Mobile application platform risk identification method and related device

By determining the adaptive feature extraction and semantic dependency analysis of user social network graphs and message data on mobile application platforms, combined with GNN and DPCNN models, the problem of high misjudgment rate of sensitive text recognition and malicious user prediction in the prior art is solved. Improved the accuracy of risk identification and model efficiency.

CN119089240APending Publication Date: 2024-12-06STATE GRID INFORMATION & TELECOMM GRP CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411076407.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-07
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The prior art has a high misjudgment rate in sensitive text recognition and malicious user prediction, and graph neural networks are difficult to learn typical features of malicious users, resulting in poor prediction results.

Method used

By determining the adaptive feature extraction of user social network graphs and message data, semantic dependency analysis is performed, semantic text graphs are constructed, and feature splicing is combined with GNN and DPCNN models to realize the identification of sensitive texts and malicious users.

Benefits of technology

It improves the accuracy of risk identification, reduces the complexity of the model and the consumption of computing resources, and enhances the learning ability of malicious user behavior characteristics.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119089240A_ABST
    Figure CN119089240A_ABST
Patent Text Reader

Abstract

The invention provides a mobile application platform risk identification method and a related device, and the method comprises the steps: carrying out the adaptive breadth feature extraction and the adaptive depth feature extraction of a user social network graph, and obtaining a first feature; performing feature extraction on the user message data to obtain a second feature; performing semantic dependency analysis on the user message data to obtain semantic text nodes with a semantic dependency relationship, performing construction based on the semantic text nodes to obtain a semantic text graph, and performing feature extraction on the semantic text graph to obtain third features; splicing the second feature and the third feature to obtain a first spliced feature, and splicing the first feature and the first spliced feature to obtain a second spliced feature; and performing sensitive text recognition based on the second splicing feature to obtain a sensitive text recognition result, and performing malicious user recognition based on the second splicing feature to obtain a malicious user recognition result. The behavior characteristics of the malicious users are extracted through joint learning, and the risk identification accuracy is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of electronic computers and Internet technology, and in particular to a mobile application platform risk identification method and related devices. Background Art

[0002] This section is intended to provide a background or context to the embodiments of the disclosure that are recited in the claims. No description herein is admitted to be prior art by inclusion in this section.

[0003] As the company's digital transformation gradually deepens, the company's various units and professional employees have increasing demands for personalized, precise, and intelligent mobile applications, and at the same time, the requirements for the construction of mobile application platforms are gradually increasing. Efficiency, security, and intelligence are the basic requirements of company employees for mobile application platforms. Facing the high standards of employees, risk monitoring and prediction of mobile application platforms have gradually become an important part of the work of mobile application platforms.

[0004] In related technologies, string matching algorithms, such as the DFA algorithm, are usually used for sensitive text recognition; and graph neural network models are usually used to predict malicious users.

[0005] However, in the solutions in the related technologies, texts with specific sensitive words are not necessarily sensitive texts, and the misjudgment rate is relatively high. In reality, the number of malicious users is far smaller than that of normal users, and the number of samples that are clearly marked as malicious users is even rarer. Under the influence of a large number of normal sample features, it is difficult for graph neural networks to learn the typical features of malicious users, resulting in unsatisfactory prediction results. Summary of the invention

[0006] In view of this, the purpose of the present disclosure is to propose a mobile application platform risk identification method and related devices, which at least to a certain extent solve one of the technical problems in the related technology.

[0007] Based on the above purpose, the first aspect of the exemplary embodiment of the present disclosure provides a mobile application platform risk identification method, including:

[0008] Determine a user social network graph, perform adaptive breadth feature extraction and adaptive depth feature extraction on the user social network graph, and obtain a first feature of the user social network graph;

[0009] Determining user message data, performing feature extraction on the user message data, and obtaining a second feature of the user message data;

[0010] Performing semantic dependency analysis on the user message data to obtain semantic text nodes with semantic dependency relationships, constructing a semantic text graph based on the semantic text nodes, and performing feature extraction on the semantic text graph to obtain a third feature of the semantic text graph;

[0011] The second feature and the third feature are spliced ​​to obtain a first spliced ​​feature, and the first feature and the first spliced ​​feature are spliced ​​to obtain a second spliced ​​feature;

[0012] Sensitive text recognition is performed based on the second splicing feature to obtain a sensitive text recognition result, and malicious user recognition is performed based on the second splicing feature to obtain a malicious user recognition result.

[0013] Based on the same inventive concept, a second aspect of the exemplary embodiment of the present disclosure provides a mobile application platform risk identification device, including:

[0014] A first feature extraction module is configured to determine a user social network graph, perform adaptive breadth feature extraction and adaptive depth feature extraction on the user social network graph, and obtain a first feature of the user social network graph;

[0015] A second feature extraction module is configured to determine user message data, perform feature extraction on the user message data, and obtain a second feature of the user message data;

[0016] A third feature extraction module is configured to perform semantic dependency analysis on the user message data to obtain semantic text nodes with semantic dependency relationships, construct a semantic text graph based on the semantic text nodes, perform feature extraction on the semantic text graph, and obtain a third feature of the semantic text graph;

[0017] A feature splicing module is configured to splice the second feature and the third feature to obtain a first splicing feature, and to splice the first feature and the first splicing feature to obtain a second splicing feature;

[0018] The risk identification module is configured to perform sensitive text identification based on the second splicing feature to obtain a sensitive text identification result, and perform malicious user identification based on the second splicing feature to obtain a malicious user identification result.

[0019] Based on the same inventive concept, a third aspect of the exemplary embodiment of the present disclosure provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method described in the first aspect is implemented.

[0020] Based on the same inventive concept, a fourth aspect of the exemplary embodiments of the present disclosure provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the method described in the first aspect.

[0021] Based on the same inventive concept, a fifth aspect of the exemplary embodiments of the present disclosure provides a computer program product, including computer program instructions. When the computer program instructions are executed on a computer, the computer executes the method described in the first aspect.

