Social media information detection method and device, electronic equipment and storage medium

Abstract

The invention provides a social media information detection method and device, electronic equipment and a storage medium, and belongs to the technical field of financial science and technology. The method comprises the steps of obtaining target social media information, wherein the target social media information is used for indicating target push content and a target object having an interaction behavior for the target push content; obtaining comment data generated by the target object before the current time; performing graph attention calculation based on the object information of the target object to obtain object graph features; performing graph attention calculation based on the comment data of the target object to obtain comment graph features; performing attention calculation based on the target content feature, the object graph feature and the comment graph feature of the target push content to obtain a target social media feature vector; the target push content is detected based on the target social media feature vector, a detection result is obtained, and the detection result is used for indicating whether the target push content is real or not. According to the invention, the detection accuracy of the abnormal social media information can be improved.

Description

Technical Field

[0001] The present application relates to the field of financial technology, and in particular to a social media information detection method and device, an electronic device and a storage medium. Background Art

[0002] With the rapid development of Internet technology, social networks have become an indispensable part of people's daily lives. It has not only changed the way people communicate, but also greatly promoted the spread of information. However, there is often some fictitious and false information spreading in social networks, which will mislead the public and reduce the accuracy of information dissemination. For example, in the financial field, the spread of untrue financial information will greatly affect the subject's judgment in various financial activities, and often cause resource losses to the subject.

[0003] Based on this, related technologies have considered using neural network models to detect the authenticity of social media information based on the analysis of the specific content of social media information or the behavior of users who publish each social media information. However, this method has the problem of being unable to obtain implicitly expressed false information and being unable to accurately identify the propagation path of false information, resulting in low accuracy in detecting abnormal social media information. Summary of the invention

[0004] The main purpose of the embodiments of the present application is to propose a social media information detection method and device, an electronic device and a storage medium, aiming to improve the accuracy of detecting abnormal social media information.

[0005] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application proposes a social media information detection method, the method comprising:

[0006] Acquire target social media information, wherein the target social media information is used to indicate target push content and target objects that have interactive behaviors with respect to the target push content;

[0007] Obtain the comment data generated by the target object before the current time;

[0008] Performing graph attention calculation based on the object information of the target object to obtain object graph features;

[0009] Performing graph attention calculation based on the comment data of the target object to obtain comment graph features;

[0010] Performing attention calculation based on the pre-acquired target content features of the target pushed content, the object graph features, and the comment graph features to obtain a target social media feature vector;

[0011] The target pushed content is detected and processed based on the target social media feature vector to obtain a detection result of the target pushed content, wherein the detection result is used to indicate whether the target pushed content is authentic.

[0012] In some embodiments, performing graph attention calculation based on the object information of the target object to obtain object graph features includes:

[0013] Constructing an object node of an object graph structure based on each of the target objects, and constructing an object edge of the object graph structure based on interaction information between every two of the target objects, so as to generate the object graph structure;

[0014] In the object graph structure, based on the object information of each target object, an initial object node feature of each object node is determined, and based on the interaction information between the target objects corresponding to two object nodes associated with each object edge, an object edge weight feature of the object edge is determined;

[0015] For a first node among the object nodes, performing attention calculation on the first node based on an initial object node feature of the first node, an initial object node feature of a second node other than the first node, and an object edge weight feature of the object edge connecting the first node and the second node, to obtain a first attention coefficient, wherein the first attention coefficient is used to indicate the degree of attention of the first node to the second node;

[0016] Performing feature updates on the initial object node feature of the first node based on the first attention coefficient to obtain a target object node feature of the first node;

[0017] Average pooling is performed on the target object node features of multiple first nodes to obtain the object graph features.

[0018] In some embodiments, the performing attention calculation on the first node based on the initial object node feature of the first node, the initial object node features of the second nodes other than the first node, and the object edge weight feature of the object edge connecting the first node and the second node to obtain the first attention coefficient includes:

[0019] Based on a preset weight matrix, feature splicing is performed on the initial object node feature of the first node and the initial object node feature of the second node to obtain an object node splicing feature;

[0020] Performing attention calculation on the object node splicing feature and the object edge weight feature to obtain the object node fusion feature;

[0021] The activation result of the object node fusion feature is normalized to obtain the first attention coefficient.

[0022] In some embodiments, the updating the initial object node feature of the first node based on the first attention coefficient to obtain the target object node feature of the first node includes:

[0023] Performing matrix multiplication on the first attention coefficient of the first node for a single second node, the initial object node feature of the single second node, and a preset weight matrix to obtain a product feature vector;

[0024] Based on the number of nodes of the second node, performing feature summation on a plurality of the product feature vectors to obtain adjacent object features of the first node;

[0025] The adjacent object features are activated to obtain the target object node features.

[0026] In some embodiments, the comment data includes each comment information generated by the target object before the current time, and the correlation between each two of the comment information;

[0027] The performing graph attention calculation based on the comment data of the target object to obtain comment graph features includes:

[0028] constructing a comment node of a comment graph structure based on each of the comment information, and constructing a comment edge of the comment graph structure based on the correlation between each two of the comment information, so as to generate the comment graph structure;

[0029] In the comment graph structure, based on each of the comment information, the initial comment node feature of each of the comment nodes is determined, and based on the correlation between the comment information corresponding to the two comment nodes associated with each of the comment edges, the comment edge weight feature of the comment edge is determined;

[0030] For a third node among the comment nodes, performing attention calculation on the third node based on an initial comment node feature of the third node, an initial comment node feature of a fourth node other than the third node, and a comment edge weight feature of the object edge connecting the third node and the fourth node, to obtain a second attention coefficient, wherein the second attention coefficient is used to indicate the degree of attention of the third node to the fourth node;

[0031] Based on the second attention coefficient, the initial comment node feature of the third node is updated to obtain a target comment node feature of the third node;

[0032] Average pooling is performed on the target comment node features of multiple third nodes to obtain the comment graph features.

[0033] In some embodiments, the attention calculation is performed based on the pre-acquired target content features of the target pushed content, the object graph features, and the comment graph features to obtain the target social media feature vector, including:

[0034] Performing feature projection on the object graph features based on a preset query projection matrix to obtain a target query feature vector;

[0035] Perform feature projection on the review graph features based on a preset value projection matrix to obtain a target value feature vector;

[0036] Performing feature projection on the target content feature based on a preset key projection matrix to obtain a target key feature vector;

[0037] Performing attention calculation on the target key feature vector and the target query feature vector to obtain an attention weight;

[0038] The attention weight and the target value feature vector are weightedly calculated to obtain the target social media feature vector.

