Object operation identification method and device, storage medium and electronic equipment

Abstract

The invention provides an object operation identification method and device, a storage medium and electronic equipment. The method comprises the steps that operation information of object operation to be recognized is acquired, and an operation feature vector of the object operation is determined according to the operation information; n projection vectors obtained by projection of the operation feature vectors in N sample spaces are obtained, the N sample spaces correspond to N operation types respectively, and the sample spaces are vector spaces determined according to multiple sample feature vectors of the corresponding operation types; the sample feature vector is a feature vector determined according to the sample information of the object operation sample of the corresponding operation type; determining a target feature vector matched with the object operation according to the N projection vectors, and inputting the target feature vector into a target classification model; and determining a target operation type of the object operation according to the target classification model. The technical problem that the financial risk identification accuracy of object operation is low is solved.

Description

Technical Field

[0001] The present invention relates to the field of computers, and in particular to a method and device for identifying object operations, a storage medium, and an electronic device. Background Art

[0002] On various internet platforms, users can perform various financial-related account operations, such as payments and transfers. Because these operations may involve financial risks (such as payment fraud and money laundering), to ensure user payment security and manage platform risks, various platforms can classify and identify various financial-related account operations. This allows for platform payment fraud detection and anti-money laundering intervention, and, when risky operations are identified, appropriate management or intervention measures are implemented.

[0003] In related fields, a trained classification model can be deployed on a platform server to process operational feature information related to account operations, thereby obtaining a classification result for the account operations and determining whether the current account operations present financial risks. Existing classification methods are typically based on raw operational feature information, which often carries a significant amount of noise. This can lead to inaccurate risk identification results for account operations based on this information.

[0004] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0005] Embodiments of the present invention provide a method and apparatus for identifying object operations, a storage medium, and an electronic device, to at least solve the technical problem of low accuracy in identifying financial risks of user operations in related fields.

[0006] According to one aspect of an embodiment of the present invention, a method for identifying an object operation is provided, comprising: obtaining operation information of an object operation to be identified, and determining an operation feature vector of the object operation based on the operation information, wherein the operation information includes account information and operation behavior information associated with the object operation; obtaining N projection vectors obtained by projecting the operation feature vector onto N sample spaces, wherein the N sample spaces correspond to N operation types, respectively, and the sample space is a vector space determined based on a plurality of sample feature vectors of the corresponding operation type, and the sample feature vector is a feature vector determined based on sample information of the object operation sample of the corresponding operation type, and N is an integer greater than or equal to 1; determining a target feature vector matching the object operation based on the N projection vectors, and inputting the target feature vector into a target classification model; and determining a target operation type of the object operation based on the target classification model.

[0007] According to another aspect of an embodiment of the present invention, an object operation recognition device is also provided, including: a first acquisition unit, used to obtain operation information of the object operation to be identified, and determine the operation feature vector of the object operation based on the above operation information, wherein the above operation information includes account information and operation behavior information associated with the above object operation; a second acquisition unit, used to obtain N projection vectors obtained by projecting the above operation feature vector onto N sample spaces, wherein the N sample spaces correspond to N operation types, respectively, and the above sample space is a vector space determined according to multiple sample feature vectors of the corresponding operation type, and the above sample feature vector is a feature vector determined according to sample information of the object operation sample of the corresponding operation type, and the above N is an integer greater than or equal to 1; a classification unit, used to determine a target feature vector matching the above object operation based on the N projection vectors, and input the above target feature vector into a target classification model; a determination unit, used to determine the target operation type of the above object operation according to the above target classification model.

[0008] Optionally, the second acquisition unit is used to: repeat the following operations until N of the sample spaces are traversed: obtain a principal component matrix matching the current sample space, wherein the principal component matrix is used to indicate the main feature distribution of the multiple sample feature vectors included in the current sample space; determine the current projection vector corresponding to the current sample space based on the operation feature vector and the principal component matrix.

[0009] Optionally, the second acquisition unit is used to: obtain a mean eigenvector matching the current sample space, wherein the mean eigenvector is a eigenvector determined based on the weighted summation result of multiple sample eigenvectors; determine the difference vector between the operation eigenvector and the mean eigenvector as a reference eigenvector; and determine the projection vector based on the transposed matrix of the principal component matrix and the reference eigenvector.

[0010] Optionally, the above-mentioned object operation recognition device also includes: a third acquisition unit, used to obtain a reference sample set consisting of multiple reference operation samples of a reference operation type, wherein the N above-mentioned operation types include the above-mentioned reference operation type; according to the above-mentioned sample information corresponding to each of the multiple reference operation samples, respectively determine the multiple reference sample feature vectors, and construct a reference sample space corresponding to the reference operation type based on the multiple reference sample feature vectors; obtain the spatial feature matrix of the above-mentioned reference sample space, and perform singular value decomposition on the above-mentioned spatial feature matrix to obtain a first reference matrix, a second reference matrix and a third reference matrix, wherein the above-mentioned first reference matrix includes left singular vectors corresponding to the above-mentioned spatial feature matrix, the above-mentioned second reference matrix includes right singular vectors corresponding to the above-mentioned spatial feature matrix, the above-mentioned first reference matrix and the above-mentioned second reference matrix are orthogonal matrices, and the above-mentioned third reference matrix is a diagonal matrix; determine the reference submatrix determined according to the above-mentioned first reference matrix as the above-mentioned principal component matrix matching the above-mentioned reference sample space, wherein the matrix dimension of the above-mentioned reference submatrix is smaller than the matrix dimension of the above-mentioned first reference matrix.

[0011] Optionally, the above-mentioned third acquisition unit is used to: determine the mean feature vector matching the above-mentioned reference sample space based on the weighted summation result between the multiple reference sample feature vectors; obtain the difference vector between the multiple reference sample feature vectors and the above-mentioned mean feature vector in the above-mentioned reference sample space; and determine the above-mentioned spatial feature matrix of the above-mentioned reference sample space based on the splicing result of the multiple difference vectors.

[0012] Optionally, the third acquisition unit is configured to: acquire a singular value sequence composed of singular values in the third reference matrix, and use the sum of squares of each singular value in the singular value sequence as a first reference value, wherein the singular value sequence includes a plurality of singular values arranged in descending order according to numerical values; acquire principal component coefficients, and acquire the product of the first reference value and the principal component coefficients as a second reference value, wherein the principal component coefficients are preset constants greater than 0 and less than 1; when the sum of squares of the first M singular values in the singular value sequence is greater than or equal to the second reference value, and the sum of squares of the first M-1 singular values in the singular value sequence is less than the second reference value, determine the reference submatrix composed of the first M column vectors in the first reference matrix as the principal component matrix, wherein the first reference matrix includes L column vectors, L is an integer greater than 1, and M is an integer greater than or equal to 1 and less than or equal to L.

[0013] Optionally, the above-mentioned classification unit includes one of the following: a first classification module, used to determine the N above-mentioned projection vectors as the above-mentioned target feature vectors that match the above-mentioned object operation; a second classification module, used to determine the above-mentioned reference projection vector as the above-mentioned target feature vector that matches the above-mentioned object operation when the vector modulus of the reference projection vector among the N above-mentioned projection vectors is greater than or equal to the target threshold; a third classification module, used to determine the weighted sum result of the above-mentioned N projection vectors as the above-mentioned target feature vector.

[0014] Optionally, the above-mentioned determination unit includes: an acquisition module for obtaining N type probabilities output by the above-mentioned target classification model based on the above-mentioned target feature vector, wherein the N above-mentioned type probabilities respectively correspond to the N above-mentioned operation types; a determination module for determining the above-mentioned operation type corresponding to the highest above-mentioned type probability among the N above-mentioned type probabilities as the above-mentioned target operation type.

[0015] Optionally, the above-mentioned determination unit is also used to: obtain N sample sets corresponding to the N above-mentioned operation types and a current classification model in a training state, wherein the above-mentioned sample sets include multiple object operation samples, and the above-mentioned object operation samples include sample information and label information; determine N spatial feature matrices according to the N above-mentioned sample sets, wherein the above-mentioned spatial feature matrices are used to characterize the sample features of the above-mentioned object operation samples in the corresponding above-mentioned sample sets; obtain a current operation feature vector of a current object operation sample from the N above-mentioned sample sets, and determine N sample projection vectors according to the above-mentioned current operation feature vector and the N above-mentioned spatial feature matrices; splice the N above-mentioned sample projection vectors to obtain a current sample feature that matches the above-mentioned current object operation sample, and obtain the classification result of the above-mentioned current classification model based on the above-mentioned current sample feature; train the above-mentioned current classification model according to the above-mentioned classification result and the comparison result of the label information of the above-mentioned current object operation sample until the above-mentioned target classification model that meets the material collection conditions is obtained.

[0016] Optionally, the above-mentioned determination unit is used to: obtain an account operation set in the target social platform within the target period, wherein the above-mentioned account operation set includes multiple account interaction operations between multiple object accounts; determine the above-mentioned object operation samples based on the operation information of the above-mentioned account interaction operations and the operation type of the above-mentioned interaction operations, wherein the above-mentioned operation information includes the account information of the above-mentioned object account associated with the above-mentioned interaction operation, and the operation behavior information of the above-mentioned interaction operation; determine N sample sets corresponding to the N types of the above-mentioned operation types according to the classification results of the above-mentioned object operation samples.

