Method and system for identifying merchant main body and storage medium

Abstract

The invention relates to a method and a system for identifying a merchant body, a computer readable storage medium for implementing the method and a computer program product. According to one aspect of the invention, the method comprises the steps: obtaining the feature data of a plurality of modes of a target merchant, and determining a classification result of the target merchant at least based on the feature data of the plurality of modes of the target merchant through a first neural network model; selectively utilizing a second neural network model to detect an operation behavior of the target merchant based on the classification result of the target merchant so as to determine an anomaly detection result of the target merchant; wherein the training data of the first neural network model comprises feature data of a plurality of modals of historical merchants and a plurality of classification tags corresponding to the historical merchants; a weight is dynamically assigned to feature data for each of a plurality of modalities of the historical merchant during training of a first neural network model based on a loss function of the first neural network model.

Description

Technical Field

[0001] The present application relates to a method and system for identifying a merchant entity, a computer-readable storage medium for implementing the method, and a computer program product. Background Art

[0002] With the rapid development of the payment and clearing industry, transactions through electronic payment have become the norm. In electronic payment, merchants refer to commercial entities that provide goods or services and receive payments through electronic payment platforms. Merchants can be of various types, such as individuals, mobile vendors, merchants with fixed business locations, etc.

[0003] Currently, it is usually necessary to formulate different risk monitoring strategies, compliance requirements, fund settlement methods, etc. for different types of merchants. In addition, it is expected to monitor the transaction activities of merchants to identify bad merchants.

[0004] Generally, merchant classification and identification mainly rely on merchant text data, that is, by collecting merchant text data and then using simple rules to classify and identify merchants, it is difficult to achieve accurate merchant classification and identification. In addition, merchant identification is currently usually only performed when the merchant joins the network, lacking effective merchant identification in subsequent transactions. Summary of the invention

[0005] In order to solve or at least alleviate one or more of the above problems, the following technical solutions are provided.

[0006] According to a first aspect of the present application, a method for identifying a merchant entity is provided, the method comprising the following steps: acquiring feature data of multiple modalities of a target merchant and using a first neural network model to determine a classification result of the target merchant at least based on the feature data of multiple modalities of the target merchant; selectively using a second neural network model to detect the operating behavior of the target merchant based on the classification result of the target merchant to determine an abnormal detection result of the target merchant; wherein the training data of the first neural network model includes feature data of multiple modalities of historical merchants and multiple classification labels corresponding to the historical merchants, and during the training of the first neural network model, weights are dynamically assigned to the feature data of each modality of the feature data of the multiple modalities of the historical merchant based on the loss function of the first neural network model.

[0007] According to the method for identifying a merchant entity described in one embodiment of the present application, the feature data of multiple modalities of the target merchant include a combination of one or more of the following: feature data of text modality obtained from text information of the target merchant; feature data of image modality obtained from image information of the target merchant; feature data of location modality obtained from location information of the target merchant.

[0008] According to the method for identifying a merchant entity described in one embodiment or any one of the above embodiments of the present application, the feature data of multiple modalities of the historical merchant include a combination of one or more of the following: feature data of the text modality obtained from the text information of the historical merchant; feature data of the image modality obtained from the image information of the historical merchant; feature data of the location modality obtained from the location information of the historical merchant.

[0009] According to the method for identifying a merchant entity according to one embodiment of the present application or any of the above embodiments, the first neural network model is a graph convolutional neural network model.

[0010] According to the method for identifying merchant entities described in one embodiment or any one of the above embodiments of the present application, during the training of the first neural network model: the target merchants and the historical merchants are used as graph nodes to construct a merchant node graph; the feature vectors of the graph nodes are determined based on the feature data of multiple modalities of the target merchants and the feature data of multiple modalities of the historical merchants; the adjacency matrix of the graph nodes is determined based on the similarity between the feature vectors of the graph nodes; and the feature vectors of the graph nodes and the adjacency matrix of the graph nodes are input into the first neural network model to update the layer node features of the first neural network model.

[0011] According to the method for identifying a merchant entity in one embodiment or any one of the above embodiments of the present application, the classification result of the target merchant is used to indicate that the category of the target merchant belongs to one of the multiple classification labels.

[0012] According to the method for identifying a merchant entity according to one embodiment of the present application or any of the above embodiments, selectively utilizing a second neural network model to detect the operating behavior of the target merchant based on the classification result of the target merchant includes: detecting the operating behavior of the target merchant in response to the classification result of the target merchant indicating that the category of the target merchant belongs to the first classification label among the multiple classification labels.

[0013] According to the method for identifying a merchant entity according to one embodiment of the present application or any of the above embodiments, the operating behavior of the target merchant includes: the modification behavior of the target merchant on one or more of text information, image information and location information, and the transaction behavior of the target merchant.

[0014] According to the method for identifying a merchant entity according to one embodiment of the present application or any of the above embodiments, the second neural network model is a recurrent neural network model.