[0039] In some embodiments, the detecting and processing the target pushed content based on the target social media feature vector to obtain the detection result of the target pushed content includes:

[0040] Mapping the target social media feature vector to a preset feature vector space to obtain a mapped social media feature vector;

[0041] Performing probability calculation on the mapped social media feature vector based on a preset function to obtain abnormal probability data of the target pushed content;

[0042] If it is determined that the abnormal probability data is less than or equal to a preset threshold, the detection result is determined as the target pushed content is authentic;

[0043] If it is determined that the abnormal probability data is greater than a preset threshold, the detection result is determined as the target pushed content is not authentic.

[0044] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application proposes a social media information detection device, the device comprising:

[0045] A first acquisition unit is used to acquire target social media information, wherein the target social media information is used to indicate target push content and a target object that has an interactive behavior with respect to the target push content;

[0046] A second acquisition unit, used to acquire comment data generated by the target object before the current time;

[0047] A first computing unit, configured to perform graph attention calculation based on the object information of the target object to obtain object graph features;

[0048] A second computing unit, configured to perform graph attention computing based on the comment data of the target object to obtain comment graph features;

[0049] A third calculation unit is used to perform attention calculation based on the pre-acquired target content features of the target pushed content, the object graph features, and the comment graph features to obtain a target social media feature vector;

[0050] A detection unit is used to perform detection processing on the target pushed content based on the target social media feature vector to obtain a detection result of the target pushed content, wherein the detection result is used to indicate whether the target pushed content is authentic.

[0051] To achieve the above-mentioned purpose, a third aspect of an embodiment of the present application proposes an electronic device, wherein the electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the social media information detection method described in the first aspect is implemented.

[0052] To achieve the above-mentioned purpose, the fourth aspect of the embodiments of the present application proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the social media information detection method described in the first aspect above is implemented.

[0053] The social media information detection method and device, electronic device and storage medium proposed in the present application obtain target social media information, wherein the target social media information is used to indicate the target push content and the target object with interactive behavior for the target push content, and obtain the comment data generated by the target object before the current time. Then, the graph attention calculation is performed based on the object information of the target object to obtain the object graph feature; the graph attention calculation is performed based on the comment data of the target object to obtain the comment graph feature, and the graph attention mechanism can be used to respectively construct an object graph structure representing the internal relationship of the object information of each target object, and a comment graph structure representing the internal relationship between each comment and the target push content. Further, the attention mechanism is used to combine the target content feature, the object graph feature, and the comment graph feature of the target push content to perform attention calculation to obtain the target social media feature vector, which can realize the feature construction of the target social media information using multi-dimensional feature information, and improve the accuracy and completeness of the feature information of the target social media feature vector used for detection. Finally, the target push content is detected and processed based on the target social media feature vector to obtain the detection result of the target push content, and the detection result is used to indicate whether the target push content is true, which can improve the accuracy of detecting abnormal social media information. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 is a flow chart of a social media information detection method provided in an embodiment of the present application;

[0055] Figure 2 yes Figure 1 Flow chart of step S103 in FIG.

[0056] Figure 3 yes Figure 2 Flow chart of step S203 in FIG.

[0057] Figure 4 yes Figure 2 Flow chart of step S204 in FIG.

[0058] Figure 5 yes Figure 1 Flow chart of step S104 in FIG.

[0059] Figure 6 yes Figure 1 Flow chart of step S105 in FIG.

[0060] Figure 7 yes Figure 1 Flow chart of step S106 in FIG.

[0061] Figure 8 is a structural diagram of a social media information detection device provided in an embodiment of the present application;

[0062] Fig. 9 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0064] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0065] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0066] The social media information detection method and device, electronic device and storage medium provided in the embodiments of the present application are specifically described through the following embodiments. First, the social media information detection method in the embodiments of the present application is described.

[0067] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0068] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0069] The social media information detection method provided in the embodiment of the present application relates to the fields of financial technology and artificial intelligence technology. The social media information detection method provided in the embodiment of the present application can be applied to a terminal, can be applied to a server side, or can be software running in a terminal or a server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or can be configured as a server cluster or a distributed system composed of multiple physical servers, or can be configured as a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the social media information detection method, etc., but is not limited to the above forms.

[0070] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0071] It should be noted that in each specific implementation of the present application, when it comes to the need to perform relevant processing based on data related to the identity or characteristics of the object, such as object information, object behavior data, object historical data, and object location information, the permission or consent of the object will be obtained first, and the collection, use, and processing of these data will comply with relevant laws, regulations, and standards. In addition, when the embodiment of the present application needs to obtain the personal information of the object, the separate permission or separate consent of the object will be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the separate permission or separate consent of the object, the necessary object-related data for the normal operation of the embodiment of the present application will be obtained.

[0072] Figure 1 is an optional flowchart of the social media information detection method provided in an embodiment of the present application. Figure 1 The method may include but is not limited to steps S101 to S106.

[0073] Step S101, obtaining target social media information;

[0074] Step S102, obtaining comment data generated by the target object before the current time;

[0075] Step S103, performing graph attention calculation based on the object information of the target object to obtain object graph features;

[0076] Step S104, performing graph attention calculation based on the comment data of the target object to obtain comment graph features;

[0077] Step S105, performing attention calculation based on the target content features, object graph features, and comment graph features of the target push content acquired in advance to obtain a target social media feature vector;

[0078] Step S106: Detect and process the target pushed content based on the target social media feature vector to obtain a detection result of the target pushed content.

[0079] The target social media information is used to indicate the target push content and the target object that has interactive behavior with respect to the target push content. The detection result is used to indicate whether the target push content is authentic.

[0080] Steps S101 to S106 shown in the embodiment of the present application are obtained by obtaining target social media information, wherein the target social media information is used to indicate the target push content and the target object with which there is an interactive behavior for the target push content, and obtaining the comment data generated by the target object before the current time. Then, the graph attention calculation is performed based on the object information of the target object to obtain the object graph feature; the graph attention calculation is performed based on the comment data of the target object to obtain the comment graph feature, and the graph attention mechanism can be used to respectively construct an object graph structure representing the intrinsic relationship of the object information of each target object, and a comment graph structure representing the intrinsic connection between each comment and the target push content. Further, using the attention mechanism, the attention calculation is performed in combination with the target content feature, the object graph feature, and the comment graph feature of the target push content to obtain the target social media feature vector, which can realize the feature construction of the target social media information using multi-dimensional feature information, and improve the accuracy and completeness of the feature information of the target social media feature vector used for detection. Finally, the target push content is detected and processed based on the target social media feature vector to obtain the detection result of the target push content, and the detection result is used to indicate whether the target push content is true, which can improve the detection accuracy of abnormal social media information.

