text, and the transaction correlation between the applications are aggregated and evaluated. This ensures that the selected target application encompasses both effective information of semantic correlation and transaction correlation, solving the problem of weak matching between the target location text and the root cause application text, and greatly improving the accuracy of application location.

[0074] Please refer to Figure 3, which is a flowchart illustrating an application positioning method provided in an embodiment of this specification. As shown in Figure 3, the method described in this embodiment may include the following steps S202-S206.

[0075] S202, Obtain descriptive text data related to the first candidate application;

[0076] Optionally, when calculating the semantic relevance between the first candidate application and the target location text, the descriptive text data related to the first candidate application is first obtained. The descriptive text data can be the application name and capability description of the first candidate application. This descriptive text data related to the first candidate application can be pre-stored in a database and directly retrieved when application location is required.

[0077] S204, perform word segmentation on the descriptive text data to obtain keyword data of the first candidate application, and use the keyword data as the related text;

[0078] In one optional implementation, the descriptive text is segmented into words, and keyword data is obtained. Word segmentation involves dividing the descriptive text. For example, a dictionary-based segmentation method can be used to extract a segmentation lexicon from a large-scale training corpus, while simultaneously calculating the word frequencies. For instance, sentence segmentation methods such as reverse maximum matching and N-shortest path segmentation can be used. Alternatively, word segmentation tools such as Jieba can be directly applied. Keyword data consists of keywords with application semantics extracted through word segmentation.

[0079] S206, Based on the matching of the relevant text and the target location text, the semantic correlation between the first candidate application and the target location text is obtained.

[0080] Once the relevant text is obtained, similarity matching is performed between the relevant text and the located text to obtain the semantic relevance between the first candidate application and the customer complaint. This semantic relevance can be calculated by training a similarity scoring model. Since the relationship between the target located text and the relevant text of the application is usually weak, the scoring model can be trained with manual supervision and parameter tuning. For example, the BM25 algorithm can be used to calculate the semantic relevance between the first candidate application and the target located text. The BM25 algorithm is a common formula used for relevance scoring; its idea is relatively simple: it mainly calculates the relevance between all words q1, q2...qn in a query and the document, and then sums the scores. Alternatively, the relevant text can be used as the query to calculate the relevance between it and the target located text to obtain the semantic relevance. Another approach is to use an unsupervised method to obtain the vector representations of the target located text and the relevant text, calculate the Jaccard similarity between the vector representations of the target located text and the relevant text, and obtain the semantic relevance based on the Jaccard similarity.

[0081] Please refer to Figure 4, which is a flowchart illustrating an application positioning method provided in an embodiment of this specification. As shown in Figure 4, the method described in this embodiment may include the following steps S302-S304.

[0082] S302, obtain the adjacent nodes of the node corresponding to the second candidate application in the traffic call graph of the first candidate application; wherein, the traffic call graph is constructed based on the traffic call relationship between the first candidate applications;

[0083] S304, the first candidate application corresponding to the adjacent node is confirmed as the third candidate application.

[0084] In one feasible implementation, a traffic call graph is first constructed based on the traffic call relationships between the first candidate applications. These relationships are derived from the call records between applications and include the caller, callee, and the traffic involved. A topological graph is then constructed based on these relationships. In the topological graph, each application is abstracted as a node, and each node has its k-order neighbors. Specifically, the direct neighbors of the current node are its 1-order neighbors, the direct neighbors of the 1-order neighbors are its 2-order neighbors, and so on. The relationships between 1-order neighbors are the strongest, and the higher the order, the weaker the relationship is likely to be. Therefore, the neighboring nodes of the nodes corresponding to the second candidate applications are identified in the traffic call graph. The first candidate applications corresponding to these neighboring nodes are then identified as the third candidate applications.

[0085] Please refer to Figure 5, which is a flowchart illustrating an application positioning method provided in an embodiment of this specification. As shown in Figure 5, the method described in this embodiment may include the following steps S402-S406.

[0086] S402, Based on the traffic call graph of the fourth candidate application, obtain the embedding vector of the first graph node corresponding to the fourth candidate application;

[0087] In the embodiments of this specification, based on the traffic call graph of the fourth candidate application, the embedding vector of the first graph node corresponding to the fourth candidate application is obtained. That is, the high-dimensional adjacency information of a node in the traffic call graph is reduced to a low-dimensional vector representation through a neural network. The node embedding can not only learn the features of each node itself, but also the neighbor relationships.

[0088] S404, Based on the link relationship graph of the fourth candidate application, obtain the second graph node embedding vector corresponding to the fourth candidate application;

[0089] Similarly, the vector representation of the node corresponding to the fourth candidate application in the link relationship graph is obtained to obtain the node embedding vector of the second graph.

[0090] S406, Based on the first graph node embedding vector and the second graph node embedding vector, the transaction correlation degree between the fourth candidate applications is obtained.

[0091] After obtaining the vector representations of the fourth candidate application in the traffic call graph and the link relationship graph, feature fusion can be performed on both. This combines the features learned from both graphs to estimate the degree of association between applications. For example, feature fusion can employ methods such as feature summation (mean, pooling, weighted summation). The embedding representation of each node corresponding to the fourth candidate application is obtained, and the similarity between the embedding representations of each node is calculated pairwise to obtain the similarity between the fourth candidate applications. A higher similarity indicates a higher degree of transactional association between the fourth candidate applications.

[0092] Please refer to Figure 6, which is a flowchart illustrating an application positioning method provided in an embodiment of this specification. As shown in Figure 6, the method described in this embodiment may include the following steps S502-S504.

[0093] S502, obtain the fifth candidate application in the fourth candidate application whose transaction correlation with the second candidate application is greater than the first preset threshold;

[0094] After obtaining the transaction correlation between the fourth candidate applications, a first preset threshold can be set to filter them. Candidate applications among the fourth candidate applications whose transaction correlation with the second candidate application is greater than the first preset threshold can be selected as the fifth candidate application. It is understandable that when some candidate applications among the fourth candidate applications have a transaction correlation with the second candidate application greater than the first preset threshold, it indicates that the application may have more transactional relationships with the second candidate application. Therefore, the target location text may also be related to this application. This method can then be used to obtain relevant applications that cannot be obtained through semantic correlation, avoiding the problem of missed selections.