[0015] According to the method for identifying a merchant entity according to one embodiment of the present application or any of the above embodiments, the training data of the first neural network model includes feature data of multiple modalities of a historical merchant at a first moment and multiple classification labels corresponding to the historical merchant, and the training data of the second neural network model includes feature data of multiple modalities of the historical merchant during each second time period of one or more second time periods, feature data of operating behaviors, and anomaly detection labels corresponding to the historical merchant.

[0016] According to the method for identifying merchant entities described in one embodiment of the present application or any of the above embodiments, during the training of the second neural network model, the convolution kernel parameter matrix during the current second time period is updated based on the feature data of multiple modalities of the historical merchant during the current second time period and the convolution kernel parameter matrix during the previous second time period.

[0017] According to the method for identifying a merchant entity described in one embodiment of the present application or any of the above embodiments, using a second neural network model to detect the operating behavior of the target merchant to determine the abnormal detection result of the target merchant includes: obtaining characteristic data of the operating behavior of the target merchant during each operating period; inputting the characteristic data of the operating behavior of the target merchant during each operating period into the second neural network model to determine the abnormal detection result of the target merchant.

[0018] According to the method for identifying merchant entities described in one embodiment or any of the above embodiments of the present application, weights are dynamically assigned to the feature data of each modality of the feature data of multiple modalities of the historical merchant based on the loss function of the first neural network model during the training of the first neural network model, including: using a linear normalization function to determine the first weight assigned to the feature data of each modality of the feature data of multiple modalities of the historical merchant and using the weighted feature data after weighting the feature data of each modality by the first weight to train the first neural network model; in response to the error loss of the loss function of the first neural network model being less than a threshold loss, using a softmax normalization function to determine the second weight assigned to the feature data of each modality of the feature data of multiple modalities of the historical merchant; and using the weighted feature data after weighting the feature data of each modality by the second weight to train the first neural network model.

[0019] According to a second aspect of the present application, a system for identifying a merchant entity is provided, the system comprising: a memory; a processor coupled to the memory; and a computer program stored on the memory and running on the processor, the running of the computer program resulting in the following operations: acquiring feature data of multiple modalities of a target merchant and using a first neural network model to determine a classification result of the target merchant at least based on the feature data of multiple modalities of the target merchant; selectively using a second neural network model based on the classification result of the target merchant to detect the operating behavior of the target merchant to determine an abnormal detection result of the target merchant; wherein the training data of the first neural network model comprises feature data of multiple modalities of historical merchants and multiple classification labels corresponding to the historical merchants, and during the training of the first neural network model, weights are dynamically assigned to the feature data of each modality of the feature data of multiple modalities of the historical merchant based on the loss function of the first neural network model.

[0020] According to a third aspect of the present application, a computer-readable storage medium is provided, which includes instructions, and the instructions, when run, execute the steps of the method for identifying a merchant entity according to the first aspect of the present application.

[0021] According to a fourth aspect of the present application, a computer program product is provided, the computer program product comprising instructions, which, when executed by a processor, implement the steps of the method for identifying a merchant entity according to the first aspect of the present application.

[0022] The scheme for identifying merchant entities proposed according to one or more embodiments of the present application can utilize a neural network model to classify merchants through the merchant's multiple modal feature data, and selectively utilize the neural network model to detect the merchant's operating behavior based on the classification results, thereby improving the coverage and accuracy of merchant entity classification and anomaly detection. The neural network model proposed according to one or more embodiments of the present application can dynamically assign weights to feature data of each modality in feature data of multiple modalities based on the loss function of the neural network model during training, thereby enhancing the neural network model's differentiated extraction and learning of features of each modality, and effectively improving the classification accuracy of the neural network model. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The above and / or other aspects and advantages of the present application will become clearer and easier to understand through the following description of various aspects in conjunction with the accompanying drawings, in which the same or similar units are represented by the same reference numerals. The accompanying drawings include:

[0024] Figure 1 A flow chart of a method for identifying a merchant entity according to one or more embodiments of the present application is shown.

[0025] Figure 2 An exemplary merchant node diagram according to an embodiment of the present application is shown.

[0026] Figure 3 A schematic diagram of the processing process of a graph convolutional neural network model according to an embodiment of the present application is shown.

[0027] Figure 4 A schematic diagram of the processing process of a recurrent neural network model according to an embodiment of the present application is shown.

[0028] Figure 5 A block diagram of a system for identifying a merchant entity according to one or more embodiments of the present application is shown. DETAILED DESCRIPTION

[0029] The present application is described more fully below with reference to the accompanying drawings in which illustrative embodiments of the present application are illustrated. However, the present application may be implemented in different forms and should not be interpreted as being limited to the embodiments given herein. The above embodiments are given to make the disclosure herein comprehensive and complete, so as to more fully convey the scope of protection of the present application to those skilled in the art.

[0030] In this specification, terms such as "comprise" and "include" indicate that in addition to the units and steps directly and explicitly stated in the specification and claims, the technical solution of the present application does not exclude the situation where there are other units and steps that are not directly or explicitly stated.