[0081] It should be noted that the social media information detection method of the embodiment of the present application can be implemented by a preset detection model, and the detection model is mainly composed of a coding network, a first graph attention network, a second graph attention network, an attention network and a linear layer. Among them, the coding network is constructed based on the DeBERTa model. Specifically, the coding network is used to encode the target pushed content to obtain the target content features. The first graph attention network is used to perform graph attention calculation based on the object information of the target object to obtain the object graph features; the second graph attention network is used to perform graph attention calculation based on the comment data of the target object to obtain the comment graph features. The attention network is used to perform attention calculation based on the target content features, object graph features, and comment graph features to obtain the target social media feature vector. The linear layer is used to detect and process the target pushed content based on the target social media feature vector to obtain the detection result of the target pushed content.

[0082] In step S101 of some embodiments, target social media information is obtained, wherein the target social media information is used to indicate target pushed content and a target object that has an interactive behavior with respect to the target pushed content.

[0083] The target social media information refers to the target push content published by a certain object on a social platform or a multimedia platform, and a collection of all information formed based on the target push content.

[0084] Target push content refers to text, video, audio and other content published by a certain object on a social platform or multimedia platform.

[0085] The target object refers to the user who browses the target pushed content on the social platform or multimedia platform and interacts with the target pushed content.

[0086] Interactive behaviors include but are not limited to clicks, comments, forwarding, favorites, and likes by the target object.

[0087] In step S102 of some embodiments, comment data generated by the target object before the current time is obtained.

[0088] Comment data refers to the specific content of comments published by the target object on social platforms or multimedia platforms for various pushed contents before the current time. The comment data includes the content identifier of the pushed content commented by the target object, the comment time, and the content of the comment, etc.

[0089] When this embodiment is implemented, since the background server of the social platform or multimedia platform will record the log data generated by all the behaviors of each target object on the social platform or multimedia platform, based on this, all the comment information generated by the target object before the current time is retrieved from the background server of the social platform or multimedia platform with authorization and permission, and the retrieved comment information is integrated into comment data.

[0090] In step S103 of some embodiments, graph attention calculation is performed based on the object information of the target object to obtain object graph features.

[0091] Object information refers to the name, terminal address, registration time, location and other information of the target object.

[0092] The object graph feature is used to indicate feature information reflected by a graph structure constructed by object information of each target object that has generated an interactive behavior with respect to the target pushed content.

[0093] See also Figure 2 In some embodiments, step S103 may include but is not limited to steps S201 to S205:

[0094] Step S201, constructing an object node of an object graph structure based on each target object, and constructing an object edge of the object graph structure based on interaction information between every two target objects, so as to generate an object graph structure;

[0095] Step S202, in the object graph structure, based on the object information of each target object, determining the initial object node feature of each object node, and based on the interaction information between the target objects corresponding to two object nodes associated with each object edge, determining the object edge weight feature of the object edge;

[0096] Step S203, for a first node in the object nodes, performing attention calculation on the first node based on an initial object node feature of the first node, an initial object node feature of a second node other than the first node, and an object edge weight feature of an object edge connecting the first node and the second node, to obtain a first attention coefficient;

[0097] Step S204, updating the initial object node feature of the first node based on the first attention coefficient to obtain the target object node feature of the first node;

[0098] Step S205 , average pooling is performed on target object node features of multiple first nodes to obtain object graph features.

[0099] Steps S201 to S205 are described in detail below.

[0100] In step S201 of some embodiments, the object graph structure refers to a structural form that represents the object information of the target object related to the target pushed content in the form of a graph structure, where the object node is used to indicate a single target object and the object edge is used to indicate whether there is interaction between the target objects and the intensity of the interaction.

[0101] The interaction information between target objects is used to reflect whether there is interaction between target objects, as well as the frequency or intensity of interaction between target objects.

[0102] When this embodiment is specifically implemented, first, for each target object, each target object is used as an object node of the object graph structure. Then, based on the interaction information between the target objects, an object is constructed between two target objects with interactive behaviors. At the same time, the number of interactions between the two target objects is extracted from the interaction information, and the number of interactions is input into a preset evaluation function. The preset evaluation function outputs a numerical value, which is used as the edge weight of the object edge. The value range of the edge weight is (0,1), and the preset evaluation function is an increasing function with the number of interactions as the independent variable and the edge weight as the dependent variable. Based on this, after constructing the object nodes and object edges, and assigning each object edge a corresponding edge weight, a relatively complete object graph structure is generated.

[0103] In step S202 of some embodiments, the initial object node features are used to indicate feature information of the target object corresponding to each object node. The object edge weight features are used to indicate the interaction strength between the target objects corresponding to two object nodes associated with the object edge in the form of a potential vector.

[0104] In the specific implementation of this embodiment, first, for each object node in the object graph structure, the object information of the target object corresponding to the object node is mapped to the latent vector space, and the object information is projected from the data space to the vector space to obtain the initial object node feature of the object node. Then, for each object edge, the edge weight is projected from the data space to the latent vector space to obtain the object edge weight feature of the object edge.

[0105] In step S203 of some embodiments, the first attention coefficient is used to indicate the degree of attention that the first node pays to the second node.

[0106] The first node refers to any one of the multiple object nodes in the object graph structure, and the second node refers to the other object nodes in the multiple object nodes in the object graph structure except the third node.

[0107] To save space, the specific process of determining the first attention coefficient in the embodiment of the present application will be described in detail below.

[0108] In step S204 of some embodiments, the target object node feature is used to indicate comprehensive feature information of other target objects associated with the target object corresponding to the first node.

[0109] To save space, the specific process of updating the feature of the initial object node of the first node based on the first attention coefficient in the embodiment of the present application will be described in detail below.

[0110] In step S205 of some embodiments, since the first node refers to any object node of the object graph structure, each object node has a corresponding target object node feature. Based on this, first, the object nodes of the object graph structure are counted to obtain the total number of object nodes. Then, the target object node features of each object node are feature spliced ​​to obtain an object splicing feature. Further, the object splicing feature is divided by the total number of object nodes to achieve average pooling of the target object node features of multiple first nodes to obtain an object graph feature.