[0095] S504, based on the second candidate application and the fifth candidate application, the target application of the target location text is obtained.

[0096] Once the fifth candidate application is selected, the target application can be generated directly based on the second and fifth candidate applications. Alternatively, the first-order neighbors (adjacent nodes) of the fifth candidate application can be obtained and used together as the target application. Furthermore, the transaction correlation between candidate applications can be scored. When there are many fifth candidate applications, the second candidate applications can be sorted first based on semantic correlation. Then, according to the sorting order, the transaction correlation scores of the fifth candidate applications corresponding to the second candidate applications can be obtained. The fifth candidate application with the higher transaction correlation score is retained and used together with the second candidate application as the target application.

[0097] Please refer to Figure 7, which is a flowchart illustrating a training method for a transaction correlation model provided in an embodiment of this specification. As shown in Figure 7, the method described in this embodiment may include the following steps S602-S608.

[0098] S602, Obtain the traffic call graph and link relationship graph corresponding to the first candidate application;

[0099] This specification provides a method for training a transaction correlation model in its embodiments. The transaction correlation model is used to calculate the transaction correlation between first candidate applications. The transaction correlation model is a graph model that calculates transaction correlation based on a traffic call graph and a link relationship graph. For example, unsupervised learning is used for model training, directly modeling the traffic call graph and link relationship graph of the first candidate applications.

[0100] S604, input the traffic call graph and the link relationship graph into a graph convolutional neural network to obtain the graph node embedding vector of the first candidate application;

[0101] Specifically, in order to learn the dual features of the traffic call graph and the link relationship graph, the traffic call graph and the link relationship graph are used together as input to the graph convolutional neural network. Through feature extraction, the graph node embedding vector of the first candidate application that integrates the dual features of the traffic call graph and the link relationship graph is obtained.

[0102] S606, calculate the similarity loss value corresponding to the graph node embedding vector using the similarity loss function;

[0103] The similarity loss value between vectors is calculated based on the graph node embedding vector, and the generated graph node embedding vector is constrained by the similarity loss function.

[0104] S608, optimize the network parameters of the graph convolutional neural network based on the similarity loss value, and obtain a pre-trained transaction association model based on the optimized graph convolutional neural network.

[0105] Specifically, after calculating the similarity loss function, it is confirmed whether the similarity loss value meets the training conditions. If not, the network parameters are adjusted and the similarity loss value is calculated using the similarity loss function until the similarity loss value of the model meets the training conditions, thus obtaining the trained transaction association model.

[0106] Please refer to Figure 8, which is a flowchart illustrating a training method for a transaction correlation model provided in an embodiment of this specification. As shown in Figure 8, the method described in this embodiment may include the following steps S702-S710.

[0107] S702, input the traffic call graph into the first graph convolutional neural network to obtain the third graph node embedding vector of the first candidate application;

[0108] In the embodiments of this specification, graph message passing is performed in an independent space, that is, graph convolutional neural networks are used to perform graph message passing in the traffic call graph. The graph nodes in the traffic call graph are represented by the first graph convolutional neural network to obtain the third graph node embedding vector of the first candidate application.

[0109] S704, input the link relationship graph into the second graph convolutional neural network to obtain the fourth graph node embedding vector of the first candidate application;

[0110] Correspondingly, a graph convolutional neural network is used to perform graph message passing in the link space on the link relationship graph. That is, the graph nodes in the link relationship graph are represented by the second graph convolutional neural network to obtain the fourth graph node embedding vector of the first candidate application.

[0111] S706, input the traffic call graph into the third graph convolutional neural network to obtain the fifth graph node embedding vector of the first candidate application;

[0112] The third graph convolutional neural network consists of two parts: one part extracts the graph node representations of the traffic call graph, and the other part extracts the graph node representations of the link relationship graph. The two parts share parameters, i.e., weight sharing, meaning they use the same filter to scan the image once and extract features once.

[0113] Therefore, the traffic call graph is input into the third graph convolutional neural network to obtain the fifth graph node embedding vector of the first candidate application. The graph convolutional neural network with shared parameters is used to perform graph message passing in the traffic-link common space on the traffic call graph.

[0114] S708, input the link relationship graph into the third graph convolutional neural network to obtain the sixth graph node embedding vector of the first candidate application;

[0115] Subsequently, the link relationship graph is input into the third graph convolutional neural network to obtain the sixth graph node embedding vector of the first candidate application. The graph convolutional neural network with shared parameters is used to perform graph message passing of traffic-link common space on the link relationship graph.

[0116] S710, the third graph node embedding vector, the fourth graph node embedding vector, the fifth graph node embedding vector, and the sixth graph node embedding vector are weighted and fused to obtain the seventh graph node embedding vector of the first candidate application.

[0117] The embedding vectors of the third, fourth, fifth, and sixth graph nodes are fused to obtain the fused seventh graph node embedding vector. For example, feature addition can be used for concatenation. The concatenated seventh graph node embedding vector is then used as the first candidate graph node embedding vector for application.

[0118] Please refer to Figure 9, which is a flowchart illustrating a training method for a transaction correlation model provided in an embodiment of this specification. As shown in Figure 9, the method described in this embodiment may include the following steps S802-S808.

[0119] S802, calculate the difference between the third graph node embedding vector and the fifth graph node embedding vector based on the first loss function to obtain the first loss value;

[0120] Understandably, to prevent the common space from being too closely related to a certain space and thus failing to learn common information, the representations learned by the same graph in the independent space and the common space should be constrained to be inconsistent. The Hilbert-Schmidt Independence Criterion (HSIC) is used to constrain the graph node embeddings learned by S702 and S706 to be independent, and the graph node embeddings learned by S704 and S708 to be independent. That is, the fifth graph node embedding vector extracted from the traffic call graph by the third graph convolutional neural network should be different from the third graph node embedding vector extracted from the traffic call graph by the first graph convolutional neural network. Therefore, the difference between the two is calculated as the first loss value of the first loss function.

[0121] S804, calculate the difference between the fourth graph node embedding vector and the sixth graph node embedding vector based on the second loss function to obtain the second loss value;

[0122] Similarly, the node embedding vector of the sixth graph extracted from the link relationship graph using the third graph convolutional neural network should be different from the node embedding vector of the fourth graph extracted from the link relationship graph using the second graph convolutional neural network. Therefore, the difference between the two is calculated as the second loss value of the second loss function. Both the first and second loss functions are independence loss functions.