[0022] From the above, it can be seen that the mobile application platform risk identification method and related device provided by the embodiment of the present disclosure include: determining a user social network graph, performing adaptive breadth feature extraction and adaptive depth feature extraction on the user social network graph, and obtaining a first feature of the user social network graph; determining user message data, performing feature extraction on the user message data, and obtaining a second feature of the user message data; performing semantic dependency analysis on the user message data, obtaining semantic text nodes with semantic dependency relationships, constructing a semantic text graph based on the semantic text nodes, and performing feature extraction on the semantic text graph to obtain a third feature of the semantic text graph; splicing the second feature and the third feature to obtain a first splicing feature, splicing the first feature and the first splicing feature to obtain a second splicing feature; performing sensitive text recognition based on the second splicing feature to obtain a sensitive text recognition result, and performing malicious user recognition based on the second splicing feature to obtain a malicious user recognition result. The present disclosure improves the accuracy of risk identification by extracting the behavioral characteristics of malicious users through joint learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the present disclosure or related technologies, the drawings required for use in the embodiments or related technical descriptions are briefly introduced below. Obviously, the drawings described below are only embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 A schematic diagram of an application scenario of a mobile application platform risk identification method provided by an exemplary embodiment of the present disclosure;

[0025] Figure 2 A flow chart of a method for identifying risks on a mobile application platform provided by an exemplary embodiment of the present disclosure;

[0026] Figure 3 A schematic diagram of a joint GNN model provided for an exemplary embodiment of the present disclosure;

[0027] Figure 4 A schematic diagram of a GeniePath model provided for an exemplary embodiment of the present disclosure;

[0028] Figure 5 A schematic diagram of a semantic text graph provided for an exemplary embodiment of the present disclosure;

[0029] Figure 6 A schematic diagram of a structure of a mobile application platform risk identification device provided by an exemplary embodiment of the present disclosure;

[0030] Figure 7 A schematic diagram of the structure of an electronic device provided by an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0031] It is understandable that before using the technical solutions disclosed in the embodiments of this application, the type, scope of use, usage scenarios, etc. of the personal information involved in this application should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0032] For example, in response to receiving an active request from a user, a prompt message is sent to the user to clearly prompt the user that the operation requested to be performed will require obtaining and using the user's personal information. Thus, the user can autonomously choose whether to provide personal information to software or hardware such as an electronic device, application, server, or storage medium that performs the operation of the technical solution of the present application according to the prompt message.

[0033] As an optional but non-limiting implementation, in response to receiving an active request from the user, the prompt information may be sent to the user in the form of a pop-up window, in which the prompt information may be presented in text form. In addition, the pop-up window may also carry a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0034] It is understandable that the above notification and the process of obtaining user authorization are merely illustrative and do not constitute a limitation on the implementation method of the present application. Other methods that meet the relevant laws and regulations may also be applied to the implementation method of the present application.

[0035] It is understandable that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) shall comply with the requirements of relevant laws, regulations and relevant provisions.

[0036] In order to make the purpose, technical solutions and advantages of the present disclosure more clear, the principles and spirit of the present disclosure will be described with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and implement the present disclosure, and are not intended to limit the scope of the present disclosure in any way. On the contrary, these embodiments are provided to make the present disclosure more thorough and complete, and to fully convey the scope of the present disclosure to those skilled in the art.

[0037] It should be understood herein that any number of elements in the drawings is for illustration rather than limitation, and any naming is only for distinction rather than having any limiting meaning.

[0038] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present disclosure should be understood by people with ordinary skills in the field to which the present disclosure belongs. The "first", "second" and similar words used in the embodiments of the present disclosure do not represent any order, quantity or importance, but are only used to distinguish different components. "Including" or "comprising" and similar words mean that the elements or objects appearing in front of the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connecting" or "connected" and similar words are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly. The article "one" or "a" before an element does not exclude the existence of multiple such elements.

[0039] The principle and spirit of the present disclosure are explained in detail below with reference to several representative embodiments of the present disclosure.

[0040] As described in the background technology, with the gradual deepening of the company's digital transformation, the company's various units and professional employees have increasing demands for personalized, precise and intelligent mobile applications, and at the same time, the requirements for the construction of mobile application platforms are gradually increasing. Efficiency, security and intelligence are the basic requirements of company employees for mobile application platforms. With high standards for employees, risk monitoring and prediction of mobile application platforms have gradually become an important part of the work of mobile application platforms.

[0041] Nowadays, with the continuous expansion of the scale of social networks on mobile application platforms, the spread of sensitive information is becoming more and more intense, and the patterns of malicious users who spread sensitive information are more difficult to grasp. Therefore, a method is needed to monitor sensitive information and mobile devices and predict the patterns of users who spread sensitive information. It is necessary to effectively monitor and filter sensitive information, and use the historical user information and historical login device information of those who spread sensitive information to predict users who are more likely to spread malicious information in the future, so as to ensure the normal and good operation of employees' digital work mode.

[0042] In order to solve the above problems, in the related technology, string (which can be understood as keywords) matching algorithms are usually used for sensitive text identification, such as the DFA algorithm. The DFA algorithm is a more traditional sensitive information identification algorithm. The advantage of this algorithm is its high detection efficiency. The disadvantage is that texts with specific sensitive words are not necessarily sensitive texts, and the misjudgment rate is high. Specifically, its risk monitoring method is: first, build a sensitive word library, then sort by the frequency of occurrence of sensitive words and build a sensitive word tree, and finally filter the target text based on the sensitive word tree to identify whether the target text is sensitive text.

[0043] In the related technology, malicious user prediction is usually completed using a classic graph neural network model. Specifically, the risk prediction method is: building a user social graph based on users and their social relationships, using a graph neural network model for feature learning and aggregation, and finally passing the feature learning results into the classifier to obtain the prediction results.

[0044] However, the inventors of the present disclosure have found that the solutions in the related art have the following shortcomings:

[0045] Its risk monitoring method is to first determine that texts containing specific sensitive words are sensitive texts, which has a high misjudgment rate. Secondly, sensitive texts that intentionally avoid specific keywords and use alternative characters need to be identified by continuously expanding the sensitive word library, which puts a lot of storage pressure.

[0046] The risk prediction methods are all aimed at classic graph neural networks and make related improvements such as increasing the number of model layers. These can improve the prediction accuracy to a certain extent, but when there is less training data, the model will degrade due to the lack of multi-perspective information. However, in reality, the number of malicious users is far less than that of normal users, and the number of samples clearly marked as malicious users is even rarer. Under the influence of a large number of normal sample features, it is difficult for classic graph neural networks to learn the typical features of malicious users, resulting in unsatisfactory prediction results.