[0111] Among them, the object graph feature of the embodiment of the present application can be expressed as follows:

[0112]

[0113] Among them, U is the object graph feature, N is the total number of object nodes in the object graph structure, and u i ′ is the target object node feature of the i-th object node in the object graph structure. i and N are positive integers, and i is less than N.

[0114] The benefit of this embodiment is that, for all target objects that have interacted with the target pushed content, an object graph structure is constructed based on the object information of the target objects and the interaction between the target objects. The complex interactive relationships between the target objects can be effectively captured based on the object graph structure, thereby analyzing the characteristics of the overall object group of target objects that have interacted with the target pushed content, and utilizing the overall object characteristics of the target object group to assist in detecting the authenticity of the target pushed content, thereby improving the accuracy of detecting abnormal social media information.

[0115] See also Figure 3 In some embodiments, step S203 may include but is not limited to steps S301 to S303:

[0116] Step S301, performing feature splicing on the initial object node feature of the first node and the initial object node feature of the second node based on a preset weight matrix to obtain an object node splicing feature;

[0117] Step S302, performing attention calculation on the object node splicing features and the object edge weight features to obtain the object node fusion features;

[0118] Step S303, normalize the activation result of the object node fusion feature to obtain the first attention coefficient.

[0119] Steps S301 to S303 are described in detail below.

[0120] In step S301 of some embodiments, a preset weight matrix is ​​used to adjust the weight of feature information of the initial object node features of the first node and the second node in the feature splicing stage. The object node splicing feature is used to indicate the result of splicing the initial object node feature of the first node with the initial object node feature of the single second node.

[0121] In the specific implementation of this embodiment, for each second node, first, a product operation is performed on the preset weight matrix and the initial object node feature of the first node to achieve feature weighting of the initial object node feature of the first node, and a first feature product result is obtained, and a product operation is performed on the preset weight matrix and the initial object node feature of the second node to achieve feature weighting of the initial object node feature of the second node, and a second feature product result is obtained. Then, vector splicing is performed on the first feature product result and the second feature product result to obtain an object node splicing feature.

[0122] In step S302 of some embodiments, when performing attention calculation on the object node splicing feature and the object edge weight feature, a dot product operation is performed on the transposed results of the object node splicing feature and the object edge weight feature to obtain the object node fusion feature.

[0123] In step S303 of some embodiments, when the activation result of the object node fusion feature is normalized, first, for each second node, the object node fusion feature of the second node is activated using a preset activation function to obtain the activation result of the object node fusion feature. Next, a preset exponential function is used to perform exponential calculation with a natural constant as the base and the activation result of the object node fusion feature as the exponent to obtain the node score of the second node. Furthermore, since each second node has a node score, based on this, the node scores of all second nodes of the first node are summed to obtain the total node score. Finally, for each second node, the node score is divided by the total node score, and the division result is used as the first attention coefficient of the second node, thereby achieving normalization of the activation result of the object node fusion feature.

[0124] Among them, the first attention coefficient of the embodiment of the present application can be expressed as follows:

[0125]

[0126] Among them, α ij is the first attention coefficient of the ith first node for the jth second node (any object node among multiple object nodes except the ith object node), i and j are both positive integers, i and j are both less than N, LeakyReLU is the preset activation function, exp(LeakyReLU(a T [Wu i ∥Wu j ])) is the node score of the jth second node; ∥ refers to the feature concatenation operation, LeakyReLU(a T [Wu i ∥Wu j ]) is the activation result of the object node fusion feature of the jth second node, a T [Wu i ∥Wu j ] is the object node fusion feature of the jth second node, a is the object edge weight feature of the object edge between the i-th first node and the j-th second node. W is the preset weight matrix, u i is the initial object node feature of the i-th first node, u j is the initial object node feature of the j-th second node. is the node set consisting of all second nodes of the i-th first node. k is the initial object node feature of the kth second node in the node set consisting of all second nodes of the ith first node. k is a positive integer, and k is less than N.

[0127] It should be noted that the total number of nodes in the node set composed of all second nodes of any first node is 1 less than the total number of object nodes in the object graph structure. When there is no object edge between any two object nodes, the edge weights of these two object nodes that are not connected by an object edge are regarded as 0.

[0128] The advantage of this embodiment is that it introduces a method that combines feature splicing, feature activation, exponential calculation and normalization operations to calculate the first attention coefficient of the first node for each second node, which can more truly and accurately reflect the degree of attention of each target object to other target objects, which is conducive to improving the calculation accuracy of the first attention coefficient.

[0129] See also Figure 4 In some embodiments, step S204 may include but is not limited to steps S401 to S403:

[0130] Step S401, performing matrix multiplication on a first attention coefficient of a first node for a single second node, an initial object node feature of the single second node, and a preset weight matrix to obtain a product feature vector;

[0131] Step S402, based on the number of nodes of the second node, perform feature summation on multiple product feature vectors to obtain adjacent object features of the first node;

[0132] Step S403: Activate the adjacent object features to obtain the target object node features.

[0133] Steps S401 to S403 are described in detail below.

[0134] In step S401 of some embodiments, for a single second node, the initial object node feature of the second node, the first attention coefficient of the first node for the second node, and the preset weight matrix are matrix multiplied to obtain a product feature vector corresponding to the second node.

[0135] In step S402 of some embodiments, the number of nodes is used to indicate the total number of other object nodes in the object graph structure except the first node. The adjacent object feature is used to indicate the result of jointly characterizing the feature characterization of the first node through the object features of other object nodes in the object graph structure except the first node.

[0136] When this embodiment is implemented, first, for the first node, the total number of second nodes corresponding to the first node is counted, and the total number refers to the total number of object nodes other than the first node in the object graph structure. Then, since each second node has a corresponding product feature vector, the total number of product feature vectors can be feature summed, and the total number of product feature vectors can be vector-added to obtain the adjacent object feature of the first node.

[0137] In step S403 of some embodiments, a preset linear activation function is used to activate the adjacent object features to achieve a linear transformation of the adjacent object features, thereby obtaining the target object node features of the first node.

[0138] The target object node feature of the embodiment of the present application can be expressed as follows:

[0139]

[0140] Among them, u i ′ is the target object node feature of the i-th object node in the object graph structure. σ is a linear activation function, W is a preset weight matrix; u jis the initial object node feature of the jth object node (the second node) in the object graph structure. ij Wu j is the product feature vector of the i-th object node (first node) for the j-th second node; α ij It is the first attention coefficient of the i-th object node (first node) with respect to the j-th second node (any object node among multiple object nodes except the i-th object node). is the adjacent object feature. It is a set formed by all object nodes except the i-th object node among multiple object nodes.