[0031] Unless otherwise specified, terms such as "first" and "second" do not indicate the order of the units in terms of time, space, size, etc., but are merely used to distinguish the units.

[0032] Hereinafter, various exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings.

[0033] Figure 1 A flow chart of a method for identifying a merchant entity according to one or more embodiments of the present application is shown.

[0034] like Figure 1 As shown in , in step S101, feature data of multiple modes of a target merchant are obtained and a first neural network model is used to determine a classification result of the target merchant based at least on the feature data of multiple modes of the target merchant, wherein the training data of the first neural network model includes feature data of multiple modes of historical merchants and multiple classification labels corresponding to the historical merchants. During the training of the first neural network model, weights are dynamically assigned to feature data of each mode of the feature data of multiple modes of the historical merchant based on the loss function of the first neural network model.

[0035] In one embodiment, the characteristic data of multiple modes of a merchant may include characteristic data of a text mode, characteristic data of an image mode, and characteristic data of a location mode of the merchant, wherein the mode may be understood as the source or expression form of the merchant information. In one embodiment, the characteristic data of the text mode may be obtained from the merchant's text information, the characteristic data of the image mode may be obtained from the merchant's image information, and the characteristic data of the location mode may be obtained from the merchant's location information. Exemplarily, the merchant's text information may include the merchant's business name, legal name, business type, etc., the merchant's image information may include the merchant's business license image, store image, etc., and the merchant's location information may include the merchant's map point of interest (POI), which may include the merchant's address and location (e.g., longitude and latitude), etc. Exemplarily, the merchant's text information, image information, and location information may be provided by the merchant when the merchant registers on the network.

[0036] Exemplarily, the text information of the merchant may be encoded to obtain feature data of the text modality, for example, the text information of the merchant may be encoded by one-hot encoding, position encoding, word embedding encoding, etc. to obtain feature data of the text modality. Exemplarily, the image information of the merchant may be encoded to obtain feature data of the image modality, for example, the image information of the merchant may be subjected to feature extraction and downsampling by convolution and pooling operations to obtain feature data of the image modality. Exemplarily, the location information of the merchant may be subjected to map encoding to obtain feature data of the location modality, for example, the location information of the merchant may be encoded by Geohash to obtain feature data of the location modality.

[0037] In one embodiment, the target merchant can be understood as a newly added merchant to be identified, and the historical merchant can be understood as an existing merchant in the database. In the training data of the first neural network model, the multiple classification labels corresponding to the historical merchants can be set differently depending on the application scenario. For example, in the identification scenario of abnormal merchants, the classification labels can be set to normal and abnormal, in the credit classification management scenario, the classification labels can be set to high credit rating, medium credit rating and low credit rating, and in the risk control scenario, the classification labels can be set to high risk, medium risk and low risk, etc.

[0038] In one embodiment, the classification result of the target merchant can be used to indicate that the category of the target merchant belongs to one of multiple classification labels, for example, to indicate that the category of the target merchant belongs to normal or abnormal, or to indicate that the category of the target merchant belongs to one of high credit level, medium credit level and low credit level.

[0039] In one embodiment, the first neural network model can be implemented as a graph convolutional neural network model. Figure 3 The schematic processing process of the graph convolutional neural network model is further described.

[0040] In one embodiment, during the training of the first neural network model, a linear normalization function may be first used to determine a first weight assigned to feature data of each of the multiple modal feature data in the training data, and the first neural network model may be trained using weighted feature data obtained by weighting the feature data of each modal by the first weight. When the error loss of the loss function of the first neural network model is less than a threshold loss, a softmax normalization function may be used to determine a second weight assigned to feature data of each of the multiple modal feature data in the training data, and the first neural network model may be trained using weighted feature data obtained by weighting the feature data of each modal by the second weight, until the first neural network model converges. In the early stages of training, the use of a linear normalization function can help the first neural network model converge faster. When the error loss of the loss function of the first neural network model is less than a threshold loss, the use of a softmax normalization function can help the first neural network model adjust the weights more finely, thereby improving the accuracy of model prediction. Exemplarily, the loss function of the first neural network model may be implemented as a cross entropy loss function, and the error loss may be a cross entropy loss. For example, the threshold loss can be determined based on the error loss of the validation set, or based on a certain percentage of the training progress. By combining the training strategy of the linear normalization function and the softmax normalization function, the advantages of each function can be brought into play at different training stages, thereby improving the overall performance of the neural network model.