[0047] At the same time, the probability of malicious users spreading noun information is much higher than that of ordinary users. Users who frequently send sensitive texts are also more likely to become malicious users. Joint learning of the two can effectively improve recognition and prediction accuracy, while traditional risk monitoring and prediction methods do not consider joint feature learning.

[0048] In order to solve the above problems, the present disclosure provides a mobile application platform risk identification solution, which specifically includes: determining a user social network graph, performing adaptive breadth feature extraction and adaptive depth feature extraction on the user social network graph, and obtaining a first feature of the user social network graph; determining user message data, performing feature extraction on the user message data, and obtaining a second feature of the user message data; performing semantic dependency analysis on the user message data, and obtaining semantic text nodes with semantic dependency relationships, constructing a semantic text graph based on the semantic text nodes, and performing feature extraction on the semantic text graph, and obtaining a third feature of the semantic text graph; splicing the second feature and the third feature to obtain a first splicing feature, and splicing the first feature and the first splicing feature to obtain a second splicing feature; performing sensitive text recognition based on the second splicing feature to obtain a sensitive text recognition result, and performing malicious user recognition based on the second splicing feature to obtain a malicious user recognition result. The present disclosure extracts the behavioral characteristics of malicious users through joint learning, thereby improving the accuracy of risk identification.

[0049] After introducing the basic principles of the present disclosure, various non-limiting embodiments of the present disclosure are described in detail below.

[0050] refer to Figure 1 , which is a schematic diagram of an application scenario of the mobile application platform risk identification method provided by an exemplary embodiment of the present disclosure.

[0051] This application scenario includes a terminal device 101, a server 102, and a data storage system 103. The terminal device 101, the server 102, and the data storage system 103 may be connected via a wired or wireless communication network to achieve data interaction.

[0052] The terminal device 101 may be an electronic device with data transmission and multimedia input / output functions close to the user side, including but not limited to a desktop computer, a mobile phone, a mobile computer, a tablet computer, a media player, a smart wearable device, a personal digital assistant (PDA) or other electronic devices capable of realizing the above functions. The electronic device may include a processor and a display screen with a touch input function, the display screen is used to present a graphical user interface, the graphical user interface can display an application interface, and the processor is used to process application data, generate a graphical user interface, and control the display of the graphical user interface on the display screen.

[0053] Server 102 and data storage system 103 can both be independent physical servers, or a server cluster or distributed system composed of multiple physical servers. They can also be cloud servers that provide basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), as well as big data and artificial intelligence platforms.

[0054] When the mobile application platform risk identification method is run on the server 102, the server 102 is used to provide a mobile application platform risk identification service to the user of the terminal device 101. The terminal device 101 is installed with a client that communicates with the server 102. The user can input user behavior data and user message data through the client. After clicking the identification button, the client sends the user behavior data and user message data to the server 102. The server 102 determines the user social network graph (based on the user behavior data, the user social network graph is constructed), and performs adaptive breadth feature extraction and adaptive depth feature extraction on the user social network graph to obtain the first feature of the user social network graph; determines the user message data , extract features from the user message data to obtain a second feature of the user message data; perform semantic dependency analysis on the user message data to obtain semantic text nodes with semantic dependency relationships, construct a semantic text graph based on the semantic text nodes, extract features from the semantic text graph to obtain a third feature of the semantic text graph; splice the second feature and the third feature to obtain a first splicing feature, splice the first feature and the first splicing feature to obtain a second splicing feature; perform sensitive text recognition based on the second splicing feature to obtain a sensitive text recognition result, perform malicious user recognition based on the second splicing feature to obtain a malicious user recognition result. The server 102 can send the sensitive text recognition result and the malicious user recognition result to the client, and the client displays the sensitive text recognition result and the malicious user recognition result to the user to help the user identify risks.

[0055] The data storage system 103 stores user behavior data and user message data.

[0056] Specifically, the present disclosure is mainly aimed at risk monitoring and prediction of mobile application platforms. The mobile application platform is responsible for managing mobile applications, and mainly includes six modules: permission management module, application management module, message management module, device management module, risk monitoring module, and risk prediction module. The platform administrator can send messages to user terminals based on mobile applications through the mobile application platform (including two methods: the platform administrator actively pushes messages and other R&D units / third parties push messages through open interfaces). The risk monitoring module needs to monitor in real time whether the sent messages contain sensitive information, and the risk prediction module needs to predict whether the current login account has a high probability of becoming a malicious user. The above messages include text, graphics, pictures, voice, audio, video, files, etc.

[0057] Combine the following Figure 1 The mobile application platform risk identification method according to the exemplary embodiment of the present disclosure is described by using the application scenario. It should be noted that the above application scenario is only shown to facilitate understanding of the spirit and principle of the present disclosure, and the embodiments of the present disclosure are not limited in this respect. On the contrary, the embodiments of the present disclosure can be applied to any applicable scenario.

[0058] refer to Figure 2 and Figure 3 The mobile application platform risk identification method provided by the exemplary embodiment of the present disclosure includes the following steps:

[0059] Step S210: determine a user social network graph, perform adaptive breadth feature extraction and adaptive depth feature extraction on the user social network graph, and obtain a first feature of the user social network graph.

[0060] In this exemplary embodiment, determining the user social network graph includes:

[0061] Determine user behavior data;

[0062] Based on the user behavior data, determine the user node and the message sending information between users;

[0063] Based on the message sending information between the users, undirected edges are added between the user nodes to construct the user social network graph.

[0064] In the specific implementation, past user behavior data and message sending data (text) are collected, and all historical messages in the previous year are counted on a daily basis. The data is preprocessed, and the user behavior data is used to build a user social network graph, and the message sending data is used to build a semantic dependency graph.