[0141] The benefit of this embodiment is that, based on the first attention coefficient of the first node to each second node and the initial object node feature of the second node, the product feature vector of each second node is calculated, and the product feature vectors of multiple second nodes are integrated into feature information of adjacent object features of the first node, which can improve the completeness and accuracy of the feature information of adjacent object features, thereby determining the target object node features of the first node based on the adjacent object features, which can improve the feature accuracy of the target object node features.

[0142] In step S104 of some embodiments, graph attention calculation is performed based on the comment data of the target object to obtain comment graph features.

[0143] The comment graph feature is used to indicate feature information reflected by a graph structure constructed by comment data information of each target object that has generated interactive behaviors on the target pushed content.

[0144] In the embodiment of the present application, the comment data includes each comment information generated by the target object before the current time, and the correlation between each two pieces of comment information.

[0145] Comment information refers to the comments generated by the target object on a certain push content before the current time.

[0146] The relevance is used to indicate the degree of relevance between each two review information posted by the target object.

[0147] In an embodiment of the present application, the correlation between each two comment information can be determined based on the similarity of the content types of the pushed content on which each two comment information is based, or the similarity of the specific content of the pushed content on which the two comment information is based, or the similarity of the content of the two comment information itself.

[0148] See also Figure 5 In some embodiments, step S104 may include but is not limited to steps S501 to S505:

[0149] Step S501, constructing a comment node of a comment graph structure based on each comment information, and constructing a comment edge of the comment graph structure based on the correlation between each pair of comment information, so as to generate a comment graph structure;

[0150] Step S502, in the comment graph structure, based on each comment information, determining the initial comment node feature of each comment node, and based on the correlation between the comment information corresponding to two comment nodes associated with each comment edge, determining the comment edge weight feature of the comment edge;

[0151] Step S503, for a third node in the comment nodes, performing attention calculation on the third node based on the initial comment node feature of the third node, the initial comment node feature of the fourth node other than the third node, and the comment edge weight feature of the object edge connecting the third node and the fourth node, to obtain a second attention coefficient;

[0152] Step S504, updating the initial comment node feature of the third node based on the second attention coefficient to obtain the target comment node feature of the third node;

[0153] Step S505, average pooling is performed on the target comment node features of multiple third nodes to obtain comment graph features.

[0154] Steps S501 to S505 are described in detail below.

[0155] In step S501 of some embodiments, the comment graph structure refers to a structural form in which multiple comment data generated by a target object are represented in the form of a graph structure. The comment node is used to indicate a single comment information published by the target object. The comment edge is used to indicate whether there is a correlation between the comment information and the quantitative result of the correlation between the comment information.

[0156] When this embodiment is implemented, the specific process of step S501 is similar to the above step S201. The difference is that step S201 generates an object graph structure based on the object information of multiple target objects, while step S501 generates a comment graph structure unique to each target object based on multiple comment information of the target object. There is only one object graph structure, but there can be multiple comment graph structures. To save space, it will not be repeated.

[0157] In step S502 of some embodiments, the initial comment node feature is used to indicate the feature information possessed by a single comment information. The comment edge weight feature is used to indicate the association strength between two comment information associated with the comment edge in the form of a latent vector.

[0158] In the specific implementation of this embodiment, the specific process of step S502 is similar to the above step S202, and will not be described in detail to save space.

[0159] In step S503 of some embodiments, the second attention coefficient is used to indicate the degree of attention that the third node pays to the fourth node.

[0160] The third node refers to any one of the multiple comment nodes in the comment graph structure, and the fourth node refers to other comment nodes in the multiple comment nodes in the comment graph structure except the third node.

[0161] In the specific implementation of this embodiment, the specific process of step S503 is similar to the above steps S301 to S303, and will not be described in detail to save space.

[0162] In step S504 of some embodiments, the target comment node feature is used to indicate comprehensive feature information of other comment information associated with the single comment information corresponding to the third node.

[0163] In the specific implementation of this embodiment, the specific process of step S504 is similar to the above steps S401 to S403, and will not be described in detail to save space.

[0164] In step S505 of some embodiments, the specific process of step S505 is similar to the above-mentioned step S205. The difference is that there is only one object graph structure in step S205, and the object graph feature is obtained by average pooling the target object node features of multiple first nodes based on this object graph structure; while there are multiple comment graph structures in step S505, and for each comment graph structure, average pooling is performed based on the target comment node features of multiple third nodes of this comment graph structure to obtain the comment graph features (preliminary graph features) of this comment graph structure, and the comment graph features (preliminary graph features) of each comment graph structure need to be average pooled again to obtain the final comment graph features. To save space, it will not be repeated.

[0165] It should be noted that when generating the comment graph structure, for each target object, a comment graph structure of the target object is generated based on the comment information and correlation of the target object, so that each target object has a corresponding comment graph structure. After generating the comment graph structure of each target object, the same operation is performed on each comment graph structure to obtain the comment graph features corresponding to each comment graph structure. Finally, the comment graph features corresponding to each comment graph structure are averaged and pooled to obtain the final comment graph features used for detection.

[0166] The benefit of this embodiment is that, for all target objects that have interacted with the target pushed content, for each target object, a comment graph structure for each target object is constructed based on the comment information generated by the target object in the past and the correlation between the comment information. The intrinsic connection between the comment information of the target object can be effectively captured based on the comment graph structure, so as to extract some feature information in the comment information of the target object that is helpful for determining the authenticity of the target pushed content, and integrate the useful comment information of all target objects to form the overall comment features of the target object group to assist in detecting the authenticity of the target pushed content, thereby improving the detection accuracy of abnormal social media information.

[0167] In step S105 of some embodiments, attention calculation is performed based on the target content features, object graph features, and comment graph features of the pre-acquired target pushed content to obtain a target social media feature vector.

[0168] The target content feature of the target pushed content is used to indicate the representation of the main content of the target pushed content in the latent vector space.

[0169] The target content features of the target pushed content are obtained by inputting the target pushed content into the DeBERTa model for encoding processing to capture the deep-level features of the text information of the target pushed content.

[0170] The target social media feature vector is used to indicate feature information determined by comprehensively combining the target push content, comment information and object information, which can be used to detect whether the target push content is authentic.