[0017] Optionally, the above-mentioned first acquisition unit is used to: when the above-mentioned object operation is an account interaction operation between the first object account and the second object account, obtain the above-mentioned account information of the above-mentioned first object account and the above-mentioned second object account, as well as the above-mentioned operation behavior information of the above-mentioned account interaction operation; map each operation information in the above-mentioned account information and the above-mentioned operation behavior information into an operation feature value, and determine the above-mentioned operation feature vector of the above-mentioned object operation based on multiple above-mentioned operation feature values.

[0018] Optionally, the above-mentioned object operation identification device also includes at least one of the following: an interception unit, used to intercept the above-mentioned account interaction operation between the above-mentioned first object account and the second object account when the above-mentioned target operation type of the above-mentioned object operation is a risk type; a first alarm unit, used to send risk behavior prompt information to the above-mentioned first object account when the above-mentioned target operation type of the above-mentioned object operation is a risk type; a second alarm unit, used to send risk behavior prompt information to the above-mentioned second object account when the above-mentioned target operation type of the above-mentioned object operation is a risk type.

[0019] According to another aspect of the embodiments of the present invention, a computer-readable storage medium is provided, in which a computer program is stored. The computer program is configured to execute the above-mentioned object operation recognition method during runtime.

[0020] According to another aspect of the embodiments of the present application, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the object operation recognition method described above.

[0021] According to another aspect of an embodiment of the present invention, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the object operation recognition method through the computer program.

[0022] In an embodiment of the present invention, operation information of an object operation to be identified is obtained, and an operation feature vector of the object operation is determined based on the operation information; N projection vectors are obtained by projecting the operation feature vector onto N sample spaces, where the N sample spaces correspond to N operation types, respectively. The sample space is a vector space determined based on multiple sample feature vectors of the corresponding operation type, and the sample feature vector is a feature vector determined based on sample information of the object operation sample of the corresponding operation type; a target feature vector matching the object operation is determined based on the N projection vectors, and the target feature vector is input into a target classification model; a target operation type of the object operation is determined based on the target classification model, thereby realizing recognition of the object operation.

[0023] In the above-mentioned object operation recognition method, the operation feature vector determined according to the operation information of the object operation is projected respectively in the sample space corresponding to different operation types, and then a plurality of projection vectors are obtained. It can be understood that the projection operation based on the vector space can significantly determine the degree of matching between the current object operation and different operation types based on the original features, that is, the projection vector obtained by the projection operation indicates the above-mentioned matching degree. At the same time, the above-mentioned projection vector can also exclude feature information that is not related to the corresponding type feature, that is, the projection vector can carry high-density feature information; and then the target feature vector determined based on the projection vector is input into the classification model for classification operation. The classification model can perform efficient classification based on the target feature vector carrying high-density feature information, and at the same time significantly improve the classification effect, thereby solving the technical problem of low accuracy in identifying financial risks of user operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0025] Figure 1 is a schematic diagram of a hardware environment of an optional object operation recognition method according to an embodiment of the present invention;

[0026] Figure 2 is a flow chart of an optional object operation recognition method according to an embodiment of the present invention;

[0027] Figure 3 is a flowchart of a training process of an optional object operation recognition method according to an embodiment of the present invention;

[0028] Figure 4 is a schematic diagram of an optional object operation recognition method according to an embodiment of the present invention;

[0029] Figure 5is a flowchart of another optional object operation recognition method according to an embodiment of the present invention;

[0030] Figure 6 A schematic structural diagram of an optional object operation recognition device according to an embodiment of the present invention;

[0031] Figure 7 FIG. 4 is a schematic structural diagram of an optional electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0032] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0033] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention 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 numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes 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.

[0034] It should be noted that in the relevant implementation methods of this application, in order to fulfill legal obligations such as financial risk control and financial fraud detection, the platform involves the acquisition and processing of relevant information and data of the target account. The relevant information and data involved in this application are all pre-acquired through methods that comply with the provisions of relevant normative legal documents, and before obtaining the above-mentioned relevant information and data, it is necessary to obtain the authorization permission of the corresponding account subject.

[0035] According to one aspect of an embodiment of the present invention, a method for identifying an object operation is provided. As an optional implementation, the method for identifying an object operation can be applied to, but is not limited to, Figure 1 In the object operation recognition system composed of the terminal device 102, the server 104 and the network 110 shown in FIG. Figure 1As shown, the terminal device 102 is connected to and communicates with the server 104 via the network 110. The network may include, but is not limited to, a wired network and a wireless network. The wired network includes a local area network, a metropolitan area network, and a wide area network. The wireless network includes Bluetooth, WIFI, and other networks that enable wireless communication. The terminal device may include, but is not limited to, at least one of the following: a mobile phone (such as an Android phone, an iOS phone, etc.), a laptop computer, a tablet computer, a PDA, an MID (Mobile Internet Device), a PAD, a desktop computer, a smart TV, an in-vehicle device, etc. The terminal device 102 may be installed with a client for providing interactive operation services between object accounts. The client may include, but is not limited to, an instant messaging client, a virtual resource service client, etc.

[0036] The terminal device 102 is also provided with a display, a processor and a memory. The display can be used to display the program interface of the client, and the processor can process and convert the operation information generated by the object operation based on the client; the memory is used to store the operation information generated by the object account in historical operations.

[0037] The server 104 may be a single server, a server cluster consisting of multiple servers, or a cloud server. The server includes a database and a processing engine. The processing engine is configured to execute the target service corresponding to the client and obtain a sample information set to determine a sample space corresponding to each of the multiple operation types.

[0038] According to one aspect of an embodiment of the present invention, the object operation recognition system may further perform the following steps: first, the terminal device 102 performs step S102 to trigger the object operation and generate operation information; then performs step S104 to send the object operation information to the server 104 via the network 110;

[0039] Next, in the server 104, steps S106 to S112 are executed to obtain operation information of the object operation to be identified, and determine an operation feature vector of the object operation based on the operation information, wherein the operation information includes account information and operation behavior information associated with the object operation; obtain N projection vectors obtained by projecting the operation feature vector onto N sample spaces, wherein the N sample spaces correspond to N operation types, respectively, the sample space is a vector space determined based on multiple sample feature vectors of the corresponding operation type, the sample feature vector is a feature vector determined based on sample information of the object operation sample of the corresponding operation type, and N is an integer greater than or equal to 1; determine a target feature vector that matches the object operation based on the N projection vectors, and input the target feature vector into a target classification model; determine a target operation type of the object operation based on the target classification model;

[0040] Next, when the target operation type is a risk type, the server 104 executes step S114 to send an alarm message to the terminal device 102 via the network 110 ; finally, the terminal device 102 executes step S116 to display the alarm message.

[0041] In an optional embodiment, when the above-mentioned target classification model is pre-running in the above-mentioned terminal device 102, the feature vectors for indicating the above-mentioned N sample spaces are pre-stored in the above-mentioned terminal device 102, and then in the process of the object account performing the object operation based on the above-mentioned terminal device 102, the above-mentioned steps S106 to S112 can be implemented through the terminal device 102 to identify the operation type of the object operation in real time, and then when a risky object operation is identified, an alarm message is immediately displayed to issue an alarm prompt.

[0042] In an embodiment of the present invention, operation information of an object operation to be identified is obtained, and an operation feature vector of the object operation is determined based on the operation information; N projection vectors are obtained by projecting the operation feature vector onto N sample spaces, where the N sample spaces correspond to N operation types, respectively. The sample space is a vector space determined based on multiple sample feature vectors of the corresponding operation type, and the sample feature vector is a feature vector determined based on sample information of the object operation sample of the corresponding operation type; a target feature vector matching the object operation is determined based on the N projection vectors, and the target feature vector is input into a target classification model; a target operation type of the object operation is determined based on the target classification model, thereby realizing recognition of the object operation.

[0043] In the above-mentioned object operation recognition method, the operation feature vector determined according to the operation information of the object operation is projected respectively in the sample space corresponding to different operation types, thereby obtaining a plurality of projection vectors. It can be understood that the projection operation based on the vector space can significantly determine the degree of matching between the current object operation and different operation types based on the original features, that is, the projection vector obtained by the projection operation indicates the above-mentioned matching degree. At the same time, the above-mentioned projection vector can also exclude feature information that is not related to the corresponding type feature, that is, the projection vector can carry high-density feature information; and then the target feature vector determined based on the projection vector is input into the classification model for classification operation. The classification model can perform efficient classification based on the target feature vector carrying high-density feature information, while significantly improving the classification effect, thereby solving the technical problem of inaccurate type recognition of existing object operations.

[0044] The above is only an example and is not limited in this embodiment.