[0065] As a specific example, the user social network graph G = (U, X, L, E), where node i is represented by u i ∈U means that the user account is used as the unique identifier, U is the set of user nodes, x i ∈R l×d for u i The feature vector of all users is composed of the feature matrix X∈R M×d , l i ∈R 1×1 for u i The labels of all users are recorded as Y∈R M×1 ,e(u i ,u j )∈E represents that user i has sent a message to user j. The social relationships of all users constitute the edge set E. The characteristic information of the edges in set E includes the unique identifier of the edge, the frequency of sending messages, and the time of the most recent message sending.

[0066] In this exemplary embodiment, the step of performing adaptive breadth feature extraction and adaptive depth feature extraction on the user social network graph to obtain a first feature of the user social network graph includes:

[0067] Based on the GeniePath model, adaptive breadth feature extraction and adaptive depth feature extraction are performed on the user social network graph to obtain the first feature of the user social network graph.

[0068] In specific implementation, the user social network graph G will complete feature learning based on the GeniePath model, including adaptive breadth feature learning and adaptive depth feature learning.

[0069] Adaptive breadth feature learning process: For the input G, taking the i-th node as an example, node i and its adjacent node j (M(i)∪i) are calculated through the attention layer to calculate the attention coefficient α(i,j) of node i and its first-order neighbor j (j∈M(i)∪i):

[0070]

[0071] In formula (1-1), w s and w d They are all trainable parameters. The attention layer first performs weighted summation of the feature vectors of adjacent nodes based on the attention mechanism, and then passes a nonlinear activation function to increase the expression ability of the attention layer. Finally, the attention coefficients of node i and node j are obtained after normalization.

[0072] Continuing with the i-th node as an example, the attention coefficients of its adjacent nodes are transformed by matrix and weighted summation to obtain the breadth feature learning result of the target node i, that is, Where t represents the number of layers from the target node i. This feature learning process can be used to determine the in-depth exploration direction of the target node:

[0073]

[0074] J i,j is the Jaccard similarity coefficient of nodes i and j. The nodes that have recently sent messages frequently to nodes i and j are constructed into a set to calculate J i,j .

[0075] Adaptive deep feature learning process: Drawing on the long-short memory mechanism of LSTM, first-order neighbor feature information based on breadth feature learning The update gate of LSTM is used to continue to add useful neighbor feature information to the memory, and the forget gate is used to filter out useless neighbor feature information, saving or forgetting the feature information of each hop neighbor, and finally completing information filtering and extraction when exploring high-order neighbors.

[0076] refer to Figure 4 , showing the overall framework of GeniePath model feature learning and aggregation, where:

[0077] Select the feature vector (h∈R 1*d ) is used as the input of the subsequent feedforward neural network. The first layer of the feedforward neural network introduces the nonlinear activation function ReLU, and the second layer only performs linear transformation. After two layers of fully connected layers, the dimension of the feature vector remains unchanged. The specific transformation process is shown in formula (1-3):

[0078] FFN(h)=max(0,h*W1+b1)*W2+b2 (1-3)

[0079] The feedforward neural network subsequently adds residual and normalization operations to accelerate model convergence and improve model stability. The residual operation is to add the feature vector h to the output FFN(h) of the feedforward neural network. The normalization calculation method is shown in (1-5), and finally the result h of the GeniePath-FFN model is obtained. output :

[0080] h output =Norm(h+FFN(h)) (1-4)

[0081]

[0082] Step S220: determine user message data, perform feature extraction on the user message data, and obtain a second feature of the user message data.

[0083] In this exemplary embodiment, the step of extracting features from the user message data to obtain a second feature of the user message data includes:

[0084] Based on the DPCNN model, feature extraction is performed on the user message data to obtain the second feature of the user message data.

[0085] Step S230: perform semantic dependency analysis on the user message data to obtain semantic text nodes with semantic dependency relationships, construct a semantic text graph based on the semantic text nodes, perform feature extraction on the semantic text graph, and obtain a third feature of the semantic text graph.

[0086] In a specific implementation, the semantic dependency analysis is performed on the user message data, including:

[0087] When the user message data is text information, semantic dependency analysis is directly performed on the text information;

[0088] When the user message data is non-text information, such as voice data, image data or video data, it is converted into text information, and then semantic dependency analysis is performed on the converted text information.

[0089] In a specific implementation, a semantic text graph is constructed based on the semantic text nodes, including:

[0090] Undirected edges are added between semantic text nodes with semantic dependencies to construct a semantic text graph.

[0091] refer to Figure 5 , as a specific example of a semantic text graph, taking the user message data "I like to eat apples" as an example for explanation:

[0092] Set input text T = {w1,w2,...,w i}, where w i Represents the i-th word in the text. It is initialized as a d-dimensional vector using a model such as BERT, and is continuously updated during training. The output is T ’ .

[0093] In this exemplary embodiment, the step of extracting features from the semantic text graph to obtain a third feature of the semantic text graph includes:

[0094] Based on the GNN model, feature extraction is performed on the semantic text graph to obtain the third feature of the semantic text graph.

[0095] In specific implementation, the GNN model uses the nodes of the semantic text graph as graph nodes, mines the deep semantic information of the sentence based on the attention mechanism, and enriches the text feature information extracted by the model. However, the local feature extraction ability of GNN is weak and it cannot obtain the local receptive field. Therefore, the GNN model is processed in parallel with the DPCNN model, which can not only improve the local text feature information extraction ability through the convolutional neural network, but also mine the deep semantic connection between words in the sentence through the graph neural network.

[0096] The DPCNN model first uses a set of convolution operations to learn the features of the input vector. The feature learning results are added to the feature learning results after two equal-length convolution operations. Then, a pooling layer with a stride of 2 is passed to reduce the complexity of the model. The equal-length convolution + 1 / 2 pooling layer operation is repeated until the corresponding dimension is reached.

[0097] The input of DPCNN model is text X = {X1,X2,...,X m}, where m is the total number of words in the input text (seq_len), X i Represents the embedding of the i-th word. Assuming that the number of channels of the convolutional layer of Region Embedding is 250 and the height is 3, X will become X′∈R after passing through the Region Embedding layer. 250*(m-3+1)*1 .