[0171] See also Figure 6 In some embodiments, step S105 may include but is not limited to steps S601 to S605:

[0172] Step S601, performing feature projection on the object graph features based on a preset query projection matrix to obtain a target query feature vector;

[0173] Step S602, performing feature projection on the review image features based on a preset value projection matrix to obtain a target value feature vector;

[0174] Step S603, performing feature projection on the target content feature based on a preset key projection matrix to obtain a target key feature vector;

[0175] Step S604, performing attention calculation on the target key feature vector and the target query feature vector to obtain an attention weight;

[0176] Step S605, performing weighted calculation on the attention weight and the target value feature vector to obtain the target social media feature vector.

[0177] Steps S601 to S605 are described in detail below.

[0178] In step S601 of some embodiments, the preset query projection matrix refers to a feature matrix that provides weight parameters of the query channel in the attention calculation. The target query feature vector is used to indicate feature information related to judging the authenticity of the target pushed content in the object information of each object that has an interactive behavior with the target pushed content.

[0179] In a specific implementation of this embodiment, when feature projection is performed on the object graph feature based on the preset query projection matrix, the query projection matrix and the object graph feature are rectangularly multiplied to obtain a target query feature vector.

[0180] In step S602 of some embodiments, the preset value projection matrix refers to a feature matrix that provides weight parameters of the value channel in the attention calculation. The target value feature vector is used to indicate feature information related to judging the authenticity of the target pushed content in the comment information of each object that has an interactive behavior with respect to the target pushed content.

[0181] In a specific implementation of this embodiment, when feature projection is performed on the comment image features based on a preset value projection matrix, rectangular multiplication is performed on the value projection matrix and the comment image features to obtain a target value feature vector.

[0182] In step S603 of some embodiments, the preset key projection matrix refers to a feature matrix that provides weight parameters of key channels in attention calculation. The target key feature vector is used to indicate feature information in the specific content of the target push content that is related to determining whether the target push content is authentic.

[0183] In a specific implementation of this embodiment, when feature projection is performed on the target content feature based on the preset key projection matrix, rectangular multiplication is performed on the key projection matrix and the target content feature to obtain a target key feature vector.

[0184] In step S604 of some embodiments, the attention weight is used to indicate the degree of attention paid by each object to the authenticity of the content of the target pushed content.

[0185] In the specific implementation of this embodiment, when the attention calculation is performed on the target key feature vector and the target query feature vector, first, the transposed result of the target query feature vector and the target key feature vector is multiplied to obtain a product result; then, the product result is divided by the square difference of the feature dimension of the target value feature vector to obtain a division result. Finally, the division result is normalized using the softmax function to obtain an attention weight.

[0186] In step S605 of some embodiments, a weighted calculation is performed on the attention weight and the target value feature vector, and the attention weight and the target value feature vector are multiplied to obtain a target social media feature vector.

[0187] The benefit of this embodiment is that, based on the attention mechanism, in-depth feature extraction of target push content, comment information and object information can be performed, which can realize comprehensive analysis of the authenticity of social media information in multiple dimensions, thereby improving the detection accuracy of abnormal and non-authentic social media information.

[0188] In step S106 of some embodiments, the target pushed content is detected based on the target social media feature vector to obtain a detection result of the target pushed content, wherein the detection result is used to indicate whether the target pushed content is authentic.

[0189] See also Figure 7 In some embodiments, step S106 may include but is not limited to steps S701 to S704:

[0190] Step S701, mapping the target social media feature vector to a preset feature vector space to obtain a mapped social media feature vector;

[0191] Step S702, performing probability calculation on the mapped social media feature vector based on a preset function to obtain abnormal probability data of the target pushed content;

[0192] Step S703: if it is determined that the abnormal probability data is less than or equal to the preset threshold, the detection result is determined as the target pushed content is authentic;

[0193] Step S704: If it is determined that the abnormal probability data is greater than a preset threshold, the detection result is determined as the target pushed content is not authentic.

[0194] Steps S701 to S704 are described in detail below.

[0195] In step S701 of some embodiments, the preset feature vector space refers to a feature space used to reduce the dimension of the target social media feature vector.

[0196] In the specific implementation of this embodiment, the target social media feature vector is mapped to a preset feature vector space, and the target social media feature vector is matrix multiplied using a preset feature matrix of the preset feature vector space to convert the target social media feature vector into a one-dimensional vector to obtain a mapped social media feature vector.

[0197] In step S702 of some embodiments, the preset function refers to a softmax function, and the abnormal probability data is used to represent the probability size of the target pushed content being true in a mathematical dimension.

[0198] In the specific implementation of this embodiment, the mapped social media feature vector is input into a preset function, the preset function generates a binary probability distribution for the target pushed content, and the abnormal probability data of the target pushed content is obtained according to the binary probability distribution.

[0199] In step S703 of some embodiments, the preset threshold is used to measure the abnormal probability data of the target pushed content and under what circumstances the target pushed content will be determined to be authentic.

[0200] In the specific implementation of this embodiment, the abnormal probability data is compared with a preset threshold. If it is determined that the abnormal probability data is less than or equal to the preset threshold, indicating that the content of the target push content is basically true, the detection result is determined as the target push content is true.

[0201] In step S704 of some embodiments, the abnormal probability data is compared with a preset threshold. If it is determined that the abnormal probability data is greater than the preset threshold, indicating that the target pushed content is likely to be fictitious, the detection result is determined as the target pushed content is not authentic.

[0202] The benefit of this embodiment is that the target social media feature vector is mapped to a low-dimensional feature space to obtain a mapped social media feature vector containing richer feature information, and a softmax function is used to calculate the abnormal probability data of the target pushed content based on the mapped social media feature vector, thereby using a threshold comparison method to obtain the detection result of the target pushed content based on the size relationship between the abnormal probability data and the preset threshold, which can improve the detection accuracy and efficiency of untrue target pushed content.

[0203] See also Figure 8 The embodiment of the present application also provides a social media information detection device, which can implement the above social media information detection method, and the device includes:

[0204] The first acquisition unit 801 is used to acquire target social media information, wherein the target social media information is used to indicate target push content and target objects that have interactive behaviors with respect to the target push content;

[0205] The second acquisition unit 802 is used to acquire the comment data generated by the target object before the current time;

[0206] A first calculation unit 803 is used to perform graph attention calculation based on object information of the target object to obtain object graph features;

[0207] The second calculation unit 804 is used to perform graph attention calculation based on the comment data of the target object to obtain comment graph features;

[0208] The third calculation unit 805 is used to perform attention calculation based on the target content features, object graph features, and comment graph features of the pre-acquired target push content to obtain a target social media feature vector;

[0209] The detection unit 806 is used to perform detection processing on the target pushed content based on the target social media feature vector to obtain a detection result of the target pushed content, wherein the detection result is used to indicate whether the target pushed content is authentic.