[0045] As an optional implementation, Figure 2 As shown, the above object operation recognition method may include the following steps:

[0046] S202, obtaining operation information of the object operation to be identified, and determining an operation feature vector of the object operation based on the operation information, wherein the operation information includes account information associated with the object operation and operation behavior information;

[0047] S204, obtaining N projection vectors obtained by projecting the operation feature vector onto N sample spaces, where the N sample spaces correspond to N operation types, respectively. The sample space is a vector space determined based on multiple sample feature vectors corresponding to the operation type. The sample feature vector is a feature vector determined based on sample information of object operation samples corresponding to the operation type. N is an integer greater than or equal to 1.

[0048] S206, determining a target feature vector that matches the object operation based on the N projection vectors, and inputting the target feature vector into a target classification model;

[0049] S208: Determine the target operation type of the object operation according to the target classification model.

[0050] First, the technical application scenario of the above-mentioned embodiment of the present application is explained. The above-mentioned object operation identification method can be used in a scenario where operation information can be obtained through account operations of an object account. In the case where the above-mentioned scenario is specifically a communication scenario of an instant messaging application, the above-mentioned object operation can be specifically an interactive operation between accounts, such as an instant conversation between accounts, virtual resource sharing between accounts, or media resource sharing between accounts; in the case where the above-mentioned scenario is specifically a virtual interactive scenario in a game application, the above-mentioned object operation can be specifically a virtual game task initiated and completed by a game account, and the acquisition and exchange operations of virtual resources triggered in the above-mentioned game task, and can also be the acquisition and exchange operations of virtual resources between different game accounts in the above-mentioned game task; in the case where the above-mentioned scenario is specifically a resource management scenario of a resource management application, the above-mentioned object operation can be specifically a resource management operation performed by the object account in the above-mentioned resource management application, for example, the transfer and exchange of virtual resources. The above application scenario is only an example and does not limit the specific application scenario of the above-mentioned embodiment of the present application.

[0051] Furthermore, in the above step S202 , the object operation to be identified and the operation information of the object operation may vary depending on the specific application scenario.

[0052] For example, when the above-mentioned application scenario is an instant messaging scenario and the above-mentioned object operation is a resource exchange operation between object accounts, the above-mentioned operation information may include the account information of the two object accounts that perform the above-mentioned resource operation, such as account identification information used to uniquely indicate the object account, account behavior information used to indicate the object operations performed by the object account in the past period of time, etc.; the above-mentioned operation information may further include the operation behavior information of the above-mentioned object operation, such as the operation amount information of the above-mentioned resource operation, operation frequency information, operation time information, etc.

[0053] For another example, when the above-mentioned application scenario is a game application scenario, and the above-mentioned object operation is a virtual game task triggered by a game account, the above-mentioned operation information may include the account information of the object account that performs the above-mentioned virtual game task, for example, account identification information used to uniquely indicate the game account, and account behavior information used to indicate the game operations performed by the object account in the past period of time, such as game win rate, game record in the game task, etc.; the above-mentioned operation information may further include operation behavior information in the above-mentioned game task, for example, the operation parameters of the object account in the above-mentioned game task, such as the number of people defeated, experience value growth rate, skill casting time information within a period of time, etc.

[0054] It can be understood that when the above-mentioned operation information is obtained, since the operation information can further include multiple information items, the information content of each of the multiple information items in the operation information can be converted into corresponding eigenvalues, respectively, so as to splice the eigenvalues corresponding to the multiple information items to obtain the operation feature vector corresponding to the above-mentioned object operation.

[0055] The operation type involved in the above step S204 can be used to indicate the risk level of the above object operation in the above platform, and can also be used to indicate the risk level for the object account.

[0056] For example, in the case where the aforementioned application scenario is an instant messaging scenario, interactive operations between accounts may include, but are not limited to, normal interaction types, first risk types, and second risk types. For example, the first risk type may be an interaction type that violates the platform's management regulations, such as posting inappropriate text or images in the aforementioned communication application, or posting illegal advertising information in the aforementioned communication application; the second risk type may be a type that violates the provisions of relevant regulatory documents, such as an interaction type containing fraudulent content, such as a payment fraud risk type or a black market risk type.

[0057] For another example, in the case where the aforementioned application scenario is a gaming application scenario, the object operations triggered by the gaming account may include, but are not limited to, normal gaming operations, illegal gaming operations, and risky gaming operations. For example, the aforementioned illegal gaming operations may specifically be cheating operations in a game (such as "cheat" behavior) or posting illegal advertising information in the aforementioned gaming application; the aforementioned risky gaming operations may be game resource exchange operations that contain fraudulent content.

[0058] In the above step S204, a sample space constructed by operation samples of multiple operation types can be obtained in advance, and the projection vectors of the above operation feature vectors in the above multiple sample spaces can be obtained. Then, the above projection vectors can be used to indicate the similarity or matching degree between the above operation feature vectors and the operation samples of each operation type, thereby improving the information density based on the original feature information.

[0059] The above-mentioned method of obtaining the projection vector of the operation characteristic vector in N sample spaces may include but is not limited to one of the following: optionally, obtaining a spatial matrix used to indicate the above-mentioned sample space (for example, the spatial matrix may be a spatial matrix obtained by splicing multiple sample characteristic vectors of the same type), and obtaining the above-mentioned projection vector based on the above-mentioned spatial matrix and the above-mentioned operation characteristic vector; for example, after obtaining the spatial matrix indicating the above-mentioned sample space, further obtaining the spatial characteristic vector of the spatial matrix based on the above-mentioned spatial matrix, and then determining the above-mentioned projection vector based on the spatial characteristic vector and the above-mentioned operation characteristic vector.

[0060] In a preferred embodiment, the above-mentioned obtaining of N projection vectors obtained by projecting the operational feature vector onto N sample spaces respectively includes: repeating the following operation until the N sample spaces are traversed:

[0061] S1, obtaining a principal component matrix matching the current sample space, wherein the principal component matrix is used to indicate the main feature distribution of multiple sample feature vectors included in the current sample space;

[0062] S2, determining a current projection vector corresponding to the current sample space according to the operating eigenvector and the principal component matrix.

[0063] It can be understood that in this embodiment, the method of obtaining N projection vectors obtained by projecting the operation feature vector onto N sample spaces can be by traversing the N sample spaces to perform projection operations with the above-mentioned operation feature vector according to the principal component matrices corresponding to each of the multiple sample spaces, thereby obtaining the above-mentioned projection vector.

[0064] In an optional implementation, the above step S206 may include one of the following:

[0065] Method 1: Determine N projection vectors as target feature vectors that match the object operation;

[0066] Method 2: when the vector modulus of the reference projection vector among the N projection vectors is greater than or equal to the target threshold, the reference projection vector is determined as the target feature vector that matches the object operation;

[0067] Method 3: The weighted sum of N projection vectors is determined as the target feature vector.

[0068] In the above-mentioned method 1, the above-mentioned N projection vectors can all be used as target feature vectors matching the object operation, that is, the above-mentioned N projection vectors can be spliced and input into the above-mentioned target classification model;

[0069] In the second method, the projection vectors that meet certain conditions among the N projection vectors can be input as target feature vectors into the target classification model. In the above embodiment, a target threshold value based on the vector modulus of the projection vector can be used as the above condition, and the projection vectors that are greater than or equal to the target threshold value can be used as the target feature vector.

[0070] In a preferred embodiment, when the number of the above-mentioned projection vectors is large, the target feature vector can be determined from the multiple projection vectors by the above-mentioned method 2, thereby simplifying the feature information to be input into the target classification model and improving the model classification efficiency;

[0071] In the third approach, a weighted sum of multiple projection vectors may be determined as the target feature vector. Optionally, the vector weights corresponding to the projection vectors may be positively correlated with the vector moduli of the projection vectors.

[0072] In another preferred embodiment, when the number of the above-mentioned projection vectors is large, the target feature vector can be determined based on multiple projection vectors through the above-mentioned method three, thereby simplifying the feature information to be input into the target classification model and improving the model classification efficiency.

[0073] Finally, as in step S208 , when the target feature vector is determined, the target operation type of the object operation is determined based on the target feature vector according to the target classification model.

[0074] When N in the above steps is 2, the target classification model can be a binary classification model, such as a logistic regression model, a support vector machine model, etc.; when N in the above steps is greater than 2, the target classification model can also be a multi-classification model, such as a decision tree model, a k-nearest neighbor model, etc. In this embodiment, the specific type of the target classification model is not limited.

[0075] In the above-mentioned object operation recognition method, the operation feature vector determined according to the operation information of the object operation is projected respectively in the sample space corresponding to different operation types, thereby obtaining a plurality of projection vectors. It can be understood that the projection operation based on the vector space can significantly determine the degree of matching between the current object operation and different operation types based on the original features, that is, the projection vector obtained by the projection operation indicates the above-mentioned matching degree. At the same time, the above-mentioned projection vector can also exclude feature information irrelevant to the corresponding type feature, that is, the projection vector can carry high-density feature information; and then the target feature vector determined based on the projection vector is input into the classification model for classification operation. The classification model can perform efficient classification based on the target feature vector carrying high-density feature information, while significantly improving the classification effect, thereby solving the technical problem of low accuracy in risk identification of user operations.