[0123] S806, calculate the difference between the fifth graph node embedding vector and the sixth graph node embedding vector based on the third loss function to obtain the third loss value;

[0124] Understandably, in order to learn information that is important in both spaces, it is necessary to constrain the representations learned by the two graphs in the common space to be consistent. Therefore, the optimization objective is to make the graph node embeddings obtained on different graphs in the common space as similar as possible. That is, the fifth loss function is the consistency loss function. The third loss value is obtained by calculating the difference between the fifth graph node embedding vector and the sixth graph node embedding vector through the fifth loss function.

[0125] S808, cluster the first candidate applications, and based on the clustering results of the first candidate applications, calculate the difference of the seventh graph node candidate vectors between the first candidate applications belonging to the same class based on the fourth loss function to obtain the fourth loss value.

[0126] Understandably, since there are transactional relationships between applications, the graph node representations of applications belonging to the same transaction category should be more consistent. Therefore, the optimization goal is to ensure that strongly connected nodes (belonging to the same transaction category) learn representations that are as similar as possible.

[0127] Specifically, since there is no unified and complete application category definition across the entire site, a graph cut-based clustering algorithm is used to perform soft clustering of applications instead of labels. It is assumed that within the traffic call set in the traffic call graph, transactions that can be clustered into one group are considered a single transaction. Soft clustering means that a sample can be divided into multiple clusters. Then, the minimum graph cut algorithm is used to divide the vertices of the graph into two disjoint subsets connected by at least one edge, thus performing clustering. After grouping closely related applications into the same category through traffic calls between applications, the difference between applications in the same category can be calculated. This fourth loss function serves as the clustering loss constraint. The difference between the candidate vectors of the seventh graph nodes corresponding to the first candidate application belonging to the same category is calculated using the fourth loss function to obtain the fourth loss value.

[0128] Referring to Figure 10, a schematic diagram of a training method for a transaction correlation model provided in this embodiment of the specification is shown. As shown in Figure 10, the input to the transaction correlation model includes the traffic call graph, the link relationship graph, and the vector form of node description keywords of the first candidate application. It can be understood that since the traffic call graph and the link relationship graph represent a candidate application with each node, the corresponding node descriptions also need to be input to confirm what application each node represents. Then, the traffic call graph is input into the first graph convolutional neural network to obtain the node embedding vector Z of the third graph. t The link relationship graph is input into the third graph's convolutional neural network to obtain the node embedding vector Z of the fourth graph. l The traffic call graph is input into the third graph convolutional neural network to obtain the node embedding vector Z of the fifth graph. ct Input the traffic call graph into the third graph convolutional neural network to obtain the node embedding vector Z of the sixth graph. cl Z is calculated using the first loss function.t and Z ct The difference is used to calculate Z using the second loss function. l and Z cl The difference can be calculated using the same L function, since both the first and second loss functions are independent loss functions. d Indicates that Z is calculated using the third loss function. ct and Z cl The difference is calculated using the fourth loss function; the difference between the seventh graph node candidate vectors after weighted fusion is calculated using the graph cut algorithm. Referring to Figure 11, which illustrates a clustering method for training a transaction correlation model according to an embodiment of this specification, the graph nodes are first soft-clustered, then minimum graph cut is performed to obtain the clustering results. The difference is calculated for the first candidates of the same class. Since the third and fourth loss functions are required to be consistent, the same L... c express.

[0129] Optionally, the method described in the embodiments of this specification may include the following steps S902-S910.

[0130] S902, if the first loss value is less than the second preset threshold, optimize the first network parameters of the first graph convolutional neural network and the third graph convolutional neural network, and proceed to the step of inputting the traffic call graph and the link relationship graph into the graph convolutional neural network to obtain the graph node embedding vector of the first candidate application, until the first loss value is greater than the second preset threshold;

[0131] Understandably, if the first loss value is less than the second preset threshold, it indicates that the representations learned for the traffic call graph in the independent space and the common space are too consistent. In this case, the network parameters of the first and third graph convolutional neural networks need to be adjusted to ensure that the node embedding vectors of the third and fifth graphs satisfy the independence principle, meaning the first loss value is greater than the second preset threshold. It should be noted that the first network parameters include the network parameters of one or more graph convolutional neural networks in the first and third graphs, and specific optimization and adjustment are required based on the magnitude of the first loss value.

[0132] S904, if the second loss value is less than the third preset threshold, optimize the second network parameters of the second graph convolutional neural network and the third graph convolutional neural network, and proceed to the step of inputting the traffic call graph and the link relationship graph into the graph convolutional neural network to obtain the graph node embedding vector of the first candidate application, until the second loss value is greater than the third preset threshold;

[0133] Correspondingly, if the second loss value is less than the third preset threshold, it indicates that the representations learned by the link relationship graph in the independent space and the common space are too consistent. Therefore, it is necessary to adjust the network parameters of the second and third graph convolutional neural networks to ensure that the embedding vectors of the fourth and fifth / sixth nodes satisfy the independence principle, i.e., that the second loss value is greater than the third preset threshold. The second network parameters include the network parameters of one or more graph convolutional neural networks in the second and third graph convolutional neural networks.

[0134] S906, if the third loss value is less than the fourth preset threshold, then optimize the third network parameters of the third graph convolutional neural network, and proceed to the step of inputting the traffic call graph and the link relationship graph into the graph convolutional neural network to obtain the graph node embedding vector of the first candidate application, until the third loss value is greater than the fourth preset threshold.

[0135] Understandably, if the third loss value is less than the fourth preset threshold, it indicates that the embedding vector of the sixth graph node extracted from the link relationship graph by the third graph convolutional neural network is too similar to the embedding vector of the fourth graph node extracted from the link relationship graph by the second graph convolutional neural network. Therefore, the third network parameters of the third graph convolutional neural network should be adjusted so that the third loss value is greater than the fourth preset threshold. Here, the third network parameters refer to the network parameters of the third graph convolutional neural network.