[0041] In one embodiment, the target merchants and historical merchants can be used as graph nodes to construct a merchant node graph, the feature vector of the graph node is determined based on the feature data of multiple modes of the target merchant and the feature data of multiple modes of the historical merchant, the adjacency matrix of the graph node is determined based on the similarity between the feature vectors of the graph nodes, and the feature vector of the graph node and the adjacency matrix of the graph node are input into the first neural network model to update the layer node features of the first neural network model. In one embodiment, in the constructed merchant node graph, the feature vector and adjacency matrix of the graph node can be determined for each modality. When determining the adjacency matrix of the graph node, for any two graph nodes of the target merchant and the historical merchant, the similarity between the feature vectors of the two graph nodes can be calculated, for example, by calculating the Euclidean distance, cosine similarity, Manhattan distance, etc., thereby determining the adjacency matrix of the graph node based on the similarity between the feature vectors of each graph node and other graph nodes. Exemplarily, the training process of the graph convolutional neural network model GCN can be represented by the following formula (1):

[0042]

[0043] Among them, H (L+1) represents the node feature matrix of the L+1th layer of the graph convolutional neural network model, α represents the weight of the feature data of each modality assigned to the target merchant during the training process, which can be expressed as {α t ,α i ,α p} T , where α t , α i , α p denote the weights assigned to the feature data of the text modality, the feature data of the image modality, and the feature data of the position modality, respectively. (L) ,W (L) ,A) text Middle, H (L) represents the node feature matrix of the text modality of the Lth layer of the graph convolutional neural network model (which is composed of the concatenation of the feature vectors of each graph node), W (L) represents the convolution kernel parameter matrix of the Lth layer of the graph convolutional neural network model, A represents the adjacency matrix of the graph nodes of the text modality; in (H (L) ,W (L) ,A) image Middle, H (L) Represents the node feature matrix of the image modality of the Lth layer of the graph convolutional neural network model, W (L) represents the convolution kernel parameter matrix of the Lth layer of the graph convolutional neural network model, A represents the adjacency matrix of the graph nodes of the image modality; in (H (L) ,W (L) ,A) poi Middle, H (L) Represents the node feature matrix of the position mode of the Lth layer of the graph convolutional neural network model, W (L) represents the convolution kernel parameter matrix of the Lth layer of the graph convolutional neural network model, and A represents the adjacency matrix of the graph nodes in the position mode.

[0044] In step S103, based on the classification result of the target merchant, the second neural network model is selectively used to detect the operation behavior of the target merchant to determine the abnormal detection result of the target merchant.

[0045] Optionally, in step S103, when the classification result of the target merchant indicates that the target merchant is normal, the second neural network model can be used to detect the operating behavior of the target merchant to determine the abnormal detection result of the target merchant; when the classification result of the target merchant indicates that the target merchant is abnormal, processing measures such as removing related products of the target merchant from the shelves, restricting merchant permissions, and suspending transactions can be taken.

[0046] In one embodiment, the target merchant's operation behavior may include: the target merchant's modification behavior of one or more of its text information, image information and location information, and the target merchant's transaction behavior (for example, including transaction amount, number of transactions, transaction method, etc.). In one embodiment, the second neural network model can be implemented as a recurrent neural network model. Figure 4 The schematic processing process of the recurrent neural network model is further described. By implementing the second neural network model as a recurrent neural network model, the operation behavior of the target merchant can be tracked and monitored after the merchant registers on the network, thereby improving the tracking and monitoring capability of the merchant's information modification or transaction behavior after the merchant registers on the network.

[0047] In one embodiment, the training data of the first neural network model may include feature data of multiple modes of the historical merchant at the first moment and multiple classification labels corresponding to the historical merchant. The first moment may be, for example, the network registration moment of the historical merchant. In one embodiment, the training data of the second neural network model may include feature data of multiple modes of the historical merchant during each second period of one or more second periods, feature data of operation behavior and abnormality detection labels corresponding to the historical merchant. The second period may be, for example, one or more periods after the historical merchant is registered on the network, and the abnormality detection labels may include, for example, normal and abnormal. Exemplarily, after the historical merchant is registered on the network, the feature data of multiple modes, feature data of operation behavior and abnormality detection labels of each historical merchant during each hour of 120 hours in 5 working days may be used as the training data of the second neural network model. Exemplarily, the feature data of operation behavior may include information modification features and transaction behavior features, wherein the information modification features may include one or more of the encoding features of the text information modified by the merchant in the current period compared with the previous period, the encoding features of the modified image information, and the encoding features of the modified location information, and the transaction behavior features may include the transaction amount, number of transactions, transaction method, etc., of the merchant in the current period.

[0048] In one embodiment, during the training of the second neural network model, the convolution kernel parameter matrix during the current second period can be updated based on the feature data of multiple modes of the historical merchant during the current second period and the convolution kernel parameter matrix during the previous second period. Exemplarily, the convolution kernel parameter matrix can be updated by the following formula (2):

[0049]

[0050] Among them, W (L) represents the convolution kernel parameter matrix of the Lth layer of the recurrent neural network model during the t period, represents the convolution kernel parameter matrix of the Lth layer of the recurrent neural network model during the t-1 period, represents the node feature matrix of the Lth layer during period t, and Φ represents the encoding function of the recurrent neural network model.