[0076] In an optional implementation, the above-mentioned determining the projection vector corresponding to the current sample space according to the operating eigenvector and the principal component matrix includes:

[0077] S1, obtaining a mean feature vector that matches the current sample space, wherein the mean feature vector is a feature vector determined by weighted summation of multiple sample feature vectors;

[0078] S2, determining the difference vector between the operational eigenvector and the mean eigenvector as the reference eigenvector;

[0079] S2, determining the projection vector according to the transposed matrix of the principal component matrix and the reference eigenvector.

[0080] It is understandable that obtaining the projection vector corresponding to each sample space can be achieved through the above steps. The above implementation is described below in conjunction with a specific process.

[0081] First, we can obtain the mean eigenvector of the current sample space matching

[0082] The specific acquisition method is as follows:

[0083] Assume that in the current sample space X p Including n1 samples x i , which is expressed as:

[0084]

[0085] The above mean eigenvector It can be obtained as follows:

[0086]

[0087] According to the operating feature vector x a and the mean eigenvector The reference eigenvector determined by the difference vector between is:

[0088] The principal component matrix in the current sample space is In the case of ap The way to obtain is as follows:

[0089]

[0090] The following describes how to obtain the principal component matrix of each sample space:

[0091] In an optional embodiment, before obtaining the principal component matrix matching the current sample space, the method further includes:

[0092] S1, obtaining a reference sample set consisting of multiple reference operation samples of a reference operation type, wherein the N operation types include the reference operation type;

[0093] S2, determining multiple reference sample feature vectors according to sample information corresponding to each of the multiple reference operation samples, and constructing a reference sample space corresponding to the reference operation type according to the multiple reference sample feature vectors;

[0094] S3, obtaining a spatial characteristic matrix of the reference sample space, and performing singular value decomposition on the spatial characteristic matrix to obtain a first reference matrix, a second reference matrix, and a third reference matrix, wherein the first reference matrix includes left singular vectors corresponding to the spatial characteristic matrix, the second reference matrix includes right singular vectors corresponding to the spatial characteristic matrix, the first reference matrix and the second reference matrix are orthogonal matrices, and the third reference matrix is a diagonal matrix;

[0095] S4. Determine a reference submatrix determined according to the first reference matrix as a principal component matrix matching the reference sample space, wherein the matrix dimension of the reference submatrix is smaller than the matrix dimension of the first reference matrix.

[0096] It should be noted that, in a preferred embodiment, the construction of the above-mentioned N sample spaces and the determination of the principal component matrix corresponding to each sample space can occur before the above-mentioned step S202, that is, the N sample spaces can be constructed in advance, and the principal component matrix corresponding to each of the N sample spaces can be obtained in advance through the above-mentioned implementation method.

[0097] In this embodiment, a sample set for each operation type can be first obtained, and then a corresponding sample space can be constructed based on the sample set for each operation type. In this embodiment, when the reference sample set and the current sample set are the same sample set, the process of obtaining the corresponding principal component matrix based on the reference sample space is described using the reference sample set as an example.

[0098] In this embodiment, the feature vectors x of multiple reference samples are determined respectively. i Afterwards, multiple reference sample feature vectors x i Construct reference sample space X p , where, when the number of reference sample feature vectors is n1,

[0099]

[0100] Then, we can obtain the spatial feature matrix of the reference sample space In the case of , the singular value decomposition of the above spatial feature matrix is performed as follows:

[0101]

[0102] Among them, U P is the first reference matrix mentioned above, ∑ P is the second reference matrix mentioned above, V P is the third reference matrix mentioned above.

[0103] Furthermore, according to the above U P The reference submatrix of Determine the above principal component matrix. It should be noted that the above reference submatrix The dimension is lower than the first reference matrix U P For example, in the first reference matrix U P When M columns of data are included, the above reference submatrix Includes L columns of data, where L can be smaller than M.

[0104] In an optional embodiment, the above-mentioned spatial feature matrix It can be directly equivalent to the above reference sample space X p ;

[0105] In another optional embodiment, when each reference operation sample in the reference sample space is configured with a weight, the spatial feature matrix According to the above reference sample space X p Determined with a weight vector indicating a sample weight for each reference operation sample;

[0106] In a preferred embodiment, the obtaining of the spatial feature matrix of the reference sample space includes:

[0107] S1, determining a mean eigenvector matching the reference sample space based on a weighted summation result between multiple reference sample eigenvectors;

[0108] S2, obtaining a difference vector between a plurality of reference sample feature vectors and a mean feature vector in the reference sample space;

[0109] S3, determining a spatial feature matrix of the reference sample space according to the splicing results of the multiple difference vectors.

[0110] In this embodiment, when each reference operation sample in the reference sample space is configured with a different weight, the mean feature vector can be calculated based on the reference sample space X p Each reference eigenvector x in i is determined by a weight vector a indicating the sample weight of each reference operation sample, namely:

[0111]

[0112] In another optional embodiment, when each reference operation sample in the reference sample space is configured with the same weight, the mean feature vector can be calculated based on the reference sample space X p Each reference eigenvector x in i The average vector is determined by:

[0113]

[0114] Next, the spatial feature matrix of the reference sample space is It can be determined based on the concatenation results of multiple difference vectors between multiple reference sample feature vectors and the mean feature vector:

[0115]

[0116] It should be noted that, in this embodiment, the spatial feature matrix is determined by the splicing result of the difference vectors. This approach can significantly reduce the amount of data required for operations. For example, if you perform related operations directly on a vector [5, 7, 8, 5, 6, 10], you need to perform related calculations on each element separately during each calculation. However, if the mean vector is [5, 5, 5, 5, 5, 5], the difference vector between the two vectors is [0, 2, 3, 0, 1, 5], which significantly reduces the amount of data in the vector. Furthermore, when performing vector or matrix calculations based on the difference vector, at least the calculations for the first and third elements can be completed quickly, thereby improving the efficiency of matrix calculations.

[0117] In an optional implementation, determining the reference submatrix determined according to the first reference matrix as a principal component matrix matching the reference sample space includes:

[0118] S1, obtaining a singular value sequence consisting of singular values in a third reference matrix, and taking the sum of the squares of each singular value in the singular value sequence as a first reference value, wherein the singular value sequence includes a plurality of singular values arranged in descending order of numerical value;

[0119] S2, obtaining a principal component coefficient, and obtaining a product of the first reference value and the principal component coefficient as a second reference value, wherein the principal component coefficient is a preset constant greater than 0 and less than 1;

[0120] S3. When the sum of the squares of the first M singular values in the singular value sequence is greater than or equal to the second reference value, and the sum of the squares of the first M-1 singular values in the singular value sequence is less than the second reference value, a reference submatrix composed of the first M column vectors in the first reference matrix is determined as a principal component matrix, wherein the first reference matrix includes L column vectors, L is an integer greater than 1, and M is an integer greater than or equal to 1 and less than or equal to L.

[0121] Specifically, in this embodiment, the spatial feature matrix When performing singular value decomposition,

[0122] Assumptions:

[0123] ∑ P= diag(σ1, σ2, …, σ m )

[0124] where σ m are the singular values arranged in descending order on the diagonal of the diagonal matrix;

[0125] Next, obtain the principal component coefficient a, where 0 < a < 1, which is used to indicate the proportion of the feature of the space indicated by the principal component matrix in the original feature space. Usually, it can be selected as 0.8 or 0.9. Next, find a positive integer L p (i.e., M in the above method), such that the following two inequalities both hold:

[0126]

[0127]

[0128] That is, determine the above positive integer L p which can be used to determine the above principal component matrix It can be understood that in the above inequality, after arranging the singular values in Σ P in descending order, find the number of the smallest singular values such that the sum of the squares occupied by these singular values is a times the sum of the squares of all singular values. Since the value of a is usually close to 1, that is, this part of the singular values can approximately replace all the singular values, and then the matrix dimension of the principal component matrix that can represent the original sample space can be determined accordingly, that is:

[0129] the first L p columns

[0130] Through the above implementation manners of the present application, the sample features of operation samples with the same operation type can be indicated by the above principal component matrix, that is, the feature information in the original sample space is highly refined. While reducing the storage and calculation dimensions, as much feature information as possible is retained, thereby improving the feature extraction efficiency of the sample space. Further, based on the projection result of the above principal component matrix and the operation feature vector, a projection vector is determined, and then the feature correlation degree between the operation feature vector and the operation samples of each operation type can be accurately extracted.

[0131] In an optional implementation manner, the above determining the target operation type of the object operation according to the target classification model includes:

[0132] S1, obtain N type probabilities output by the target classification model based on the target feature vector, where the N type probabilities respectively correspond to N operation types; <​S2: Determine the operation type corresponding to the highest type probability among the N type probabilities as the target operation type.

[0134] It can be understood that in this embodiment, when the target classification model is a multi-classification model, the target classification model can output N type probabilities based on the target feature vector, and then determine the operation type of the object operation according to each probability;

[0135] In the case where the target classification model is a binary classification model, the target classification model may actually output a probability p corresponding to the first type. Since the target classification model is a binary classification model, it is equivalent to outputting a probability 1-p corresponding to the second type at the same time.