[0136] S908, if the fourth loss value is greater than the fifth preset threshold, then optimize the fourth network parameters of the first graph convolutional neural network, the second graph convolutional neural network and the third graph convolutional neural network, and proceed to the step of inputting the traffic call graph and the link relationship graph into the graph convolutional neural network to obtain the graph node embedding vector of the first candidate application, until the fourth loss value is less than the fifth preset threshold;

[0137] If the fourth loss value is greater than the fifth preset threshold, it indicates that the graph node embedding vectors belonging to the same class in the fused seventh graph node embedding vector are not sufficiently approximate. Therefore, it is necessary to constrain the fourth loss value between graph node embedding vectors of the same class to be less than the fifth preset threshold. This is because the seventh graph node embedding vector is a fusion of the third, fourth, fifth, and sixth graph node embedding vectors. Therefore, during parameter tuning, optimization is required for the first, second, and third graph convolutional neural networks. The fourth network parameter refers to one or more of the network parameters corresponding to the first, second, and third graph convolutional neural networks, and can be a collective term including the network parameters of the first, second, and third graph convolutional neural networks.

[0138] It should be noted that the parameter tuning order of steps S902-S908 is not limited. For example, the network parameters of one graph convolutional neural network can be fixed while the parameters of the other two graph convolutional neural networks are adjusted.

[0139] S910, based on the first network parameters, the second network parameters, the third network parameters and the fourth network parameters, the optimized first graph convolutional neural network, the second graph convolutional neural network and the third graph convolutional neural network are obtained, and a pre-trained transaction correlation model is obtained.

[0140] After adjusting the network parameters of the first, second, and third convolutional neural networks, a transaction correlation model can be obtained. When any one or more candidate applications from the first candidate applications are input, the transaction correlation between them can be obtained.

[0141] It should be noted that the above preset thresholds can all be determined according to actual needs.

[0142] In the embodiments of this specification, a transaction correlation model based on a link relationship graph and a traffic call graph is constructed. Through graph message passing in both independent and common spaces, and constrained by a similarity loss function, an unsupervised pre-trained transaction correlation model is obtained. This model allows for the extraction of graph node embedding vectors from the graph convolutional neural network within the transaction correlation model, and the calculation of transaction correlation based on these vectors. Because text matching methods only consider semantic correlation and ignore many key transaction correlations, they are ineffective in addressing the problem of weak matching between target text and application, resulting in low accuracy. However, by training the transaction correlation model, preliminary results can now be provided for applications that were previously unrecommended. Furthermore, by aggregating and evaluating the semantic correlation between the application and the target text, as well as the transaction correlation between applications, the problem of weak matching between the target application and the root cause application is solved, thereby improving the accuracy of application location.

[0143] The application positioning device provided in the embodiments of this specification will now be described in detail with reference to Figure 12. It should be noted that the application positioning device in Figure 12 is used to execute the methods of the embodiments shown in Figures 2-11 of this specification. For ease of explanation, only the parts related to the embodiments of this specification are shown. For specific technical details not disclosed, please refer to the embodiments shown in Figures 2-11 of this specification.

[0144] Please refer to Figure 12, which shows a schematic diagram of the structure of an application positioning device provided in an exemplary embodiment of this specification. This application positioning device can be implemented as all or part of a device through software, hardware, or a combination of both. The device 1 includes a semantic correlation module 11, a first confirmation module 12, a second confirmation module 13, a transaction correlation module 14, and a positioning module 15.

[0145] The semantic relevance module 11 is used to obtain the semantic relevance between the first candidate application and the target location text based on the target location text and the relevant text of the first candidate application.

[0146] The first confirmation module 12 is used to confirm the second candidate application in the first candidate application based on the semantic relevance.

[0147] The second confirmation module 13 is used to obtain a third candidate application associated with the second candidate application from the first candidate application;

[0148] The transaction correlation module 14 is used to obtain the transaction correlation between the fourth candidate applications based on the traffic call graph and link relationship graph of the fourth candidate application.

[0149] The positioning module 15 is used to obtain the target application of the target positioning text from the fourth candidate application based on the transaction correlation degree.

[0150] Optionally, the semantic relevance module 11 is specifically used to obtain descriptive text data related to the first candidate application;

[0151] The descriptive text data is segmented to obtain keyword data for the first candidate application, and the keyword data is used as the relevant text.

[0152] Based on the matching of the relevant text and the target location text, the semantic correlation between the first candidate application and the target location text is obtained.

[0153] Optionally, the second confirmation module 13 is specifically used to obtain the adjacent nodes of the node corresponding to the second candidate application in the traffic call graph of the first candidate application; wherein, the traffic call graph is constructed based on the traffic call relationship between the first candidate applications;

[0154] The first candidate application corresponding to the adjacent node is confirmed as the third candidate application.

[0155] Optionally, the transaction correlation module 14 is specifically used to obtain the first graph node embedding vector corresponding to the fourth candidate application based on the traffic call graph of the fourth candidate application.

[0156] Based on the link relationship graph of the fourth candidate application, the embedding vector of the second graph node corresponding to the fourth candidate application is obtained;

[0157] Based on the first graph node embedding vector and the second graph node embedding vector, the transaction correlation degree between the fourth candidate applications is obtained.

[0158] Optionally, the positioning module 15 is specifically used to obtain a fifth candidate application in the fourth candidate application whose transaction correlation with the second candidate application is greater than a first preset threshold.

[0159] Based on the second candidate application and the fifth candidate application, the target application of the target location text is obtained.

[0160] Optionally, please refer to Figure 13, which is a structural schematic diagram of an application positioning device provided in an embodiment of this specification. As shown in Figure 13, the device 2 includes an acquisition module 21, an embedding vector calculation module 22, a loss function calculation module 23, and an optimization module 24.

[0161] Module 21 is used to obtain the traffic call graph and link relationship graph corresponding to the first candidate application;

[0162] Embedded vector calculation module 22 is used to input the traffic call graph and the link relationship graph into a graph convolutional neural network to obtain the graph node embedding vector of the first candidate application;

[0163] Loss function calculation module 23 is used to calculate the similarity loss value corresponding to the graph node embedding vector through the similarity loss function;

[0164] The optimization module 24 is used to optimize the network parameters of the graph convolutional neural network based on the similarity loss value, and to obtain a pre-trained transaction correlation model based on the optimized graph convolutional neural network.