[0098] The input of the equal-length convolutional layer is X′, and the output is X″∈R 250*(m-3+1)*1 This paper uses a convolution kernel of length 3. After two equal-length convolution operations, X″ i Can express X′ i and X′ i-1 , X′ i-2 , X′ i+1 , X′ i+2 The characteristic information of the five word positions, X″ i The semantic feature information expressed is richer. It should be noted that activation is performed before the convolution operation:

[0099] X′=WRelu(X′)+b (1-6)

[0100] To prevent the gradient vanishing problem, X′ and X″ are residually connected, and the connection result is sent to the 1 / 2 pooling layer to compress the length of the input feature vector to half of the original length to obtain longer semantic feature information. Then two layers of equal-length convolutional layers are connected, as shown in the figure above, and repeated until the length of m is 1. The output result is X″′∈R 250*1 , the output of the DPCNN model is X″′∈R channel*1 .

[0101] The GNN model input is a semantic dependency graph, and a graph convolutional neural network is constructed to complete sensitive information recognition. In order to capture important information as much as possible in limited learning data, a cross-layer aggregation path is adopted to directly merge the second-order neighbors and the first-order neighbors into the same layer to avoid information loss caused by information transmission from layer to layer, and to complete the neighbor sorting by calculating the attention coefficient (the first-order neighbors are sorted first, and the first-order important neighbors are selected and then the attention coefficient between them and the neighbors is calculated. The second-order neighbors with higher attention coefficients are selected and merged into the same layer to form an important neighbor set of the central node). The specific calculation is shown in the above formula (1-1). According to the attention coefficient, the priority sequence of neighbors N(u)=N is directly generated. 1 (u)∪N 2 (u).

[0102] Continuing with the example of the i-th node as the central node, the attention coefficients of its adjacent nodes are transformed by matrix and weighted summation to obtain the weighted aggregate feature learning result of the target node i, that is,

[0103]

[0104] The ROOT node aggregates the richest semantic feature information of the text. This paper selects the feature vector of the ROOT node (h∈R 1*d ) is used as the input of the subsequent feedforward neural network. The first layer of the feedforward neural network introduces the nonlinear activation function ReLU, and the second layer only performs linear transformation. After two layers of fully connected layers, the dimension of the feature vector remains unchanged. The specific transformation process is shown in formula (1-3):

[0105] FFN(h)=max(0,h*W1+b1)*W2+b2 (1-8)

[0106] The feedforward neural network then adds normalization operations to accelerate model convergence and improve model stability. The normalization calculation method is shown in (1-5), and the result y of the GNN model is finally obtained. output :

[0107]

[0108] Step S240: splice the second feature and the third feature to obtain a first spliced ​​feature, and splice the first feature and the first spliced ​​feature to obtain a second spliced ​​feature.

[0109] The outputs of the GNN model and the DPCNN model are horizontally spliced ​​to obtain the final output result h′ of the feature learning stage output .

[0110] h′ output =Concat(youtput +X″′ T ) (1-8)

[0111] The output h of the GeniePath model output And the output h′ of the GNN model output The horizontal splicing serves as the input of the subsequent two classifiers.

[0112] R output =Concat(h output +h′ output ) (1-9)

[0113] Step S250: performing sensitive text recognition based on the second splicing feature to obtain a sensitive text recognition result, and performing malicious user recognition based on the second splicing feature to obtain a malicious user recognition result.

[0114] In this exemplary embodiment, the sensitive text recognition is performed based on the second splicing feature to obtain a sensitive text recognition result, including:

[0115] Processing the second splicing feature based on a first preset function to obtain a first linear transformation result;

[0116] Processing the first linear transformation result based on a first preset activation function to obtain a first intermediate output;

[0117] Processing the first intermediate output based on a second preset function to obtain a second linear transformation result;

[0118] Processing the second linear transformation result based on a second preset activation function to obtain the sensitive text recognition result;

[0119] The performing malicious user identification based on the second splicing feature to obtain a malicious user identification result includes:

[0120] Processing the second splicing feature based on a third preset function to obtain a third linear transformation result;

[0121] Processing the third linear transformation result based on a third preset activation function to obtain a second intermediate output;

[0122] Processing the second intermediate output based on a fourth preset function to obtain a fourth linear transformation result;

[0123] The fourth linear transformation result is processed based on a fourth preset activation function to obtain the malicious user identification result.

[0124] In specific implementation, this joint model has two classifiers, one for monitoring whether there is sensitive text, and the other for predicting whether the user is a malicious user. The two classifiers have the same structure, and the input is the output R of the Concat layer. output ∈R 1*(2*d+channel) , the classifier is divided into two layers as a whole. The specific process is as follows: the first layer of the classifier inputs I output for The weight w, after the linear transformation shown in formula (1-10), will pass through the activation function, the main purpose of which is to enhance the nonlinear modeling ability of the model, see (1-11); the input I of the second layer output is the output Y of the first layer output , after linear transformation, it also passes through the activation function, and the final result is Y∈R 1*1 ,; The two classifier outputs represent the probability that the input text is sensitive text and the user is malicious.

[0125] Y output =w*I output +b (1-10)

[0126] Y=σ( Y output ) (1-11)

[0127] refer to Figure 3 , showing the structure of the joint GNN model for risk monitoring and prediction, where:

[0128] A joint GNN model is constructed, which includes a node embedding layer, a GNN layer, and a fully connected layer. The node embedding layer maps the nodes in the graph to vector representations; the GNN layer updates the node representation by adaptively aggregating the information of neighboring nodes; and the fully connected layer is used for the final classification task.

[0129] Joint training of sensitive word prediction and malicious user identification: For the sensitive word prediction task, text nodes containing sensitive words are used as positive samples, and other text nodes are used as negative samples; for the malicious user identification task, user nodes with malicious behavior in the historical records are used as positive samples, and other user nodes are used as negative samples. Joint training of the two tasks is achieved by splicing the trained features in the GNN model.

[0130] Use the trained joint GNN model to monitor and predict user behaviors and texts on the mobile application platform in real time. For new user behaviors and texts, add them to the graph structure data and use the GNN model for prediction.