[0136] In a specific manner, when the target classification model is a logistic regression model, a method for determining the target operation type of the object operation according to the target classification model is as follows:

[0137] p=sigmoid(ω T x+b)

[0138] Among them, ω and b are learnable model parameters.

[0139] In this embodiment, the threshold value p set in advance can be used to pred , if the target classification model outputs p>p pred , it can be determined that the current object operation is an operation of the first operation type; otherwise, it can be determined that the current object operation is an operation of the second operation type.

[0140] In an optional embodiment, before obtaining the operation information of the object operation to be identified and determining the operation feature vector of the object operation according to the operation information, the method further includes:

[0141] S1, obtaining N sample sets corresponding to N operation types and a current classification model in training, wherein the sample sets include multiple object operation samples, and the object operation samples include sample information and label information;

[0142] S2, determining N spatial feature matrices according to the N sample sets, wherein the spatial feature matrices are used to represent sample features of the object operation samples in the corresponding sample sets;

[0143] S3, obtaining a current operation feature vector of a current object operation sample from the N sample sets, and determining N sample projection vectors according to the current operation feature vector and the N spatial feature matrices;

[0144] S4, concatenating the N sample projection vectors to obtain a current sample feature that matches the current object operation sample, and obtaining a classification result of the current classification model based on the current sample feature;

[0145] S5, training the current classification model according to the classification result and the comparison result of the label information of the current object operation sample until a target classification model that meets the material collection conditions is obtained.

[0146] It should be noted that, in this embodiment, the above training process can be specifically a gradient descent training method, and in the above training process, the loss function determined can be specifically:

[0147]

[0148] In other words, in this embodiment, the training process can be based on multiple rounds of training of the sample set. In each round of training, it is necessary to obtain the training results of each sample, and then obtain the training loss of the current round through the above loss function.

[0149] In an optional manner, when the training loss does not meet the preset loss, the model parameters ω and b of the current classification model may be adjusted;

[0150] In another optional embodiment, when the training loss does not meet the preset loss, the principal component coefficients used to determine the principal component matrix may be adjusted while adjusting the model parameters ω and b of the current classification model.

[0151] In another optional embodiment, the training process of the above classification model can be as follows Figure 3 As shown:

[0152] S302, obtaining the trained first classification recognition model and the current principal component coefficients;

[0153] S304, obtaining the principal component matrix of each sample space according to the principal component coefficients;

[0154] The specific method of obtaining each principal component matrix can be carried out in the manner described in the previous embodiment of the specification, and will not be repeated here;

[0155] S306, determining a target eigenvector for each training sample based on the principal component matrix of each sample space and each sample eigenvector;

[0156] S308, obtaining recognition evaluation parameters of the first classification recognition model according to the target feature vector and sample label of each training sample;

[0157] It should be noted that the evaluation parameter may be the classification accuracy of the first classification recognition model, or may be the improvement rate of the classification accuracy of the first classification recognition model;

[0158] S310: Is the recognition evaluation parameter less than or equal to the target parameter? If the recognition evaluation parameter is less than or equal to the target parameter, execute S312 to adjust the principal component parameter; and repeat steps S304 to S310.

[0159] When the recognition parameter is greater than the target parameter, execute S314 to obtain the second classification recognition model to be trained and the target principal component parameters after training; then execute S316 to use the target principal component parameters to determine the principal component matrix of each sample space and the target feature vector of each training sample; finally, execute S318 to train the second classification model according to the target feature vector and sample label of each training sample.

[0160] In a preferred embodiment, the model scale of the first classification recognition model completed by the above training can be larger than the model scale of the second classification recognition model, and then the appropriate principal component coefficients are determined through the large model that already has a good classification effect, and then the small-scale second classification recognition model to be trained is trained according to the appropriate principal component coefficients, thereby reducing the model scale of the classification model and improving the classification efficiency.

[0161] Furthermore, in this embodiment, the above-mentioned principal component matrix can be used to characterize the above-mentioned sample space. In addition, the above-mentioned second classification recognition model is a smaller-scale recognition model with high recognition efficiency. The above-mentioned principal component matrix, target principal component parameters and second classification recognition model can be deployed on the client, thereby realizing real-time classification and risk identification of account operations of the object account on the client.

[0162] In an optional implementation, the obtaining of N sample sets corresponding to the N operation types includes:

[0163] S1, obtaining an account operation set in a target social platform within a target period, wherein the account operation set includes multiple account interaction operations between multiple target accounts;

[0164] S2, determining an object operation sample based on operation information of the account interaction operation and the operation type of the interaction operation, wherein the operation information includes account information of the object account associated with the interaction operation and operation behavior information of the interaction operation;

[0165] S3: Determine N sample sets corresponding to the N operation types according to the classification results of the object operation samples.

[0166] In this embodiment, the above-mentioned multiple sample sets can be determined by obtaining a set of account operations within a target period. Specifically, when the number of types N is 2, the above-mentioned operation types can be specifically normal operation types and abnormal operation types. The above-mentioned abnormal operation type can be specifically account operations suspected of resource fraud on social platforms.

[0167] In this embodiment, since the operation methods of the above-mentioned fraudulent operations and the operation details of normal account operations will also change over time, in the process of determining the above-mentioned sample sets, the most recent account operation sets can be preferentially selected according to a certain period to determine the normal sample set and the abnormal sample set. By updating the sample sets according to a certain period, the accuracy of operation classification is improved.

[0168] In an optional embodiment, the step of obtaining the operation information of the object operation to be identified and determining the operation feature vector of the object operation according to the operation information includes:

[0169] S1, when the object operation is an account interaction operation between a first object account and a second object account, obtaining account information of the first object account and the second object account, as well as operation behavior information of the account interaction operation;

[0170] S2, mapping each operation information in the account information and the operation behavior information into an operation feature value, and determining an operation feature vector of the object operation based on the multiple operation feature values.

[0171] Optionally, after determining the target operation type of the object operation according to the target classification model, the method further includes at least one of the following:

[0172] Method 1: When the target operation type of the object operation is a risk type, intercept the account interaction operation between the first object account and the second object account;

[0173] Method 2: When the target operation type of the object operation is a risk type, a risk behavior prompt message is sent to the first object account;

[0174] Method three: when the target operation type of the object operation is a risk type, sending risk behavior prompt information to the second object account.

[0175] The above implementation is explained below with reference to payment scenarios in instant messaging applications. It is understandable that in instant messaging applications, payment or transfer operations can be performed between target accounts, and some non-compliant accounts may obtain virtual resources through payment or transfer operations of ordinary accounts, thereby completing resource fraud.

[0176] In the above-mentioned instant messaging platform, when the above-mentioned payment or transfer operation occurs, relevant operation information of the above-mentioned payment or transfer operation can be obtained, and identification can be performed through the method described in the above-mentioned embodiment to determine whether the above-mentioned payment or transfer operation is an abnormal (risky) operation. Then, if the above-mentioned payment or transfer operation is an abnormal (risky) operation, the above-mentioned operation can be intercepted or risk warning information can be sent to the corresponding object account.

[0177] Correspondingly, when the object operation is a payment or transfer operation, the operation information may specifically include three categories: account information of the first object account, account information of the second object account, and operation information of the payment operation or transfer operation. The account information of the first object account may include account identification, historical payment / transfer records, account risk level, etc.; the account information of the second object account may include account identification, historical payment / transfer records, account risk level, etc.; the operation information of the payment operation or transfer operation may include operation time, operation frequency, and the account IP address corresponding to the timestamp of the operation.

[0178] It should be noted that in this embodiment, the relevant object operations or operation information involved are all pre-acquired through methods specified in relevant normative documents, and before obtaining the above-mentioned object operations or operation information, it is necessary to obtain authorization permission from the corresponding account subject.

[0179] The following combination Figure 4 An application scenario of this application is described.

[0180] In this implementation, for each transaction generated in the target application, the fraud risk of the transaction is detected by online deployment based on this model and supporting strategies. When a transaction is identified as having a high fraud risk, the system will remind the user or block it to prevent the loss of resources in the target account. The specific process is as follows: Figure 4 As shown:

[0181] First, the model is trained using a large amount of data and case studies. The model's effectiveness is then evaluated and validated over time. Finally, the model scores are deployed to the server for policy implementation. At the policy level, corresponding supporting policies are deployed simultaneously. When a user conducts a transaction, the model and policy monitor the transaction for fraud risk. If a transaction is identified as fraudulent, an alert or blocking action is taken to the affected account, thereby reducing the likelihood of risky transactions.

[0182] In another optional implementation, two principal component matrices can be used to represent the positive sample space and the negative sample space respectively. When the trained target classification recognition model is a small-scale recognition model with high recognition efficiency, the principal component matrices, principal component parameters and target classification recognition models corresponding to the above positive and negative spaces can be deployed on the client, thereby realizing real-time risk identification of payment operations of the object account on the client.

[0183] The following combination Figure 5 A complete implementation of the present application is described.