[0165] Optionally, the embedding vector calculation module 22 is specifically used to input the traffic call graph into the first graph convolutional neural network to obtain the third graph node embedding vector of the first candidate application;

[0166] The link relationship graph is input into the second graph convolutional neural network to obtain the fourth graph node embedding vector of the first candidate application;

[0167] The traffic call graph is input into the third graph convolutional neural network to obtain the fifth graph node embedding vector of the first candidate application;

[0168] The link relationship graph is input into the third graph convolutional neural network to obtain the sixth graph node embedding vector of the first candidate application;

[0169] The third, fourth, fifth, and sixth graph node embedding vectors are weighted and fused to obtain the seventh graph node embedding vector of the first candidate application.

[0170] Optionally, the loss function calculation module 23 is specifically used to calculate the difference between the third graph node embedding vector and the fifth graph node embedding vector based on the first loss function to obtain the first loss value;

[0171] The second loss value is obtained by calculating the difference between the fourth graph node embedding vector and the sixth graph node embedding vector based on the second loss function.

[0172] The difference between the fifth graph node embedding vector and the sixth graph node embedding vector is calculated based on the third loss function to obtain the third loss value;

[0173] Cluster the first candidate applications, and based on the clustering results of the first candidate applications, calculate the difference of the candidate vectors of the seventh graph nodes between the first candidate applications belonging to the same class based on the fourth loss function to obtain the fourth loss value.

[0174] Optionally, the optimization module 24 is specifically used to optimize the first network parameters of the first graph convolutional neural network and the third graph convolutional neural network if the first loss value is less than the second preset threshold, and then proceed to the step of inputting the traffic call graph and the link relationship graph into the graph convolutional neural network to obtain the graph node embedding vector of the first candidate application, until the first loss value is greater than the second preset threshold.

[0175] If the second loss value is less than the third preset threshold, then optimize the second network parameters of the second graph convolutional neural network and the third graph convolutional neural network, and proceed to the step of inputting the traffic call graph and the link relationship graph into the graph convolutional neural network to obtain the graph node embedding vector of the first candidate application, until the second loss value is greater than the third preset threshold;

[0176] If the third loss value is less than the fourth preset threshold, then optimize the third network parameters of the third graph convolutional neural network, and proceed to the step of inputting the traffic call graph and the link relationship graph into the graph convolutional neural network to obtain the graph node embedding vector of the first candidate application, until the third loss value is greater than the fourth preset threshold.

[0177] If the fourth loss value is greater than the fifth preset threshold, then optimize the fourth network parameters of the first graph convolutional neural network, the second graph convolutional neural network and the third graph convolutional neural network, and proceed to the step of inputting the traffic call graph and the link relationship graph into the graph convolutional neural network to obtain the graph node embedding vector of the first candidate application, until the fourth loss value is less than the fifth preset threshold.

[0178] Based on the first network parameters, the second network parameters, the third network parameters, and the fourth network parameters, optimized first graph convolutional neural networks, second graph convolutional neural networks, and third graph convolutional neural networks are obtained, resulting in a pre-trained transaction correlation model.

[0179] It should be noted that the application positioning device provided in the above embodiments is only illustrated by the division of the above functional modules when executing the application positioning method. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the application positioning device and the application positioning method embodiments provided in the above embodiments belong to the same concept, and the implementation process can be found in the method embodiments, which will not be repeated here.

[0180] The embodiment numbers in this specification are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0181] This specification also provides a computer storage medium storing a computer program. When the computer program is executed by a processor, it implements the application location method as shown in the embodiments of Figures 2-12 above. For the specific execution process, please refer to the detailed description of the embodiments shown in Figures 2-12, which will not be repeated here.

[0182] Please refer to Figure 14, which shows a schematic diagram of the structure of an electronic device provided in an exemplary embodiment of this specification. The electronic device in this specification may include one or more of the following components: a processor 110, a memory 120, an input device 130, an output device 140, and a bus 150. The processor 110, memory 120, input device 130, and output device 140 may be connected to each other via the bus 150.

[0183] Processor 110 may include one or more processing cores. Processor 110 connects to various parts of the electronic device using various interfaces and lines, and executes various functions of terminal 100 and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 120, and by calling data stored in memory 120. Optionally, processor 110 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). Processor 110 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user page, and applications; the GPU is responsible for rendering and drawing the displayed content; and the modem handles wireless communication. It is understood that the modem may also not be integrated into processor 110 and may be implemented separately using a communication chip.

[0184] The memory 120 may include random access memory (RAM) or read-only memory (ROM). Optionally, the memory 120 may include non-transitory computer-readable storage medium. The memory 120 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 120 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for implementing at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the various method embodiments described above, etc. The operating system may be the Android system, including systems deeply developed based on the Android system, the iOS system developed by Apple Inc., including systems deeply developed based on the iOS system, or other systems.

[0185] The memory 120 can be divided into operating system space and user space. The operating system runs in the operating system space, while native and third-party applications run in user space. To ensure that different third-party applications can achieve good running performance, the operating system allocates corresponding system resources for each application. However, different application scenarios within the same third-party application have different requirements for system resources. For example, in local resource loading scenarios, third-party applications have high requirements for disk read speed; in animation rendering scenarios, third-party applications have high requirements for GPU performance. Since the operating system and third-party applications are independent of each other, the operating system often cannot promptly perceive the current application scenario of a third-party application, resulting in the operating system's inability to adapt system resources accordingly.

[0186] In order for the operating system to distinguish the specific application scenarios of third-party applications, it is necessary to establish data communication between the third-party applications and the operating system. This would allow the operating system to obtain the current scenario information of the third-party applications at any time, and then perform targeted system resource adaptation based on the current scenario.

[0187] The input device 130 is used to receive input instructions or data, and includes, but is not limited to, a keyboard, mouse, camera, microphone, or touch device. The output device 140 is used to output instructions or data, and includes, but is not limited to, a display device and a speaker. In one example, the input device 130 and the output device 140 can be combined, and the input device 130 and the output device 140 can be a touch display screen.

[0188] The touch display screen can be designed as a full-screen, curved screen, or irregularly shaped screen. It can also be designed as a combination of a full-screen and a curved screen, or a combination of an irregularly shaped screen and a curved screen; however, this specification does not limit the specific design of the embodiments.

[0189] In addition, those skilled in the art will understand that the structure of the electronic device shown in the above figures does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown, or combine certain components, or have different component arrangements. For example, the electronic device may also include radio frequency circuits, input units, sensors, audio circuits, WiFi modules, power supplies, Bluetooth modules, etc., which will not be described in detail here.