[0210] In some embodiments, the first computing unit 803 includes:

[0211] A generation module (not shown), configured to construct object nodes of an object graph structure based on each target object, and to construct object edges of the object graph structure based on interaction information between every two target objects, so as to generate an object graph structure;

[0212] a determination module (not shown) for determining, in the object graph structure, an initial object node feature of each object node based on object information of each target object, and determining an object edge weight feature of the object edge based on interaction information between target objects corresponding to two object nodes associated with each object edge;

[0213] a calculation module (not shown), configured to perform attention calculation on a first node among the object nodes based on an initial object node feature of the first node, initial object node features of second nodes other than the first node, and an object edge weight feature of an object edge connecting the first node and the second node, to obtain a first attention coefficient, wherein the first attention coefficient is used to indicate the degree of attention of the first node to the second node;

[0214] An updating module (not shown), configured to update the initial object node feature of the first node based on the first attention coefficient to obtain the target object node feature of the first node;

[0215] A pooling module (not shown) is used to perform average pooling on target object node features of multiple first nodes to obtain object graph features.

[0216] In some embodiments, a computing module (not shown) is used to:

[0217] Based on a preset weight matrix, feature splicing is performed on the initial object node feature of the first node and the initial object node feature of the second node to obtain an object node splicing feature;

[0218] Perform attention calculation on the object node splicing features and object edge weight features to obtain the object node fusion features;

[0219] The activation results of the object node fusion features are normalized to obtain the first attention coefficient.

[0220] In some embodiments, an update module (not shown) is used to:

[0221] Performing matrix multiplication on a first attention coefficient of the first node for a single second node, an initial object node feature of the single second node, and a preset weight matrix to obtain a product feature vector;

[0222] Based on the number of nodes of the second node, feature summing is performed on a plurality of product feature vectors to obtain adjacent object features of the first node;

[0223] The adjacent object features are activated to obtain the target object node features.

[0224] In some embodiments, the comment data includes each comment information generated by the target object before the current time, and the correlation between each two comment information;

[0225] The second calculation unit 804 is used for:

[0226] Constructing a comment node of the comment graph structure based on each comment information, and constructing a comment edge of the comment graph structure based on the correlation between each two comment information, so as to generate a comment graph structure;

[0227] In the comment graph structure, based on each comment information, the initial comment node feature of each comment node is determined, and based on the correlation between the comment information corresponding to the two comment nodes associated with each comment edge, the comment edge weight feature of the comment edge is determined;

[0228] For a third node in the comment nodes, an attention calculation is performed on the third node based on an initial comment node feature of the third node, an initial comment node feature of a fourth node other than the third node, and a comment edge weight feature of an object edge connecting the third node and the fourth node, to obtain a second attention coefficient, wherein the second attention coefficient is used to indicate the degree of attention of the third node to the fourth node;

[0229] Based on the second attention coefficient, the initial comment node feature of the third node is updated to obtain the target comment node feature of the third node;

[0230] The target comment node features of multiple third nodes are average pooled to obtain the comment graph features.

[0231] In some embodiments, the third computing unit 805 is used to:

[0232] Perform feature projection on the object graph features based on the preset query projection matrix to obtain the target query feature vector;

[0233] Perform feature projection on the review graph features based on the preset value projection matrix to obtain the target value feature vector;

[0234] Perform feature projection on the target content features based on a preset key projection matrix to obtain a target key feature vector;

[0235] Perform attention calculation on the target key feature vector and the target query feature vector to obtain the attention weight;

[0236] The attention weight and target value feature vector are weightedly calculated to obtain the target social media feature vector.

[0237] In some embodiments, the detection unit 806 is used to:

[0238] Mapping the target social media feature vector to a preset feature vector space to obtain a mapped social media feature vector;

[0239] Based on the preset function, the probability of mapping the social media feature vector is calculated to obtain the abnormal probability data of the target push content;

[0240] If it is determined that the abnormal probability data is less than or equal to the preset threshold, the detection result is determined as the target pushed content is authentic;

[0241] If it is determined that the abnormal probability data is greater than a preset threshold, the detection result is determined as the target pushed content is not authentic.

[0242] The specific implementation of the social media information detection device is basically the same as the specific implementation of the above-mentioned social media information detection method, and will not be repeated here.

[0243] The embodiment of the present application also provides an electronic device, the electronic device comprising: a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for realizing connection and communication between the processor and the memory, and the program is executed by the processor to realize the above-mentioned social media information detection method. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0244] See also Fig. 9 , Fig. 9 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0245] The processor 901 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (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 the present application;

[0246] The memory 902 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store an operating system and other applications. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 902, and the processor 901 calls and executes the social media information detection method of the embodiment of this application;

[0247] Input / output interface 903, used to implement information input and output;

[0248] Communication interface 904, used to realize communication interaction between the device and other devices, which can be realized by wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);

[0249] A bus 905 that transmits information between various components of the device (e.g., the processor 901, the memory 902, the input / output interface 903, and the communication interface 904);

[0250] The processor 901 , the memory 902 , the input / output interface 903 and the communication interface 904 are connected to each other in communication within the device via a bus 905 .

[0251] An embodiment of the present application also provides a computer-readable storage medium, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the above-mentioned social media information detection method.

[0252] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0253] The social media information detection method, social media information detection device, electronic device and computer-readable storage medium provided by the embodiment of the present application obtain target social media information, wherein the target social media information is used to indicate the target push content and the target object with interactive behavior for the target push content, and obtain the comment data generated by the target object before the current time. Then, the graph attention calculation is performed based on the object information of the target object to obtain the object graph feature; the graph attention calculation is performed based on the comment data of the target object to obtain the comment graph feature, and the graph attention mechanism can be used to respectively construct an object graph structure representing the internal relationship of the object information of each target object, and a comment graph structure representing the internal relationship between each comment and the target push content. Further, the attention mechanism is used to combine the target content feature, the object graph feature, and the comment graph feature of the target push content to perform attention calculation to obtain the target social media feature vector, which can realize the feature construction of the target social media information using multi-dimensional feature information, and improve the accuracy and completeness of the feature information of the target social media feature vector used for detection. Finally, the target push content is detected and processed based on the target social media feature vector to obtain the detection result of the target push content, and the detection result is used to indicate whether the target push content is true, which can improve the detection accuracy of abnormal social media information.