[0131] From the above, it can be seen that the mobile application platform risk identification method provided by the embodiment of the present disclosure includes: determining a user social network graph, performing adaptive breadth feature extraction and adaptive depth feature extraction on the user social network graph, and obtaining a first feature of the user social network graph; determining user message data, performing feature extraction on the user message data, and obtaining a second feature of the user message data; performing semantic dependency analysis on the user message data, obtaining semantic text nodes with semantic dependency relationships, constructing a semantic text graph based on the semantic text nodes, and performing feature extraction on the semantic text graph to obtain a third feature of the semantic text graph; splicing the second feature and the third feature to obtain a first splicing feature, splicing the first feature and the first splicing feature to obtain a second splicing feature; performing sensitive text recognition based on the second splicing feature to obtain a sensitive text recognition result, and performing malicious user recognition based on the second splicing feature to obtain a malicious user recognition result. The present disclosure extracts the behavioral characteristics of malicious users through joint learning to improve the accuracy of risk identification.

[0132] Specifically, the prediction performance is improved: for sensitive information with a small number of samples and malicious users who spread sensitive information, the prediction performance is improved by jointly training sensitive word prediction and malicious user identification.

[0133] Reduced model complexity: Compared with training two independent models separately, joint training can significantly reduce model complexity and reduce the consumption of computing resources and storage space.

[0134] Improved real-time performance: Using real-time updated graph structure data and GNN models for risk monitoring and prediction can timely detect and handle potential risks.

[0135] It should be noted that the method of the embodiment of the present disclosure can be performed by a single device, such as a computer or a server. The method of the present embodiment can also be applied in a distributed scenario and completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only perform one or more steps in the method of the embodiment of the present disclosure, and the multiple devices will interact with each other to complete the described method.

[0136] It should be noted that the above describes some embodiments of the present disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the above embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0137] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present disclosure also provides a mobile application platform risk identification device.

[0138] refer to Figure 6 The mobile application platform risk identification device provided by the exemplary embodiment of the present disclosure includes the following modules:

[0139] A first feature extraction module 610 is configured to determine a user social network graph, perform adaptive breadth feature extraction and adaptive depth feature extraction on the user social network graph, and obtain a first feature of the user social network graph;

[0140] The second feature extraction module 620 is configured to determine the user message data, perform feature extraction on the user message data, and obtain a second feature of the user message data;

[0141] The third feature extraction module 630 is configured to perform semantic dependency analysis on the user message data to obtain semantic text nodes with semantic dependency relationships, construct a semantic text graph based on the semantic text nodes, and perform feature extraction on the semantic text graph to obtain a third feature of the semantic text graph;

[0142] The feature splicing module 640 is configured to splice the second feature and the third feature to obtain a first spliced ​​feature, and splice the first feature and the first spliced ​​feature to obtain a second spliced ​​feature;

[0143] The risk identification module 650 is configured to perform sensitive text identification based on the second splicing feature to obtain a sensitive text identification result, and perform malicious user identification based on the second splicing feature to obtain a malicious user identification result.

[0144] In some exemplary embodiments, the first feature extraction module 610 is specifically configured to:

[0145] Determine user behavior data;

[0146] Based on the user behavior data, determine the user node and the message sending information between users;

[0147] Based on the message sending information between the users, undirected edges are added between the user nodes to construct the user social network graph.

[0148] In some exemplary embodiments, the first feature extraction module 610 is specifically configured to:

[0149] Based on the GeniePath model, adaptive breadth feature extraction and adaptive depth feature extraction are performed on the user social network graph to obtain the first feature of the user social network graph.

[0150] In some exemplary embodiments, the second feature extraction module 620 is specifically configured to:

[0151] Based on the DPCNN model, feature extraction is performed on the user message data to obtain the second feature of the user message data.

[0152] In some exemplary embodiments, the third feature extraction module 630 is specifically configured to:

[0153] Based on the GNN model, feature extraction is performed on the semantic text graph to obtain the third feature of the semantic text graph.

[0154] In some exemplary embodiments, the risk identification module 650 is specifically configured to:

[0155] Processing the second splicing feature based on a first preset function to obtain a first linear transformation result;

[0156] Processing the first linear transformation result based on a first preset activation function to obtain a first intermediate output;

[0157] Processing the first intermediate output based on a second preset function to obtain a second linear transformation result;

[0158] Processing the second linear transformation result based on a second preset activation function to obtain the sensitive text recognition result;

[0159] The performing malicious user identification based on the second splicing feature to obtain a malicious user identification result includes:

[0160] Processing the second splicing feature based on a third preset function to obtain a third linear transformation result;

[0161] Processing the third linear transformation result based on a third preset activation function to obtain a second intermediate output;

[0162] Processing the second intermediate output based on a fourth preset function to obtain a fourth linear transformation result;

[0163] The fourth linear transformation result is processed based on a fourth preset activation function to obtain the malicious user identification result.

[0164] For the convenience of description, the above device is described by dividing it into various modules according to its functions. Of course, when implementing the present disclosure, the functions of each module can be implemented in the same or multiple software and / or hardware.

[0165] The device of the above embodiment is used to implement the corresponding mobile application platform risk identification method in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here.

[0166] Based on the same inventive concept, corresponding to any of the above-mentioned embodiments and methods, the present disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the mobile application platform risk identification method described in any of the above embodiments is implemented.

[0167] Figure 7 A more specific schematic diagram of the hardware structure of an electronic device provided in this embodiment is shown, and the device may include: a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040 are connected to each other through the bus 1050 in the device.

[0168] The processor 1010 can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0169] The memory 1020 may be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 may store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program codes are stored in the memory 1020 and are called and executed by the processor 1010.

[0170] The input / output interface 1030 is used to connect the input / output module to realize information input and output. The input / output module can be configured in the device as a component (not shown in the figure), or it can be externally connected to the device to provide corresponding functions. The input device may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device may include a display, a speaker, a vibrator, an indicator light, etc.

[0171] The communication interface 1040 is used to connect a communication module (not shown) to realize communication interaction between the device and other devices. The communication module can realize communication through a wired mode (such as USB, network cable, etc.) or a wireless mode (such as mobile network, WIFI, Bluetooth, etc.).

[0172] The bus 1050 includes a path that transmits information between the various components of the device (eg, the processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040).

[0173] It should be noted that, although the above device only shows the processor 1010, the memory 1020, the input / output interface 1030, the communication interface 1040 and the bus 1050, in the specific implementation process, the device may also include other components necessary for normal operation. In addition, it can be understood by those skilled in the art that the above device may also only include the components necessary for implementing the embodiments of the present specification, and does not necessarily include all the components shown in the figure.