[0190] In the electronic device shown in Figure 14, the processor 110 can be used to call the computer program stored in the memory 120 and specifically perform the following operations:

[0191] Based on the target location text and the relevant text of the first candidate application, the semantic correlation between the first candidate application and the target location text is obtained;

[0192] Based on the semantic relevance, a second candidate application is identified from the first candidate applications;

[0193] Obtain a third candidate application associated with the second candidate application from the first candidate application;

[0194] Based on the traffic call graph and link relationship graph of the fourth candidate application, the transaction correlation degree between the fourth candidate applications is obtained. The fourth candidate application includes the second candidate application and the third candidate application.

[0195] The target application for the target location text is obtained from the fourth candidate application based on the transaction correlation.

[0196] In one embodiment, when the processor 110 executes the semantic correlation between the first candidate application and the customer complaint based on the target location text and the relevant text of the first candidate application, it specifically performs the following operations:

[0197] Obtain descriptive text data related to the first candidate application;

[0198] The descriptive text data is segmented to obtain keyword data for the first candidate application, and the keyword data is used as the relevant text.

[0199] Based on the matching of the relevant text and the target location text, the semantic correlation between the first candidate application and the target location text is obtained.

[0200] In one embodiment, when the processor 110 executes the operation of retrieving a third candidate application associated with the second candidate application from the first candidate application, it specifically performs the following operations:

[0201] Obtain the adjacent nodes of the corresponding node of the second candidate application in the traffic call graph of the first candidate application; wherein, the traffic call graph is constructed based on the traffic call relationship between the first candidate applications;

[0202] The first candidate application corresponding to the adjacent node is confirmed as the third candidate application.

[0203] In one embodiment, when the processor 110 executes the traffic call graph and link relationship graph based on the fourth candidate application to obtain the transaction correlation between the fourth candidate applications, it specifically performs the following operations:

[0204] Based on the traffic call graph of the fourth candidate application, the embedding vector of the first graph node corresponding to the fourth candidate application is obtained;

[0205] Based on the link relationship graph of the fourth candidate application, the embedding vector of the second graph node corresponding to the fourth candidate application is obtained;

[0206] Based on the first graph node embedding vector and the second graph node embedding vector, the transaction correlation degree between the fourth candidate applications is obtained.

[0207] In one embodiment, when the processor 110 executes a target application that obtains the target location text from the fourth candidate application based on the transaction correlation, it specifically performs the following operations:

[0208] Obtain a fifth candidate application from the fourth candidate application whose transaction correlation with the second candidate application is greater than a first preset threshold;

[0209] Based on the second candidate application and the fifth candidate application, the target application of the target location text is obtained.

[0210] Optionally, the processor 110 can be used to call a computer program stored in the memory 120 and specifically perform the following operations:

[0211] Obtain the traffic call graph and link relationship graph corresponding to the first candidate application;

[0212] The traffic call graph and the link relationship graph are input into a graph convolutional neural network to obtain the graph node embedding vector of the first candidate application;

[0213] The similarity loss value corresponding to the graph node embedding vector is calculated using a similarity loss function;

[0214] The network parameters of the graph convolutional neural network are optimized based on the similarity loss value, and a pre-trained transaction correlation model is obtained based on the optimized graph convolutional neural network.

[0215] In one embodiment, when the processor 110 executes the operation of inputting the traffic call graph and the link relationship graph into a graph convolutional neural network to obtain the graph node embedding vector of the first candidate application, it specifically performs the following operations:

[0216] The traffic call graph is input into the first graph convolutional neural network to obtain the third graph node embedding vector of the first candidate application;

[0217] The link relationship graph is input into the second graph convolutional neural network to obtain the fourth graph node embedding vector of the first candidate application;

[0218] The traffic call graph is input into the third graph convolutional neural network to obtain the fifth graph node embedding vector of the first candidate application;

[0219] The link relationship graph is input into the third graph convolutional neural network to obtain the sixth graph node embedding vector of the first candidate application;

[0220] The third, fourth, fifth, and sixth graph node embedding vectors are weighted and fused to obtain the seventh graph node embedding vector of the first candidate application.

[0221] In one embodiment, when the processor 110 calculates the similarity loss value corresponding to the graph node embedding vector using a similarity loss function, it specifically performs the following operations:

[0222] The first loss value is obtained by calculating the difference between the embedding vector of the third graph node and the embedding vector of the fifth graph based on the first loss function.

[0223] The second loss value is obtained by calculating the difference between the fourth graph node embedding vector and the sixth graph node embedding vector based on the second loss function.

[0224] The difference between the fifth graph node embedding vector and the sixth graph node embedding vector is calculated based on the third loss function to obtain the third loss value;

[0225] Cluster the first candidate applications, and based on the clustering results of the first candidate applications, calculate the difference of the candidate vectors of the seventh graph nodes between the first candidate applications belonging to the same class based on the fourth loss function to obtain the fourth loss value.

[0226] In one embodiment, when the processor 110 optimizes the network parameters of the graph convolutional neural network based on the similarity loss value and obtains a pre-trained transaction correlation model based on the optimized graph convolutional neural network, it specifically performs the following operations:

[0227] If the first loss value is less than the second preset threshold, then optimize the first network parameters of the first graph convolutional neural network and the third graph convolutional neural network, and proceed to the step of inputting the traffic call graph and the link relationship graph into the graph convolutional neural network to obtain the graph node embedding vector of the first candidate application, until the first loss value is greater than the second preset threshold.

[0228] If the second loss value is less than the third preset threshold, then optimize the second network parameters of the second graph convolutional neural network and the third graph convolutional neural network, and proceed to the step of inputting the traffic call graph and the link relationship graph into the graph convolutional neural network to obtain the graph node embedding vector of the first candidate application, until the second loss value is greater than the third preset threshold;

[0229] If the third loss value is less than the fourth preset threshold, then optimize the third network parameters of the third graph convolutional neural network, and proceed to the step of inputting the traffic call graph and the link relationship graph into the graph convolutional neural network to obtain the graph node embedding vector of the first candidate application, until the third loss value is greater than the fourth preset threshold.