[0051] Optionally, in step S103, characteristic data of the target merchant's operation behavior during each operation period may be obtained, and the characteristic data of the target merchant's operation behavior during each operation period may be input into the second neural network model to determine an abnormal detection result of the target merchant, such as a normal or abnormal detection result. For example, after the target merchant is registered on the network, characteristic data of the target merchant's operation behavior may be obtained during each hour of the 24 hours of the day, and the obtained characteristic data of the operation behavior may be input into the second neural network model to determine an abnormal detection result of the target merchant, thereby achieving tracking and monitoring of the merchant after the registration on the network.

[0052] The method for identifying merchant entities proposed according to one or more embodiments of the present application can utilize a neural network model to classify merchants through the merchant's multiple modal feature data, and selectively utilize the neural network model to detect the merchant's operating behavior based on the classification results, thereby improving the coverage and accuracy of merchant entity classification and anomaly detection. The neural network model proposed according to one or more embodiments of the present application can dynamically assign weights to feature data of each modality in feature data of multiple modalities based on the loss function of the neural network model during training, thereby enhancing the neural network model's differentiated extraction and learning of features of each modality, and effectively improving the classification accuracy of the neural network model.

[0053] Figure 2 An exemplary merchant node diagram according to an embodiment of the present application is shown.

[0054] like Figure 2 As shown in , the merchant node graph 200 includes nodes E1, E2, E3, E4, ..., Em and nodes N1, N2, N3, ..., Nk, wherein nodes E1, E2, E3, E4, ..., Em represent target merchants, and nodes N1, N2, N3, ..., Nk represent historical merchants. Figure 2 In , I, T and P represent image modality, text modality and position modality respectively.

[0055] exist Figure 2 In the example, the target merchant's nodes E1, E2, E3, E4, ..., Em and the historical merchant's nodes N1, N2, N3, ..., Nk can establish connecting edges via the image modality I, text modality T and location modality P, that is, Figure 2 The straight line with an arrow shown in .

[0056] The connection edge established between the target merchant's node and the historical merchant's node via the image modality I may indicate that the target merchant and the historical merchant have a correlation in the image modality, and the correlation may be determined based on the similarity between the feature vector of the target merchant's node and the feature vector of the historical merchant's node in the image modality. Figure 2 As shown, a connection edge is established between the target merchant's node E1 and the historical merchant's node N1 via the image modality I, which may indicate that the target merchant represented by the node E1 and the historical merchant represented by the node N1 are correlated in terms of image modality.

[0057] The connection edge established between the target merchant's node and the historical merchant's node via the text modality T may indicate that the target merchant and the historical merchant have a correlation in terms of text modality, and the correlation may be determined based on the similarity between the feature vector of the target merchant's node and the feature vector of the historical merchant's node in the text modality. Figure 2 As shown, a connection edge is established between the target merchant's node E3 and the historical merchant's node N2 via the text modality T, which may indicate that the target merchant represented by the node E3 and the historical merchant represented by the node N2 are related in terms of text modality.

[0058] The connection edge established between the target merchant's node and the historical merchant's node via the location modality P may indicate that the target merchant and the historical merchant have a correlation in the location modality, and the correlation may be determined based on the similarity between the feature vector of the target merchant's node and the feature vector of the historical merchant's node in the location modality. Figure 2 As shown, a connection edge is established between the node Em of the target merchant and the node NK of the historical merchant via the position mode P, which can indicate that the target merchant represented by the node Em and the historical merchant represented by the node Nk are related in terms of text mode.

[0059] It should be noted that the target merchant and the historical merchant may be related in multiple modes. Figure 2 As shown, a connection edge is established between the target merchant's node E3 and the historical merchant's node N2 via both the image modality I and the text modality T, which may indicate that the target merchant represented by the node E3 and the historical merchant represented by the node N2 are correlated in terms of the image modality and the text modality.

[0060] exist Figure 2In the merchant node graph 200 shown in , by establishing connecting edges between the node of the target merchant and the node N of the historical merchant via the image modality I, the text modality T, and the position modality P, the adjacency matrix of the graph nodes can be determined for each modality. Exemplarily, in the adjacency matrix for the image modality I, the element located in the first row and the first column can be set to 1, which is used to indicate that the target merchant represented by the node E1 and the historical merchant represented by the node N1 are correlated in terms of the image modality, and the element located in the first row and the second column can be set to 0, which is used to indicate that the target merchant represented by the node E1 and the historical merchant represented by the node N2 are not correlated in terms of the image modality.

[0061] It should be noted that Figure 2 The merchant node graph according to one embodiment of the present application is shown only as an example. Without departing from the spirit and scope of the present application, the merchant node graph may be shown in other forms, such as other forms of bipartite directed graphs.

[0062] Figure 3 A schematic diagram of the processing process of a graph convolutional neural network model according to an embodiment of the present application is shown.