[0184] S502, principal component learning of positive and negative samples;

[0185] Specifically, assume that there are n1 positive samples (fraudulent transaction samples) and n0 negative samples (normal transaction samples), and the feature dimension corresponding to each sample is m (that is, an m-dimensional vertical feature vector).

[0186] Corresponding to the positive sample set (positive), each sample is represented by x i Represented. Then the entire positive sample set can be expressed as:

[0187]

[0188] First, remove the mean of the entire positive sample set. The mean of the positive sample set is defined as follows:

[0189]

[0190] The result after removing the mean of the entire positive sample set is as follows:

[0191]

[0192] right Doing mathematical SVD decomposition, we can get:

[0193]

[0194] Among them, U P and V P are two orthogonal matrices, Σ P It is a diagonal matrix, and the singular values on the diagonal are arranged from large to small;

[0195] Assumptions:

[0196] Σ P =diag(σ1,σ2,…,σ m )

[0197] Obtain the principal component coefficient \(a\), where \(0 < a < 1\), which is used to indicate the proportion of the feature in the original feature space represented by the principal component matrix. Usually, it can be selected as \(0.8\) or \(0.9\). Then, find a positive integer \(L\) p (i.e., \(M\) in the above method) such that the following two inequalities both hold:

[0198]

[0199]

[0200] It can be understood that in the above inequalities, after arranging the singular values in \(\sum\) P in descending order, find the number of the smallest singular values such that the sum of the squares of these singular values is \(a\) times the sum of the squares of all singular values. Since the value of \(a\) is usually close to \(1\), that is, this part of the singular values can approximately replace all the singular values.

[0201] Furthermore, the matrix dimension of the principal component matrix that can represent the original sample space can be determined by this value, that is:

[0202] The first \(L\) p columns

[0203] The principal component matrix of the negative sample set can be obtained in the same way

[0204] S504, sample projection calculation;

[0205] For all samples \(x\) in the training set i , including positive and negative samples, calculate their projections in the positive sample principal component space and in the negative sample principal component space respectively.

[0206] The calculation method of the projection of the sample in the positive sample principal component space is as follows: First subtract the mean of the positive sample space, and then multiply by the transpose of the positive sample principal component space matrix, that is:

[0207]

[0208] The calculation method of the projection of the sample in the negative sample principal component space is similar, that is:

[0209]

[0210] S506, merging projections of positive and negative sample spaces;

[0211] Specifically, vertically merge the projection vectors of the sample in the positive sample principal component space and the projection vector of the sample in the negative sample principal component space into a feature vector, which completely contains all the features of the sample in the positive and negative spaces, that is:

[0212]

[0213] S508, logistic regression classification model training;

[0214] The feature vector obtained in step S506 above is Input into the logistic regression model for classification training. The probability calculation formula of the classification model output is as follows:

[0215] p=sigmoid(ω T x+b)

[0216] Where ω and b are learnable model parameters.

[0217] The corresponding loss function is:

[0218]

[0219] Wherein, y is the label corresponding to the sample. It should be noted that, in the training process of this embodiment, the training and learning process of the parameters adopts a gradient descent method.

[0220] S510, performing risk detection based on the model results;

[0221] Specifically, when a new payment operation is detected, the projection vector is calculated for the corresponding original feature vector using the method described in step S504 above, and the projection vectors are merged using the method of S506 and then input into the target classification model trained in S506.

[0222] The target classification model can output an expected score. pred , if p>p pred , the model predicts that the current payment operation has high risk, and can then output corresponding abnormal risk prompts, such as reminding the user to obtain interception operations, and then avoid resource loss caused by risky operations.

[0223] Through the above-mentioned implementation of the present application, principal component learning is performed on normal and abnormal transaction operations through principal component analysis. The principal component matrix obtained from the learning is used to identify the corresponding principal component space (including the principal component space of normal transactions and the principal component space of abnormal transactions); then, the samples are projected into the principal component space of normal transactions and the principal component space of fraudulent transactions respectively to obtain corresponding projection vectors, and the projection vectors are merged and input into the logistic regression model for binary classification training. During the training process, a score threshold is set, and the transaction samples to be predicted are predicted using the trained model. Transactions with scores above the threshold are considered abnormal transactions.

[0224] During the recognition process, principal component analysis (PCA) is used to project the original feature vectors of the features. This not only removes invalid features and possible noise from the data, but also reduces the dimensionality of the feature vectors, improving the computational efficiency of training and prediction. Furthermore, the projection method and model structure are simple, eliminating the need for cumbersome computational burdens in the training and prediction processes and not relying on complex model structures and operations, facilitating real-time deployment.

[0225] In the above implementation, the principal component analysis method is used to learn the principal component spaces of normal transactions and abnormal transactions respectively, and the transaction samples to be predicted are projected into the two spaces, removing the interference of invalid features and noise, and performing fraud model prediction based on the projection. The fraud risk of the transaction is evaluated and intervened based on the classification results of the model to improve the accuracy and stability of recognition and reduce the risk of fraud and loss of the platform and users.

[0226] In the above-mentioned implementation of this application, in order to fulfill legal obligations such as financial risk control and financial fraud detection, the platform involves the acquisition and processing of relevant information and data of the target account. The above-mentioned relevant information and data are all pre-acquired through methods specified in relevant normative legal documents, and before obtaining the above-mentioned relevant information and data, authorization permission from the corresponding account subject is required.

[0227] It should be noted that for the aforementioned method embodiments, for simplicity of description, they are all expressed as a series of action combinations. However, those skilled in the art should be aware that the present invention is not limited by the order of the actions described, because according to the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0228] According to another aspect of the present invention, there is also provided an object operation recognition device for implementing the above object operation recognition method. Figure 6 As shown, the device includes:

[0229] A first acquiring unit 602 is configured to acquire operation information of an object operation to be identified, and determine an operation feature vector of the object operation based on the operation information, wherein the operation information includes account information and operation behavior information associated with the object operation;

[0230] A second obtaining unit 604 is configured to obtain N projection vectors obtained by projecting the operation feature vector onto N sample spaces, where the N sample spaces correspond to N operation types, respectively. The sample spaces are vector spaces determined based on multiple sample feature vectors corresponding to the operation types, and the sample feature vectors are feature vectors determined based on sample information of object operation samples corresponding to the operation types. N is an integer greater than or equal to 1.

[0231] A classification unit 606 is configured to determine a target feature vector that matches the object operation based on the N projection vectors, and input the target feature vector into a target classification model;

[0232] The determining unit 608 is configured to determine the target operation type of the object operation according to the target classification model.

[0233] Optionally, the second acquisition unit 604 is used to repeat the following operations until N of the sample spaces are traversed: obtaining a principal component matrix that matches the current sample space, wherein the principal component matrix is used to indicate the main feature distribution of the multiple sample feature vectors included in the current sample space; determining the current projection vector corresponding to the current sample space based on the operation feature vector and the principal component matrix.

[0234] Optionally, the second acquisition unit 604 is used to: obtain a mean eigenvector matching the current sample space, wherein the mean eigenvector is a eigenvector determined based on the weighted summation result of multiple sample eigenvectors; determine the difference vector between the operation eigenvector and the mean eigenvector as a reference eigenvector; and determine the projection vector based on the transposed matrix of the principal component matrix and the reference eigenvector.

[0235] Optionally, the above-mentioned object operation recognition device also includes: a third acquisition unit, used to obtain a reference sample set consisting of multiple reference operation samples of a reference operation type, wherein the N above-mentioned operation types include the above-mentioned reference operation type; according to the above-mentioned sample information corresponding to each of the multiple reference operation samples, respectively determine the multiple reference sample feature vectors, and construct a reference sample space corresponding to the reference operation type based on the multiple reference sample feature vectors; obtain the spatial feature matrix of the above-mentioned reference sample space, and perform singular value decomposition on the above-mentioned spatial feature matrix to obtain a first reference matrix, a second reference matrix and a third reference matrix, wherein the above-mentioned first reference matrix includes left singular vectors corresponding to the above-mentioned spatial feature matrix, the above-mentioned second reference matrix includes right singular vectors corresponding to the above-mentioned spatial feature matrix, the above-mentioned first reference matrix and the above-mentioned second reference matrix are orthogonal matrices, and the above-mentioned third reference matrix is a diagonal matrix; determine the reference submatrix determined according to the above-mentioned first reference matrix as the above-mentioned principal component matrix matching the above-mentioned reference sample space, wherein the matrix dimension of the above-mentioned reference submatrix is smaller than the matrix dimension of the above-mentioned first reference matrix.

[0236] Optionally, the above-mentioned third acquisition unit is used to: determine the mean feature vector matching the above-mentioned reference sample space based on the weighted summation result between the multiple reference sample feature vectors; obtain the difference vector between the multiple reference sample feature vectors and the above-mentioned mean feature vector in the above-mentioned reference sample space; and determine the above-mentioned spatial feature matrix of the above-mentioned reference sample space based on the splicing result of the multiple difference vectors.