[0230] If the fourth loss value is greater than the fifth preset threshold, then optimize the fourth network parameters of the first graph convolutional neural network, the second graph convolutional neural network and the third graph convolutional neural network, and proceed to the step of inputting the traffic call graph and the link relationship graph into the graph convolutional neural network to obtain the graph node embedding vector of the first candidate application, until the fourth loss value is less than the fifth preset threshold.

[0231] Based on the first network parameters, the second network parameters, the third network parameters, and the fourth network parameters, optimized first graph convolutional neural networks, second graph convolutional neural networks, and third graph convolutional neural networks are obtained, resulting in a pre-trained transaction correlation model.

[0232] In the embodiments of this specification, the semantic correlation between the first candidate application and the target location text is obtained by using the relevant text of the first candidate application based on the target location text and the first candidate application. Based on the semantic correlation, the second candidate application is confirmed in the first candidate application. The third candidate application associated with the second candidate application is obtained from the first candidate application. The transaction correlation between the fourth candidate applications is obtained based on the traffic call graph and link relationship graph of the fourth candidate application. The fourth candidate application includes the second candidate application and the third candidate application. Based on the transaction correlation, the target application of the target location text is obtained from the fourth candidate application. By calculating the semantic relevance between the target location text and candidate applications, as well as the transactional relevance between candidate applications, the semantic relevance between the application and the target location text and the transactional relevance between applications are aggregated and evaluated. This ensures that the selected target applications encompass both semantic and transactional relevance information, solving the problem of weak matching between the target location text and the root cause application text, and greatly improving the accuracy of application location. Furthermore, by constructing a transactional relevance model based on a link relationship graph and a traffic call graph, and through graph message passing in both independent and common spaces, constrained by a similarity loss function, an unsupervised pre-trained transactional relevance model is obtained. This allows the extraction of graph node embedding vectors from the graph convolutional neural network in the transactional relevance model. Transactional relevance is then calculated based on these graph node embedding vectors, providing preliminary results for applications that were previously unrecommended. Finally, by aggregating and evaluating the semantic relevance between the application and the target location text and the transactional relevance between applications, the problem of weak matching between the target location application and the root cause application is solved, thereby improving the accuracy of application location.

[0233] Additionally, this specification provides a computer program product that includes a computer program that, when executed by a processor of an electronic device, enables the processor to at least implement the methods provided in the embodiments shown in Figures 2 to 12 above.

[0234] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0235] The above-disclosed embodiments are merely preferred embodiments of this specification and should not be construed as limiting the scope of this specification. Therefore, any equivalent variations made in accordance with the claims of this specification shall still fall within the scope of this specification.

Claims

1. An application location method, comprising: Based on the target location text and the relevant text of the first candidate application, the semantic correlation between the first candidate application and the target location text is obtained; The target location text is extracted based on the user's complaint information; Based on the semantic relevance, a second candidate application is identified from the first candidate applications; Obtain a third candidate application associated with the second candidate application from the first candidate application; Based on the traffic call graph and link relationship graph of the fourth candidate application, the transaction correlation degree between the fourth candidate applications is obtained. The fourth candidate application includes the second candidate application and the third candidate application. The traffic call graph of the fourth candidate application is constructed based on the traffic call relationship between the fourth candidate applications; The traffic call relationship is obtained based on the call records between the fourth candidate applications; The link relationship diagram of the fourth candidate application is constructed based on the connections of the fourth candidate application on a predefined transaction link; Obtain a fifth candidate application from the fourth candidate application whose transaction correlation with the second candidate application is greater than a first preset threshold; Based on the second candidate application and the fifth candidate application, the target application of the target location text is obtained.

2. The method as described in claim 1, wherein obtaining the semantic correlation degree between the first candidate application and the target location text based on the target location text and the related text of the first candidate application includes: Obtain descriptive text data related to the first candidate application; The descriptive text data is segmented to obtain keyword data for the first candidate application, and the keyword data is used as the relevant text. Based on the matching of the relevant text and the target location text, the semantic correlation between the first candidate application and the target location text is obtained.

3. The method of claim 1, wherein obtaining the third candidate application associated with the second candidate application from the first candidate application comprises: Obtain the adjacent nodes of the node corresponding to the second candidate application in the traffic call graph of the first candidate application; wherein, the traffic call graph is constructed based on the traffic call relationship between the first candidate applications; and confirm the first candidate application corresponding to the adjacent node as the third candidate application.

4. The method as described in claim 1, wherein obtaining the transaction correlation degree between the fourth candidate applications based on the traffic call graph and link relationship graph of the fourth candidate application includes: Based on the traffic call graph of the fourth candidate application, the embedding vector of the first graph node corresponding to the fourth candidate application is obtained; Based on the link relationship graph of the fourth candidate application, the second graph node embedding vector corresponding to the fourth candidate application is obtained; based on the first graph node embedding vector and the second graph node embedding vector, the transaction correlation degree between the fourth candidate applications is obtained.

5. A training method for a transaction correlation model, comprising: Obtain the traffic call graph and link relationship graph corresponding to the first candidate application; The traffic call graph and the link relationship graph are input into a graph convolutional neural network to obtain the graph node embedding vector of the first candidate application; the traffic call graph of the first candidate application is constructed based on the traffic call relationship between the first candidate applications. The traffic call relationship is obtained based on the call records between the first candidate applications; The link relationship diagram of the first candidate application is constructed based on the connections of the first candidate application on a predefined transaction link; The similarity loss value corresponding to the graph node embedding vector is calculated using a similarity loss function; The network parameters of the graph convolutional neural network are optimized based on the similarity loss value, and a pre-trained transaction correlation model is obtained based on the optimized graph convolutional neural network. The transaction correlation model is used to determine the transaction correlation between the fourth candidate applications based on the traffic call graph and link relationship graph of the fourth candidate application. The transaction correlation degree is used to determine a fifth candidate application from the fourth candidate application whose transaction correlation degree with the second candidate application is greater than a first preset threshold; the second candidate application is confirmed from the first candidate application based on the semantic correlation degree of the target location text and the relevant text of the first candidate application. The target location text is extracted based on the user's complaint information; the fifth candidate application and the second candidate application are the target applications of the target location text.