[0063] exist Figure 3 In the processing process 300 of the graph convolutional neural network model shown in , the text information, image information and location information of the target merchant can be obtained. Exemplarily, the text information, image information and location information of the target merchant can be obtained when the target merchant registers on the network. Exemplarily, the text information of the target merchant may include the merchant's business name, legal name, business type, etc., the image information of the target merchant may include the merchant's business license image, store image, etc., and the location information of the target merchant may include the merchant's map POI, which may include the merchant's address and location (such as longitude and latitude), etc. Exemplarily, the text information, image information and location information of the target merchant can be provided by the merchant when the merchant registers on the network.

[0064] After obtaining the text information, image information and location information of the target merchant, the text information of the target merchant can be encoded by a text encoder to obtain feature data of the text modality. For example, the text information of the target merchant can be encoded by a single hot encoder, a position encoder, a word embedding encoder, etc. to obtain feature data of the text modality. The image information of the target merchant can be encoded by an image encoder to obtain feature data of the image modality. For example, the image encoder can extract features and downsample the image information of the target merchant through convolution and pooling operations to obtain feature data of the image modality. The location information of the target merchant can be map-encoded by a position encoder to obtain feature data of the location modality. For example, the location information of the target merchant can be encoded by a Geohash encoder to obtain feature data of the location modality. Then, the feature data of the text modality, the feature data of the image modality and the feature data of the location modality of the target merchant can be input into the graph convolutional neural network model to obtain the classification result of the target merchant.

[0065] Figure 4 A schematic diagram of the processing process of a recurrent neural network model according to an embodiment of the present application is shown.

[0066] exist Figure 4 In the processing process 400 of the recurrent neural network model shown in , the feature data input to the initial input layer of the target merchant during each operation period can be obtained. Figure 4 The feature data x input to the initial input layer are obtained during the t-1 period, t period, t+1 period and t+2 period shown in t-1 、x t 、x t+1 and x t+2 , which may include feature data of multiple modalities of target merchants obtained in respective time periods.

[0067] exist Figure 4 middle, and can represent the feature data input to the first hidden layer during the t-1 period, the t period, the t+1 period, and the t+2 period, respectively, which can respectively represent the feature data x input to the initial input layer t-1 、x t 、x t+1 and x t+2 The information modification feature y during the period t-1, period t, period t+1 and period t+2 t-1 ,y t ,y t+1 and t+2 Take the t period as an example, it can be obtained by the following formula (3):

[0068]

[0069] Among them, F concat represents the concatenation function, represents the feature data input to the first hidden layer during the t period, x t represents the feature data input to the initial input layer during the t period, y t Represents the information modification characteristics during period t.

[0070] Continue as Figure 4 As shown in and The convolution kernel parameter matrix of the first hidden layer of the recurrent neural network model during the t-1 period, the t period, the t+1 period, and the t+2 period can be represented respectively, and The convolution kernel parameter matrix of the second hidden layer of the recurrent neural network model during the t-1 period, the t period, the t+1 period and the t+2 period can be represented respectively. The convolution kernel parameter matrix of each period can be determined by the feature data input to the first hidden layer of the current period and the convolution kernel parameter matrix of the previous period, for example, by referring to the above formula (2), where Φ represents the encoding function of the recurrent neural network model.

[0071] exist Figure 4 middle, and can represent the feature data input to the second hidden layer during the t-1 period, t period, t+1 period and t+2 period respectively. Taking the t period as an example, it can be obtained by the following formula (4):

[0072]

[0073] Among them, F concat represents the concatenation function, σ represents the nonlinear classification activation function, represents the degree matrix (used to normalize the adjacency matrix), represents the adjacency matrix of the processed graph nodes (i.e., the 0s on the diagonal of the adjacency matrix are changed to 1s), represents the feature data input to the first hidden layer during the t period, Φ represents the encoding function of the recurrent neural network model, represents the convolution kernel parameter matrix of the first hidden layer of the recurrent neural network model during the t-1 period, z t Represents the characteristics of trading behavior during period t.

[0074] Continue as Figure 4As shown in , the feature data output by the second hidden layer is input into the classification function to obtain the abnormality detection result of the target merchant.

[0075] Figure 5 A block diagram of a system for identifying a merchant entity according to one or more embodiments of the present application is shown.

[0076] like Figure 5 As shown in FIG. 5 , a system 500 for identifying a merchant entity includes a memory 510, a processor 520, and a computer program 530 stored in the memory 510 and executable on the processor 520. The processor 520 executes the computer program 530 to implement a method for identifying a merchant entity according to one aspect of the present application.

[0077] The present application may also be implemented as a computer-readable storage medium, the computer storage medium including instructions, the instructions when running, executing a method for identifying a merchant entity according to one aspect of the present application. In addition, the present application may also be implemented as a computer-readable storage medium, the computer storage medium including instructions, the instructions when running, executing a method for identifying a merchant entity according to one aspect of the present application.