[0254] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0255] It can be understood by those skilled in the art that Figure 1-7 The technical solutions shown in the figure do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than those shown in the figure, or a combination of certain steps, or different steps.

[0256] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0257] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0258] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0259] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0260] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0261] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0262] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0263] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0264] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A social media information detection method, characterized in that: The social media information detection method comprises: Acquire target social media information, wherein the target social media information is used to indicate target push content and target objects that have interactive behaviors with respect to the target push content; Obtain the comment data generated by the target object before the current time; Performing graph attention calculation based on the object information of the target object to obtain object graph features; Performing graph attention calculation based on the comment data of the target object to obtain comment graph features; Performing attention calculation based on the pre-acquired target content features of the target pushed content, the object graph features, and the comment graph features to obtain a target social media feature vector; The target pushed content is detected and processed based on the target social media feature vector to obtain a detection result of the target pushed content, wherein the detection result is used to indicate whether the target pushed content is authentic.

2. The method according to claim 1, characterized in that The performing graph attention calculation based on the object information of the target object to obtain the object graph feature includes: Constructing an object node of an object graph structure based on each of the target objects, and constructing an object edge of the object graph structure based on interaction information between every two of the target objects, so as to generate the object graph structure; In the object graph structure, based on the object information of each target object, an initial object node feature of each object node is determined, and based on the interaction information between the target objects corresponding to two object nodes associated with each object edge, an object edge weight feature of the object edge is determined; For a first node among the object nodes, performing attention calculation on the first node based on an initial object node feature of the first node, an initial object node feature of a second node other than the first node, and an object edge weight feature of the object edge connecting the first node and the second node, to obtain a first attention coefficient, wherein the first attention coefficient is used to indicate the degree of attention of the first node to the second node; Performing feature updates on the initial object node feature of the first node based on the first attention coefficient to obtain a target object node feature of the first node; Average pooling is performed on the target object node features of multiple first nodes to obtain the object graph features.

3. The method according to claim 2, characterized in that The step of performing attention calculation on the first node based on the initial object node feature of the first node, the initial object node features of the second nodes other than the first node, and the object edge weight feature of the object edge connecting the first node and the second node to obtain a first attention coefficient includes: Based on a preset weight matrix, feature splicing is performed on the initial object node feature of the first node and the initial object node feature of the second node to obtain an object node splicing feature; Performing attention calculation on the object node splicing feature and the object edge weight feature to obtain the object node fusion feature; The activation result of the object node fusion feature is normalized to obtain the first attention coefficient.

4. The method according to claim 2, characterized in that: The updating the initial object node feature of the first node based on the first attention coefficient to obtain the target object node feature of the first node includes: Performing matrix multiplication on the first attention coefficient of the first node for a single second node, the initial object node feature of the single second node, and a preset weight matrix to obtain a product feature vector; Based on the number of nodes of the second node, performing feature summation on a plurality of the product feature vectors to obtain adjacent object features of the first node; The adjacent object features are activated to obtain the target object node features.

5. The method according to claim 1, characterized in that The comment data includes each comment information generated by the target object before the current time, and the correlation between each two pieces of comment information; The performing graph attention calculation based on the comment data of the target object to obtain comment graph features includes: constructing a comment node of a comment graph structure based on each of the comment information, and constructing a comment edge of the comment graph structure based on the correlation between each two of the comment information, so as to generate the comment graph structure; In the comment graph structure, based on each of the comment information, the initial comment node feature of each of the comment nodes is determined, and based on the correlation between the comment information corresponding to the two comment nodes associated with each of the comment edges, the comment edge weight feature of the comment edge is determined; For a third node among the comment nodes, performing attention calculation on the third node based on an initial comment node feature of the third node, an initial comment node feature of a fourth node other than the third node, and a comment edge weight feature of the object edge connecting the third node and the fourth node, to obtain a second attention coefficient, wherein the second attention coefficient is used to indicate the degree of attention of the third node to the fourth node; Based on the second attention coefficient, the initial comment node feature of the third node is updated to obtain a target comment node feature of the third node; Average pooling is performed on the target comment node features of multiple third nodes to obtain the comment graph features.

6. The method according to claim 1, characterized in that The attention calculation is performed based on the target content features of the target pushed content, the object graph features, and the comment graph features obtained in advance to obtain the target social media feature vector, including: Performing feature projection on the object graph features based on a preset query projection matrix to obtain a target query feature vector; Perform feature projection on the review graph features based on a preset value projection matrix to obtain a target value feature vector; Performing feature projection on the target content feature based on a preset key projection matrix to obtain a target key feature vector; Performing attention calculation on the target key feature vector and the target query feature vector to obtain an attention weight; The attention weight and the target value feature vector are weightedly calculated to obtain the target social media feature vector.

7. The method according to any one of claims 1 to 6, characterized in that: The detecting and processing the target pushed content based on the target social media feature vector to obtain the detection result of the target pushed content includes: Mapping the target social media feature vector to a preset feature vector space to obtain a mapped social media feature vector; Performing probability calculation on the mapped social media feature vector based on a preset function to obtain abnormal probability data of the target pushed content; If it is determined that the abnormal probability data is less than or equal to a preset threshold, the detection result is determined as the target pushed content is authentic; If it is determined that the abnormal probability data is greater than a preset threshold, the detection result is determined as the target pushed content is not authentic.

8. A social media information detection device, characterized in that: The social media information detection device comprises: A first acquisition unit is used to acquire target social media information, wherein the target social media information is used to indicate target push content and a target object that has an interactive behavior with respect to the target push content; A second acquisition unit, used to acquire comment data generated by the target object before the current time; A first computing unit, configured to perform graph attention calculation based on the object information of the target object to obtain object graph features; A second computing unit, configured to perform graph attention computing based on the comment data of the target object to obtain comment graph features; A third calculation unit is used to perform attention calculation based on the pre-acquired target content features of the target pushed content, the object graph features, and the comment graph features to obtain a target social media feature vector; A detection unit is used to perform detection processing on the target pushed content based on the target social media feature vector to obtain a detection result of the target pushed content, wherein the detection result is used to indicate whether the target pushed content is authentic.

9. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the social media information detection method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the social media information detection method according to any one of claims 1 to 7 is implemented.

Images