6. The method of claim 5, wherein the graph convolutional neural network comprises a first graph convolutional neural network, a second graph convolutional neural network, and a third graph convolutional neural network; the step of inputting the traffic call graph and the link relationship graph into the graph convolutional neural network to obtain the graph node embedding vector of the first candidate application comprises: The traffic call graph is input into the first graph convolutional neural network to obtain the third graph node embedding vector of the first candidate application; The link relationship graph is input into a second graph convolutional neural network to obtain the fourth graph node embedding vector of the first candidate application; the traffic call graph is input into a third graph convolutional neural network to obtain the fifth graph node embedding vector of the first candidate application; the link relationship graph is input into a third graph convolutional neural network to obtain the sixth graph node embedding vector of the first candidate application; the third graph node embedding vector, the fourth graph node embedding vector, the fifth graph node embedding vector, and the sixth graph node embedding vector are weighted and fused to obtain the seventh graph node embedding vector of the first candidate application.

7. The method as described in claim 6, wherein the similarity loss function includes a first loss function, a second loss function, a third loss function, and a fourth loss function; and the similarity loss value includes a first loss value, a second loss value, a third loss value, and a fourth loss value; The step of calculating the similarity loss value corresponding to the graph node embedding vector using a similarity loss function includes: calculating the difference between the third graph node embedding vector and the fifth graph node embedding vector based on a first loss function to obtain a first loss value; calculating the difference between the fourth graph node embedding vector and the sixth graph node embedding vector based on a second loss function to obtain a second loss value; calculating the difference between the fifth graph node embedding vector and the sixth graph node embedding vector based on a third loss function to obtain a third loss value; clustering the first candidate applications; and based on the clustering results of the first candidate applications, calculating the difference between the seventh graph node candidate vectors of the first candidate applications belonging to the same class based on a fourth loss function to obtain a fourth loss value.

8. The method of claim 7, wherein optimizing the network parameters of the graph convolutional neural network based on the similarity loss value, and obtaining a pre-trained transaction association model based on the optimized graph convolutional neural network, comprises: If the first loss value is less than the second preset threshold, then the first network parameters of the first graph convolutional neural network and the third graph convolutional neural network are optimized, and the process proceeds to the step of inputting the traffic call graph and the link relationship graph into the graph convolutional neural network to obtain the graph node embedding vector of the first candidate application, until the first loss value is greater than the second preset threshold; if the second loss value is less than the third preset threshold, then the second network parameters of the second graph convolutional neural network and the third graph convolutional neural network are optimized, and the process proceeds to the step of inputting the traffic call graph and the link relationship graph into the graph convolutional neural network to obtain the graph node embedding vector of the first candidate application, until the second loss value is greater than the third preset threshold; if the third loss value is less than the fourth preset threshold, then the third network parameters of the third graph convolutional neural network are optimized, and the process proceeds to the step of inputting the traffic call graph and the link relationship graph into the graph convolutional neural network to obtain the graph node embedding vector of the first candidate application, until the second loss value is greater than the third preset threshold; if the third loss value is less than the fourth preset threshold, then ... process proceed The process involves inputting the traffic call graph and the link relationship graph into a graph convolutional neural network to obtain the graph node embedding vector of the first candidate application, until the third loss value is greater than a fourth preset threshold. If the fourth loss value is greater than a fifth preset threshold, the fourth network parameters of the first, second, and third graph convolutional neural networks are optimized, and the process returns to the step of inputting the traffic call graph and the link relationship graph into a graph convolutional neural network to obtain the graph node embedding vector of the first candidate application, until the fourth loss value is less than the fifth preset threshold. Based on the first, second, third, and fourth network parameters, the optimized first, second, and third graph convolutional neural networks are obtained, resulting in a pre-trained transaction correlation model.

9. An application positioning device, the device comprising: The semantic relevance module is used to obtain the semantic relevance between the first candidate application and the target location text based on the target location text and the relevant text of the first candidate application. The target location text is extracted based on the user's complaint information; The first confirmation module is used to confirm the second candidate application in the first candidate application based on the semantic relevance. The second confirmation module is used to obtain a third candidate application associated with the second candidate application from the first candidate application; The transaction correlation module is used to obtain the transaction correlation between the fourth candidate applications based on the traffic call graph and link relationship graph of the fourth candidate application. The traffic call graph of the fourth candidate application is constructed based on the traffic call relationship between the fourth candidate applications; The traffic call relationship is obtained based on the call records between the fourth candidate applications; The link relationship diagram of the fourth candidate application is constructed based on the connections of the fourth candidate application on a predefined transaction link; The positioning module is used to obtain a fifth candidate application in the fourth candidate application whose transaction correlation with the second candidate application is greater than a first preset threshold. Based on the second candidate application and the fifth candidate application, the target application of the target location text is obtained.

10. An application positioning device, the device comprising: The acquisition module is used to acquire the traffic call graph and link relationship graph corresponding to the first candidate application; The traffic call graph of the first candidate application is constructed based on the traffic call relationship between the first candidate applications; The traffic call relationship is obtained based on the call records between the first candidate applications; The link relationship diagram of the first candidate application is constructed based on the connections of the first candidate application on a predefined transaction link; The embedding vector calculation module is used to input the traffic call graph and the link relationship graph into the graph convolutional neural network to obtain the graph node embedding vector of the first candidate application; The loss function calculation module is used to calculate the similarity loss value corresponding to the graph node embedding vector through the similarity loss function; the optimization module is used to optimize the network parameters of the graph convolutional neural network based on the similarity loss value, and obtain a pre-trained transaction association model based on the optimized graph convolutional neural network. The transaction correlation model is used to determine the transaction correlation between the fourth candidate applications based on the traffic call graph and link relationship graph of the fourth candidate application. The transaction correlation degree is used to determine a fifth candidate application from the fourth candidate application whose transaction correlation degree with the second candidate application is greater than a first preset threshold; the second candidate application is confirmed from the first candidate application based on the semantic correlation degree of the target location text and the relevant text of the first candidate application. The target location text is extracted based on the user's complaint information; the fifth candidate application and the second candidate application are the target applications of the target location text.

11. An electronic device, comprising: Processor and memory; The memory stores a computer program adapted to be loaded by the processor and to execute the steps of the method as described in any one of claims 1 to 8.

12. A computer storage medium storing a computer program that, when executed by a processor, implements the steps of the method as claimed in any one of claims 1 to 8.

13. A computer program product, comprising: A computer program, when executed by a processor of an electronic device, causes the processor to perform the method as described in any one of claims 1 to 8.

Images