[0078] In the applicable situation, hardware, software or a combination of hardware and software can be used to realize the various embodiments provided by the application. Moreover, in the applicable situation, without departing from the scope of the application, the various hardware components and / or software components set forth herein can be combined into a composite component comprising software, hardware and / or both. In the applicable situation, without departing from the scope of the application, the various hardware components and / or software components set forth herein can be divided into subcomponents comprising software, hardware or both. In addition, in the applicable situation, it is contemplated that the software component can be implemented as a hardware component, and vice versa.

[0079] Software (such as program code and / or data) according to the present application can be stored on one or more computer storage media. It is also contemplated that the software identified herein can be implemented using one or more general or special computers and / or systems, networked and / or otherwise. Where applicable, the order of the various steps described herein can be changed, combined into composite steps and / or divided into sub-steps to provide the features described herein.

[0080] The embodiments and examples set forth herein are provided to best illustrate embodiments according to the present application and its specific applications, and thereby enable those skilled in the art to implement and use the present application. However, those skilled in the art will appreciate that the above description and examples are provided only for ease of illustration and example. The description set forth is not intended to cover all aspects of the present application or to limit the present application to the precise form disclosed.

Claims

1. A method for identifying a merchant entity, characterized in that: The method comprises the following steps: Acquire feature data of multiple modalities of a target merchant and use a first neural network model to determine a classification result of the target merchant based at least on the feature data of multiple modalities of the target merchant; Selectively using a second neural network model to detect the operation behavior of the target merchant based on the classification result of the target merchant to determine an abnormal detection result of the target merchant; The training data of the first neural network model includes feature data of multiple modalities of historical merchants and multiple classification labels corresponding to the historical merchants. During the training of the first neural network model, weights are dynamically assigned to the feature data of each modality of the feature data of the multiple modalities of the historical merchants based on the loss function of the first neural network model.

2. The method according to claim 1, wherein the characteristic data of the target merchant in multiple modes comprises a combination of one or more of the following: Feature data of text modality obtained from the text information of the target merchant; Feature data of an image modality obtained from the image information of the target merchant; Feature data of the location modality obtained from the location information of the target merchant.

3. The method according to claim 1, wherein the characteristic data of multiple modalities of the historical merchant comprises a combination of one or more of the following: Feature data of text modality obtained from the text information of the historical merchant; Feature data of image modality obtained from image information of the historical merchant; Feature data of the location modality obtained from the location information of the historical merchants.

4. The method according to claim 1, wherein the first neural network model is a graph convolutional neural network model.

5. The method of claim 1, wherein during training of the first neural network model: Using the target merchant and the historical merchant as graph nodes to construct a merchant node graph; Determining a feature vector of the graph node based on the feature data of multiple modes of the target merchant and the feature data of multiple modes of the historical merchant; Determining an adjacency matrix of the graph nodes based on similarities between feature vectors of the graph nodes; as well as The feature vector of the graph node and the adjacency matrix of the graph node are input into the first neural network model to update the layer node features of the first neural network model. 6 . The method according to claim 1 , wherein the classification result of the target merchant is used to indicate that the category of the target merchant belongs to one of the multiple classification labels.

7. The method according to claim 1, wherein selectively using a second neural network model to detect the operation behavior of the target merchant based on the classification result of the target merchant comprises: In response to the classification result of the target merchant indicating that the category of the target merchant belongs to the first classification label among the multiple classification labels, the operation behavior of the target merchant is detected.

8. The method according to claim 2, wherein the target merchant's operation behavior includes: The target merchant's modification behavior of one or more of the text information, image information and location information, and the target merchant's transaction behavior.

9. The method according to claim 1, wherein the second neural network model is a recurrent neural network model.

10. The method according to claim 1, wherein the training data of the first neural network model includes feature data of multiple modalities of historical merchants at a first moment and multiple classification labels corresponding to the historical merchants, and the training data of the second neural network model includes feature data of multiple modalities of historical merchants during each second time period of one or more second time periods, feature data of operating behaviors, and anomaly detection labels corresponding to the historical merchants.

11. The method according to claim 10, wherein during the training of the second neural network model, the convolution kernel parameter matrix during the current second time period is updated based on the feature data of multiple modalities of the historical merchant during the current second time period and the convolution kernel parameter matrix during the previous second time period.

12. The method according to claim 1, wherein using a second neural network model to detect the operation behavior of the target merchant to determine the abnormal detection result of the target merchant comprises: Acquire characteristic data of the operation behavior of the target merchant during each operation period; The characteristic data of the operation behavior of the target merchant during each operation period is input into the second neural network model to determine the abnormal detection result of the target merchant.

13. The method of claim 1, wherein dynamically assigning weights to feature data of each modality of feature data of multiple modalities of the historical merchant based on a loss function of the first neural network model during training of the first neural network model comprises: Using a linear normalization function to determine a first weight assigned to feature data of each modality of feature data of a plurality of modalities of the historical merchant and using weighted feature data obtained by weighting the feature data of each modality using the first weight to train the first neural network model; In response to the error loss of the loss function of the first neural network model being less than a threshold loss, using a softmax normalization function to determine a second weight assigned to feature data of each of the multiple modalities of feature data of the historical merchant; and The first neural network model is trained using weighted feature data obtained by weighting feature data of each modality using the second weight.