[0237] Optionally, the third acquisition unit is configured to: acquire a singular value sequence composed of singular values in the third reference matrix, and use the sum of squares of each singular value in the singular value sequence as a first reference value, wherein the singular value sequence includes a plurality of singular values arranged in descending order according to numerical values; acquire principal component coefficients, and acquire the product of the first reference value and the principal component coefficients as a second reference value, wherein the principal component coefficients are preset constants greater than 0 and less than 1; when the sum of squares of the first M singular values in the singular value sequence is greater than or equal to the second reference value, and the sum of squares of the first M-1 singular values in the singular value sequence is less than the second reference value, determine the reference submatrix composed of the first M column vectors in the first reference matrix as the principal component matrix, wherein the first reference matrix includes L column vectors, L is an integer greater than 1, and M is an integer greater than or equal to 1 and less than or equal to L.

[0238] Optionally, the above-mentioned classification unit 606 includes one of the following: a first classification module, used to determine the N above-mentioned projection vectors as the above-mentioned target feature vectors that match the above-mentioned object operation; a second classification module, used to determine the above-mentioned reference projection vector as the above-mentioned target feature vector that matches the above-mentioned object operation when the vector modulus of the reference projection vector in the N above-mentioned projection vectors is greater than or equal to the target threshold; a third classification module, used to determine the weighted sum result of the above-mentioned N above-mentioned projection vectors as the above-mentioned target feature vector.

[0239] Optionally, the above-mentioned determination unit 608 includes: an acquisition module for obtaining N type probabilities output by the above-mentioned target classification model based on the above-mentioned target feature vector, wherein the N above-mentioned type probabilities respectively correspond to the N above-mentioned operation types; a determination module for determining the above-mentioned operation type corresponding to the highest above-mentioned type probability among the N above-mentioned type probabilities as the above-mentioned target operation type.

[0240] Optionally, the above-mentioned determination unit 608 is also used to: obtain N sample sets corresponding to the N above-mentioned operation types and a current classification model in a training state, wherein the above-mentioned sample sets include multiple object operation samples, and the above-mentioned object operation samples include sample information and label information; determine N spatial feature matrices according to the N above-mentioned sample sets, wherein the above-mentioned spatial feature matrices are used to characterize the sample features of the above-mentioned object operation samples in the corresponding above-mentioned sample sets; obtain a current operation feature vector of a current object operation sample from the N above-mentioned sample sets, and determine N sample projection vectors according to the above-mentioned current operation feature vector and the N above-mentioned spatial feature matrices; splice the N above-mentioned sample projection vectors to obtain a current sample feature that matches the above-mentioned current object operation sample, and obtain the classification result of the above-mentioned current classification model based on the above-mentioned current sample feature; train the above-mentioned current classification model according to the above-mentioned classification result and the comparison result of the label information of the above-mentioned current object operation sample until the above-mentioned target classification model that meets the material collection conditions is obtained.

[0241] Optionally, the above-mentioned determination unit 608 is used to: obtain an account operation set in the target social platform within the target period, wherein the above-mentioned account operation set includes multiple account interaction operations between multiple object accounts; determine the above-mentioned object operation samples based on the operation information of the above-mentioned account interaction operations and the operation type of the above-mentioned interaction operations, wherein the above-mentioned operation information includes the account information of the above-mentioned object account associated with the above-mentioned interaction operation, and the operation behavior information of the above-mentioned interaction operation; determine N sample sets corresponding to the N types of the above-mentioned operation types according to the classification results of the above-mentioned object operation samples.

[0242] Optionally, the first acquisition unit 602 is used to: when the object operation is an account interaction operation between a first object account and a second object account, obtain the account information of the first object account and the second object account, as well as the operation behavior information of the account interaction operation; map each item of operation information in the account information and the operation behavior information into an operation feature value, and determine the operation feature vector of the object operation based on multiple operation feature values.

[0243] Optionally, the above-mentioned object operation identification device also includes at least one of the following: an interception unit, used to intercept the above-mentioned account interaction operation between the above-mentioned first object account and the second object account when the above-mentioned target operation type of the above-mentioned object operation is a risk type; a first alarm unit, used to send risk behavior prompt information to the above-mentioned first object account when the above-mentioned target operation type of the above-mentioned object operation is a risk type; a second alarm unit, used to send risk behavior prompt information to the above-mentioned second object account when the above-mentioned target operation type of the above-mentioned object operation is a risk type.

[0244] Optionally, in this embodiment, the embodiments to be implemented by the above-mentioned various unit modules can refer to the above-mentioned various method embodiments, which will not be repeated here.

[0245] According to another aspect of the embodiment of the present invention, an electronic device for implementing the above object operation recognition method is also provided. The electronic device may be Figure 7 The terminal device or server shown in FIG. This embodiment is described by taking the electronic device as a terminal device as an example. Figure 7 As shown, the electronic device includes a memory 702 and a processor 704. The memory 702 stores a computer program, and the processor 704 is configured to execute the steps in any of the above method embodiments through the computer program.

[0246] Optionally, in this embodiment, the electronic device may be located in at least one network device among a plurality of network devices of a computer network.

[0247] Optionally, in this embodiment, the processor may be configured to execute the following steps through a computer program:

[0248] S1, obtaining operation information of an object operation to be identified, and determining an operation feature vector of the object operation based on the operation information, wherein the operation information includes account information and operation behavior information associated with the object operation;

[0249] S2, obtaining N projection vectors obtained by projecting the operation feature vector onto N sample spaces, respectively, where the N sample spaces correspond to N operation types, the sample spaces are vector spaces determined based on multiple sample feature vectors of the corresponding operation types, the sample feature vectors are feature vectors determined based on sample information of object operation samples of the corresponding operation types, and N is an integer greater than or equal to 1;

[0250] S3, determining a target feature vector that matches the object operation based on the N projection vectors, and inputting the target feature vector into a target classification model;

[0251] S4, determining the target operation type of the object operation according to the target classification model.

[0252] Alternatively, those skilled in the art will appreciate that Figure 7 The structure shown is for illustration only, and the electronic device may also be a vehicle-mounted terminal, a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a PDA, a mobile Internet device (MID), a PAD, or other terminal devices. Figure 7 It does not limit the structure of the above electronic device. For example, the electronic device may also include Figure 7 More or fewer components (such as network interfaces, etc.) as shown in, or with Figure 7 Different configurations shown.

[0253] Among them, the memory 702 can be used to store software programs and modules, such as the program instructions / modules corresponding to the object operation recognition method and device in the embodiment of the present invention. The processor 704 executes various functional applications and data processing by running the software programs and modules stored in the memory 702, that is, realizing the above-mentioned object operation recognition method. The memory 702 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 702 may further include a memory remotely located relative to the processor 704, and these remote memories may be connected to the terminal via a network. Examples of the above-mentioned networks include but are not limited to the Internet, corporate intranets, local area networks, mobile communication networks and combinations thereof. Among them, the memory 702 can be used specifically, but not limited to, to store file information such as target logical files. As an example, such as Figure 7As shown, the memory 702 may include, but is not limited to, the first acquisition unit 602, the second acquisition unit 604, the classification unit 606, and the determination unit 608 in the object operation recognition device. In addition, it may also include, but is not limited to, other module units in the object operation recognition device, which will not be repeated in this example.

[0254] Optionally, the transmission device 706 is configured to receive or send data via a network. Specific examples of the network may include a wired network and a wireless network. In one embodiment, the transmission device 706 includes a network interface controller (NIC), which can be connected to other network devices and a router via a network cable to communicate with the Internet or a local area network. In one embodiment, the transmission device 706 is a radio frequency (RF) module, which is configured to communicate with the Internet wirelessly.

[0255] In addition, the electronic device further includes: a display 708 and a connection bus 710 for connecting various module components in the electronic device.

[0256] In other embodiments, the terminal device or server may be a node in a distributed system, wherein the distributed system may be a blockchain system, and the blockchain system may be a distributed system formed by connecting multiple nodes through network communication. The nodes may form a peer-to-peer network, and any computing device, such as a server, terminal, or other electronic device, may become a node in the blockchain system by joining the peer-to-peer network.

[0257] According to one aspect of the present application, a computer program product is provided, comprising a computer program / instructions containing program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via a communication component and / or installed from a removable medium. When the computer program is executed by a central processing unit, the various functions provided in the embodiments of the present application are performed.

[0258] The serial numbers of the above embodiments of the present invention are for description only and do not represent the advantages or disadvantages of the embodiments.

[0259] According to one aspect of the present application, a computer-readable storage medium is provided, and a processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the above-mentioned object operation recognition method.

[0260] Optionally, in this embodiment, the computer-readable storage medium may be configured to store a computer program for performing the following steps:

[0261] S1, obtaining operation information of an object operation to be identified, and determining an operation feature vector of the object operation based on the operation information, wherein the operation information includes account information and operation behavior information associated with the object operation;

[0262] S2, obtaining N projection vectors obtained by projecting the operation feature vector onto N sample spaces, respectively, where the N sample spaces correspond to N operation types, the sample spaces are vector spaces determined based on multiple sample feature vectors of the corresponding operation types, the sample feature vectors are feature vectors determined based on sample information of object operation samples of the corresponding operation types, and N is an integer greater than or equal to 1;

[0263] S3, determining a target feature vector that matches the object operation based on the N projection vectors, and inputting the target feature vector into a target classification model;

[0264] S4, determining the target operation type of the object operation according to the target classification model.