[0174] The electronic device of the above embodiment is used to implement the corresponding mobile application platform risk identification method in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here.

[0175] The memory 1020 stores machine-readable instructions executable by the processor 1010. When the electronic device is running, the processor 1010 communicates with the memory 1020 via the bus 1030, so that the processor 1010 executes the following instructions when running:

[0176] Determine a user social network graph, perform adaptive breadth feature extraction and adaptive depth feature extraction on the user social network graph, and obtain a first feature of the user social network graph;

[0177] Determining user message data, performing feature extraction on the user message data, and obtaining a second feature of the user message data;

[0178] Performing semantic dependency analysis on the user message data to obtain semantic text nodes with semantic dependency relationships, constructing a semantic text graph based on the semantic text nodes, and performing feature extraction on the semantic text graph to obtain a third feature of the semantic text graph;

[0179] The second feature and the third feature are spliced ​​to obtain a first spliced ​​feature, and the first feature and the first spliced ​​feature are spliced ​​to obtain a second spliced ​​feature;

[0180] Sensitive text recognition is performed based on the second splicing feature to obtain a sensitive text recognition result, and malicious user recognition is performed based on the second splicing feature to obtain a malicious user recognition result.

[0181] In a possible implementation, in the instructions executed by the processor 1010, determining the user social network graph includes:

[0182] Determine user behavior data;

[0183] Based on the user behavior data, determine the user node and the message sending information between users;

[0184] Based on the message sending information between the users, undirected edges are added between the user nodes to construct the user social network graph.

[0185] In a possible implementation, in the instructions executed by the processor 1010, the step of performing adaptive breadth feature extraction and adaptive depth feature extraction on the user social network graph to obtain a first feature of the user social network graph includes:

[0186] Based on the GeniePath model, adaptive breadth feature extraction and adaptive depth feature extraction are performed on the user social network graph to obtain the first feature of the user social network graph.

[0187] In a possible implementation manner, in the instructions executed by the processor 1010, the extracting features of the user message data to obtain the second feature of the user message data includes:

[0188] Based on the DPCNN model, feature extraction is performed on the user message data to obtain the second feature of the user message data.

[0189] In a possible implementation, in the instructions executed by the processor 1010, the step of extracting features from the semantic text graph to obtain a third feature of the semantic text graph includes:

[0190] Based on the GNN model, feature extraction is performed on the semantic text graph to obtain the third feature of the semantic text graph.

[0191] In a possible implementation manner, in the instructions executed by the processor 1010, the step of performing sensitive text recognition based on the second splicing feature to obtain a sensitive text recognition result includes:

[0192] Processing the second splicing feature based on a first preset function to obtain a first linear transformation result;

[0193] Processing the first linear transformation result based on a first preset activation function to obtain a first intermediate output;

[0194] Processing the first intermediate output based on a second preset function to obtain a second linear transformation result;

[0195] Processing the second linear transformation result based on a second preset activation function to obtain the sensitive text recognition result;

[0196] The performing malicious user identification based on the second splicing feature to obtain a malicious user identification result includes:

[0197] Processing the second splicing feature based on a third preset function to obtain a third linear transformation result;

[0198] Processing the third linear transformation result based on a third preset activation function to obtain a second intermediate output;

[0199] Processing the second intermediate output based on a fourth preset function to obtain a fourth linear transformation result;

[0200] The fourth linear transformation result is processed based on a fourth preset activation function to obtain the malicious user identification result.

[0201] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present disclosure also provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the mobile application platform risk identification method described in any of the above embodiments.

[0202] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0203] The above-mentioned non-transitory computer-readable storage medium can be any available medium or data storage device that can be accessed by a computer, including but not limited to magnetic storage (such as floppy disks, hard disks, magnetic tapes, magneto-optical disks (MO), etc.), optical storage (such as CD, DVD, BD, HVD, etc.), and semiconductor storage (such as ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)), etc.

[0204] The computer instructions stored in the storage medium of the above embodiment are used to enable the computer to execute the mobile application platform risk identification method described in any embodiment in the above exemplary method part, and have the beneficial effects of the corresponding method embodiment, which will not be repeated here.

[0205] Based on the same inventive concept, corresponding to the mobile application platform risk identification method described in any of the above embodiments, the present disclosure also provides a computer program product, which includes computer program instructions. In some embodiments, the computer program instructions can be executed by one or more processors of a computer so that the computer and / or the processor executes the mobile application platform risk identification method. Corresponding to the execution subject corresponding to each step in each embodiment of the mobile application platform risk identification method, the processor that executes the corresponding step may belong to the corresponding execution subject.

[0206] The computer program product of the above embodiment is used to enable the computer and / or the processor to execute the mobile application platform risk identification method described in any of the above embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0207] Those skilled in the art will appreciate that the embodiments of the present disclosure may be implemented as a system, method, or computer program product. Therefore, the present disclosure may be specifically implemented in the following forms, namely: complete hardware, complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." In addition, in some embodiments, the present disclosure may also be implemented in the form of a computer program product in one or more computer-readable media, which contains computer-readable program code.

[0208] Any combination of one or more computer-readable media may be used. A computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive examples) of computer-readable storage media may include, for example: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device.

[0209] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0210] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0211] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0212] It should be understood that each box in the flowchart and / or block diagram and the combination of boxes in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer or other programmable data processing device to produce a machine, and these computer program instructions are executed by a computer or other programmable data processing device to produce a device that implements the functions / operations specified in the boxes in the flowchart and / or block diagram.

[0213] These computer program instructions may also be stored in a computer-readable medium that enables a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable medium produce a product that includes an instruction device that implements the functions / operations specified in the blocks in the flowchart and / or block diagram.

[0214] Computer program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operational steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby enabling the instructions executed on the computer or other programmable device to provide a process for implementing the functions / operations specified in the blocks in the flowchart and / or block diagram.

[0215] In addition, although the operations of the disclosed method are described in a particular order in the accompanying drawings, this does not require or imply that the operations must be performed in this particular order, or that all the operations shown must be performed to achieve the desired results. On the contrary, the steps depicted in the flow chart can be performed in a different order. Additionally or alternatively, some steps can be omitted, multiple steps can be combined into one step, and / or one step can be decomposed into multiple steps.