14. A system for identifying a merchant entity, characterized in that: The system comprises: Memory; a processor coupled to the memory; and A computer program stored on the memory and running on the processor, the execution of the computer program causing the following operations: Acquire feature data of multiple modalities of a target merchant and use a first neural network model to determine a classification result of the target merchant based at least on the feature data of multiple modalities of the target merchant; Selectively using a second neural network model to detect the operation behavior of the target merchant based on the classification result of the target merchant to determine an abnormal detection result of the target merchant; The training data of the first neural network model includes feature data of multiple modalities of historical merchants and multiple classification labels corresponding to the historical merchants. During the training of the first neural network model, weights are dynamically assigned to the feature data of each modality of the feature data of the multiple modalities of the historical merchants based on the loss function of the first neural network model.

15. The system according to claim 14, wherein the characteristic data of the target merchant in multiple modes comprises a combination of one or more of the following: Feature data of text modality obtained from the text information of the target merchant; Feature data of an image modality obtained from the image information of the target merchant; Feature data of the location modality obtained from the location information of the target merchant.

16. The system according to claim 14, wherein the characteristic data of the multiple modalities of the historical merchants include a combination of one or more of the following: Feature data of text modality obtained from the text information of the historical merchant; Feature data of image modality obtained from image information of the historical merchant; Feature data of the location modality obtained from the location information of the historical merchants.

17. The system of claim 14, wherein the first neural network model is a graph convolutional neural network model.

18. The system of claim 14, wherein execution of the computer program results in, during training of the first neural network model: Using the target merchant and the historical merchant as graph nodes to construct a merchant node graph; Determining a feature vector of the graph node based on the feature data of multiple modes of the target merchant and the feature data of multiple modes of the historical merchant; Determining an adjacency matrix of the graph nodes based on similarities between feature vectors of the graph nodes; as well as The feature vector of the graph node and the adjacency matrix of the graph node are input into the first neural network model to update the layer node features of the first neural network model.

19. The system according to claim 14, wherein the classification result of the target merchant is used to indicate that the category of the target merchant belongs to one of the multiple classification labels.

20. The system according to claim 14, wherein the operation of the computer program causes the selective use of a second neural network model to detect the operation behavior of the target merchant based on the classification result of the target merchant, comprising: In response to the classification result of the target merchant indicating that the category of the target merchant belongs to the first classification label among the multiple classification labels, the operation behavior of the target merchant is detected.

21. The system according to claim 15, wherein the target merchant's operation behavior includes: The target merchant's modification behavior of one or more of the text information, image information and location information, and the target merchant's transaction behavior.

22. The system of claim 14, wherein the second neural network model is a recurrent neural network model.

23. A system according to claim 14, wherein the training data of the first neural network model includes feature data of multiple modalities of historical merchants at a first moment and multiple classification labels corresponding to the historical merchants, and the training data of the second neural network model includes feature data of multiple modalities of historical merchants during each second time period of one or more second time periods, feature data of operating behaviors and anomaly detection labels corresponding to the historical merchants.

24. A system according to claim 23, wherein the execution of the computer program causes the convolution kernel parameter matrix during the current second time period to be updated based on the feature data of multiple modalities of the historical merchant during the current second time period and the convolution kernel parameter matrix during the previous second time period during the training of the second neural network model.

25. The system according to claim 14, wherein the operation of the computer program results in using a second neural network model to detect the operation behavior of the target merchant to determine an abnormal detection result of the target merchant, including: Acquire characteristic data of the operation behavior of the target merchant during each operation period; The characteristic data of the operation behavior of the target merchant during each operation period is input into the second neural network model to determine the abnormal detection result of the target merchant.

26. The system of claim 14, wherein the execution of the computer program results in dynamically assigning weights to feature data of each of the plurality of modalities of feature data of the historical merchant based on a loss function of the first neural network model during training of the first neural network model, comprising: Using a linear normalization function to determine a first weight assigned to feature data of each modality of feature data of a plurality of modalities of the historical merchant and using weighted feature data obtained by weighting the feature data of each modality using the first weight to train the first neural network model; In response to the error loss of the loss function of the first neural network model being less than a threshold loss, using a softmax normalization function to determine a second weight assigned to feature data of each of the multiple modalities of feature data of the historical merchant; and The first neural network model is trained using weighted feature data obtained by weighting feature data of each modality using the second weight.

27. A computer-readable storage medium, characterized in that: The computer storage medium includes instructions, which, when executed, execute the method for identifying a merchant entity according to any one of claims 1-13.

28. A computer program product, characterized in that The computer program product comprises instructions, which, when executed by a processor, implement the method for identifying a merchant entity according to any one of claims 1 to 13.

Images