[0265] Optionally, in this embodiment, a person of ordinary skill in the art may understand that all or part of the steps in the various methods of the above embodiments may be completed by instructing the hardware related to the terminal device through a program, and the program may be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0266] If the integrated units in the above embodiments are implemented in the form of software functional units and sold or used as independent products, they can be stored in the above-mentioned computer-readable storage medium. Based on this understanding, the technical solution of the present invention, 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. This computer software product is stored in a storage medium and includes a number of instructions for causing one or more computer devices (such as personal computers, servers, or network devices) to execute all or part of the steps of the above-mentioned methods in various embodiments of the present invention.

[0267] In the above embodiments of the present invention, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0268] In the several embodiments provided in this application, it should be understood that the disclosed client can be implemented in other ways. Among them, the device embodiments described above are merely illustrative. For example, the division of the above-mentioned units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0269] The units described above as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0270] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0271] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for identifying object operations, characterized in that: include: Acquiring operation information of the object operation to be identified, and determining an operation feature vector of the object operation based on the operation information, wherein the operation information includes account information and operation behavior information associated with the object operation; Obtaining N projection vectors obtained by projecting the operation feature vector onto N sample spaces, respectively, where the N sample spaces correspond to N operation types, respectively, the sample spaces are vector spaces determined based on multiple sample feature vectors corresponding to the operation types, the sample feature vectors are feature vectors determined based on sample information of object operation samples corresponding to the operation types, and N is an integer greater than or equal to 1; Determine a target feature vector that matches the object operation based on the N projection vectors, and input the target feature vector into a target classification model; A target operation type of the object operation is determined according to the target classification model.

2. The method according to claim 1, characterized in that Obtaining N projection vectors obtained by projecting the operational feature vector onto N sample spaces includes: Repeat the following steps until N sample spaces are traversed: Obtaining a principal component matrix matching the current sample space, wherein the principal component matrix is used to indicate the main feature distribution of the plurality of sample feature vectors included in the current sample space; A current projection vector corresponding to the current sample space is determined according to the operational eigenvector and the principal component matrix.

3. The method according to claim 2, characterized in that Determining the projection vector corresponding to the current sample space according to the operational eigenvector and the principal component matrix includes: Obtaining a mean feature vector that matches the current sample space, wherein the mean feature vector is a feature vector determined based on a weighted summation result of a plurality of the sample feature vectors; Determine a difference vector between the operational eigenvector and the mean eigenvector as a reference eigenvector; The projection vector is determined according to a transposed matrix of the principal component matrix and the reference eigenvector.

4. The method according to claim 2, characterized in that Before obtaining the principal component matrix matching the current sample space, the method further includes: Acquire a reference sample set consisting of a plurality of reference operation samples of a reference operation type, wherein the N operation types include the reference operation type; Determining a plurality of reference sample feature vectors respectively according to the sample information corresponding to each of the plurality of reference operation samples, and constructing a reference sample space corresponding to the reference operation type according to the plurality of reference sample feature vectors; Obtaining a spatial characteristic matrix of the reference sample space, and performing singular value decomposition on the spatial characteristic matrix to obtain a first reference matrix, a second reference matrix, and a third reference matrix, wherein the first reference matrix includes left singular vectors corresponding to the spatial characteristic matrix, the second reference matrix includes right singular vectors corresponding to the spatial characteristic matrix, the first reference matrix and the second reference matrix are orthogonal matrices, and the third reference matrix is a diagonal matrix; A reference submatrix determined according to the first reference matrix is determined as the principal component matrix matching the reference sample space, wherein a matrix dimension of the reference submatrix is smaller than a matrix dimension of the first reference matrix.

5. The method according to claim 4, characterized in that The acquiring of the spatial feature matrix of the reference sample space comprises: Determining a mean feature vector matching the reference sample space according to a weighted summation result between a plurality of reference sample feature vectors; Obtaining difference vectors between a plurality of reference sample feature vectors and the mean feature vector in the reference sample space; The spatial feature matrix of the reference sample space is determined according to a splicing result of a plurality of the difference vectors.

6. The method according to claim 4, characterized in that The step of determining the reference submatrix determined according to the first reference matrix as the principal component matrix matching the reference sample space comprises: Obtaining a singular value sequence composed of singular values in the third reference matrix, and using the sum of squares of each singular value in the singular value sequence as a first reference value, wherein the singular value sequence includes a plurality of singular values arranged in descending order of numerical value; Obtaining a principal component coefficient, and obtaining a product of the first reference value and the principal component coefficient as a second reference value, wherein the principal component coefficient is a preset constant greater than 0 and less than 1; When the sum of the squares of the first M singular values in the singular value sequence is greater than or equal to the second reference value, and the sum of the squares of the first M-1 singular values in the singular value sequence is less than the second reference value, the reference submatrix composed of the first M column vectors in the first reference matrix is determined as the principal component matrix, wherein the first reference matrix includes L column vectors, L is an integer greater than 1, and M is an integer greater than or equal to 1 and less than or equal to L.

7. The method according to claim 1, characterized in that Determining a target feature vector matching the object operation according to the N projection vectors includes one of the following: Determining N of the projection vectors as the target feature vectors that match the object operation; In a case where a vector modulus of a reference projection vector among the N projection vectors is greater than or equal to a target threshold, determining the reference projection vector as the target feature vector that matches the object operation; A weighted sum of the N projection vectors is determined as the target feature vector.

8. The method according to claim 1, characterized in that The determining the target operation type of the object operation according to the target classification model includes: Obtaining N type probabilities output by the target classification model based on the target feature vector, wherein the N type probabilities correspond to the N operation types respectively; The operation type corresponding to the highest type probability among the N type probabilities is determined as the target operation type.

9. The method according to claim 8, characterized in that Before acquiring the operation information of the object operation to be identified and determining the operation feature vector of the object operation according to the operation information, the method further includes: Obtaining N sample sets corresponding to the N operation types and a current classification model in training, wherein the sample sets include a plurality of object operation samples, and the object operation samples include sample information and label information; Determining N spatial feature matrices respectively according to the N sample sets, wherein the spatial feature matrices are used to characterize sample features of the object operation samples in the corresponding sample sets; Obtaining a current operation feature vector of a current object operation sample from the N sample sets, and determining N sample projection vectors respectively according to the current operation feature vector and the N spatial feature matrices; splicing the N sample projection vectors to obtain a current sample feature that matches the current object operation sample, and obtaining a classification result of the current classification model based on the current sample feature; The current classification model is trained according to the classification result and the comparison result of the label information of the current object operation sample until the target classification model that meets the material collection conditions is obtained.

10. The method according to claim 9, characterized in that Obtaining N sample sets corresponding to the N types of operations respectively includes: Obtaining an account operation set on a target social platform within a target period, wherein the account operation set includes multiple account interaction operations between multiple target accounts; Determining the object operation sample according to operation information of the account interaction operation and the operation type of the interaction operation, wherein the operation information includes account information of the object account associated with the interaction operation and operation behavior information of the interaction operation; N sample sets corresponding to the N operation types are determined according to the classification results of the object operation samples.

11. The method according to claim 1, wherein The acquiring operation information of the object operation to be identified and determining the operation feature vector of the object operation according to the operation information includes: In a case where the object operation is an account interaction operation between a first object account and a second object account, obtaining the account information of the first object account and the second object account, and the operation behavior information of the account interaction operation; Each item of operation information in the account information and the operation behavior information is mapped into an operation feature value, and the operation feature vector of the object operation is determined according to a plurality of the operation feature values.

12. The method according to claim 11, characterized in that After determining the target operation type of the object operation according to the target classification model, the method further includes at least one of the following: When the target operation type of the object operation is a risk type, intercepting the account interaction operation between the first object account and the second object account; When the target operation type of the object operation is a risk type, sending risk behavior prompt information to the first object account; When the target operation type of the object operation is a risk type, risk behavior prompt information is sent to the second object account.

13. An object operation recognition device, characterized in that: include: a first acquiring unit, configured to acquire operation information of an object operation to be identified, and determine an operation feature vector of the object operation based on the operation information, wherein the operation information includes account information and operation behavior information associated with the object operation; a second acquiring unit, configured to acquire N projection vectors obtained by projecting the operation feature vector onto N sample spaces, respectively, wherein the N sample spaces correspond to N operation types, respectively, the sample spaces are vector spaces determined based on a plurality of sample feature vectors corresponding to the operation types, the sample feature vectors are feature vectors determined based on sample information of object operation samples of the corresponding operation types, and N is an integer greater than or equal to 1; a classification unit, configured to determine a target feature vector matching the object operation based on the N projection vectors, and input the target feature vector into a target classification model; A determining unit is configured to determine a target operation type of the object operation according to the target classification model.

14. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored program, wherein the method according to any one of claims 1 to 12 is executed when the program is executed.

15. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 12 are implemented.

16. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to execute the method according to any one of claims 1 to 12 through the computer program.

Images