[0216] The flowchart and block diagram in the accompanying drawings illustrate the possible architecture, functions and operations of the system, method and computer program product according to various embodiments of the present application. Wherein, each box in the flowchart or block diagram can represent a module, a program segment, or a part of the code, and the above-mentioned module, program segment, or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0217] It should be noted that, although several modules or units of the equipment for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into being embodied by multiple modules or units.

[0218] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present application (including the claims) is limited to these examples. In line with the concept of the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

[0219] In addition, to simplify the description and discussion, and in order not to make the embodiments of the present application difficult to understand, the known power supply / ground connection with the integrated circuit (IC) chip and other components may or may not be shown in the provided drawings. In addition, the device can be shown in the form of a block diagram to avoid making the embodiments of the present application difficult to understand, and this also takes into account the fact that the details of the implementation of these block diagram devices are highly dependent on the platform to be implemented in the embodiments of the present application (that is, these details should be fully within the scope of understanding of those skilled in the art). In the case of elaborating specific details (e.g., circuits) to describe exemplary embodiments of the present application, it is obvious to those skilled in the art that the embodiments of the present application can be implemented without these specific details or when these specific details are changed. Therefore, these descriptions should be considered to be illustrative rather than restrictive.

[0220] Although the present application has been described in conjunction with specific embodiments of the present application, many replacements, modifications and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may use the embodiments discussed.

[0221] The embodiments of the present application are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present application should be included in the scope of protection of the present application.

[0222] Although the spirit and principle of the present disclosure have been described with reference to several specific embodiments, it should be understood that the present disclosure is not limited to the disclosed specific embodiments, and the division of various aspects does not mean that the features in these aspects cannot be combined to benefit, and such division is only for the convenience of expression. The present disclosure is intended to cover various modifications and equivalent arrangements included in the spirit and scope of the attached claims. The scope of the attached claims conforms to the broadest interpretation, thereby including all such modifications and equivalent structures and functions.

Claims

1. A mobile application platform risk identification method, characterized in that: include: Determine a user social network graph, perform adaptive breadth feature extraction and adaptive depth feature extraction on the user social network graph, and obtain a first feature of the user social network graph; Determining user message data, performing feature extraction on the user message data, and obtaining a second feature of the user message data; Performing semantic dependency analysis on the user message data to obtain semantic text nodes with semantic dependency relationships, constructing a semantic text graph based on the semantic text nodes, and performing feature extraction on the semantic text graph to obtain a third feature of the semantic text graph; The second feature and the third feature are spliced ​​to obtain a first spliced ​​feature, and the first feature and the first spliced ​​feature are spliced ​​to obtain a second spliced ​​feature; Sensitive text recognition is performed based on the second splicing feature to obtain a sensitive text recognition result, and malicious user recognition is performed based on the second splicing feature to obtain a malicious user recognition result.

2. The method according to claim 1, characterized in that Determining the user social network graph includes: Determine user behavior data; Based on the user behavior data, determine the user node and the message sending information between users; Based on the message sending information between the users, undirected edges are added between the user nodes to construct the user social network graph.

3. The method according to claim 1, characterized in that The step of performing adaptive breadth feature extraction and adaptive depth feature extraction on the user social network graph to obtain a first feature of the user social network graph includes: Based on the GeniePath model, adaptive breadth feature extraction and adaptive depth feature extraction are performed on the user social network graph to obtain the first feature of the user social network graph.

4. The method according to claim 1, characterized in that The extracting features of the user message data to obtain a second feature of the user message data includes: Based on the DPCNN model, feature extraction is performed on the user message data to obtain the second feature of the user message data.

5. The method according to claim 1, characterized in that The step of extracting features from the semantic text graph to obtain a third feature of the semantic text graph includes: Based on the GNN model, feature extraction is performed on the semantic text graph to obtain the third feature of the semantic text graph.

6. The method according to claim 1, characterized in that The step of performing sensitive text recognition based on the second splicing feature to obtain a sensitive text recognition result includes: Processing the second splicing feature based on a first preset function to obtain a first linear transformation result; Processing the first linear transformation result based on a first preset activation function to obtain a first intermediate output; Processing the first intermediate output based on a second preset function to obtain a second linear transformation result; Processing the second linear transformation result based on a second preset activation function to obtain the sensitive text recognition result; The performing malicious user identification based on the second splicing feature to obtain a malicious user identification result includes: Processing the second splicing feature based on a third preset function to obtain a third linear transformation result; Processing the third linear transformation result based on a third preset activation function to obtain a second intermediate output; Processing the second intermediate output based on a fourth preset function to obtain a fourth linear transformation result; The fourth linear transformation result is processed based on a fourth preset activation function to obtain the malicious user identification result.

7. A mobile application platform risk identification device, characterized in that: include: A first feature extraction module is configured to determine a user social network graph, perform adaptive breadth feature extraction and adaptive depth feature extraction on the user social network graph, and obtain a first feature of the user social network graph; A second feature extraction module is configured to determine user message data, perform feature extraction on the user message data, and obtain a second feature of the user message data; A third feature extraction module is configured to perform semantic dependency analysis on the user message data to obtain semantic text nodes with semantic dependency relationships, construct a semantic text graph based on the semantic text nodes, perform feature extraction on the semantic text graph, and obtain a third feature of the semantic text graph; A feature splicing module is configured to splice the second feature and the third feature to obtain a first splicing feature, and to splice the first feature and the first splicing feature to obtain a second splicing feature; The risk identification module is configured to perform sensitive text identification based on the second splicing feature to obtain a sensitive text identification result, and perform malicious user identification based on the second splicing feature to obtain a malicious user identification result.

8. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to cause a computer to execute the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that The method comprises computer program instructions, which, when executed on a computer, cause the computer to execute the method according to any one of claims 1 to 6.

Citation Information

Patent Citations

  • Social network garbage user detection method integrated with multiple algorithms

    CN106296422A

  • Malicious information detection method, device and system based on multi-feature fusion

    CN115759110A

  • Visualization of participant relationships and sentiment for electronic messaging

    US20100223581A1