Risk control model modeling, risk control method and device

Abstract

One or more embodiments of the specification disclose a risk control model modeling method and device. The method comprises: obtaining a risk control event feature matrix corresponding to each risk control event sequence in a risk control event sequence set, performing linear transformation processing on the risk control event feature matrix based on a pre-trained risk control event weight matrix to obtain a risk control event query matrix, a risk control event key matrix and a risk control event value matrix, determining a first shared key matrix corresponding to the risk control event sequence set according to the risk control event key matrix corresponding to the risk control event feature matrix, and calculating a risk control event fusion feature matrix corresponding to each risk control event sequence according to the risk control event query matrix, the first shared key matrix and the risk control event value matrix, wherein the risk control event fusion feature matrix is used to describe sequence representation of the risk control event sequence, and then a risk control model for risk control is constructed based on the risk control event fusion feature matrix.

Description

TECHNICAL FIELD

[0001] The present specification relates to the technical field of computer technology, and particularly relates to a risk control model modeling method and device and a risk control method. BACKGROUND

[0002] At present, with the continuous development of computer technology, researchers can analyze personnel behavior sequences to mine whether personnel behaviors are risky. For example, a risk control model often uses a risk control event sequence to model, to identify whether similar events are risk events. In implementation, a representation of the risk control event sequence needs to be extracted, so as to construct a risk control model using the sequence representation.

[0003] In related technologies, a sequence representation of a risk control event sequence can be extracted respectively to construct a risk control model. However, this sequence representation extraction method ignores the association between multiple risk control event sequences, so that the accuracy of the constructed risk control model is low. Therefore, there is an urgent need to provide a more accurate sequence representation extraction scheme. SUMMARY

[0004] In one aspect, one or more embodiments of the present specification provide a risk control model modeling method, including: obtaining a risk control event feature matrix corresponding to each risk control event sequence in a risk control event sequence set, respectively performing linear transformation processing on the risk control event feature matrix based on a pre-trained risk control event weight matrix, to obtain a risk control event query matrix, a risk control event key matrix and a risk control event value matrix, determining a first shared key matrix corresponding to the risk control event sequence set according to the risk control event key matrix corresponding to the risk control event feature matrix, the first shared key matrix being obtained based on fusing the risk control event key matrix corresponding to each risk control event sequence, for each risk control event sequence, calculating a risk control event fusion feature matrix corresponding to the risk control event sequence according to the risk control event query matrix, the first shared key matrix and the risk control event value matrix, the risk control event fusion feature matrix being used to describe a sequence representation of the risk control event sequence, and constructing a risk control model for risk control based on the risk control event fusion feature matrix.

[0005] In another aspect, one or more embodiments of the present specification provide a sequence representation extraction method, comprising: obtaining an event feature matrix corresponding to each event sequence in a set of to-be-processed sequences, performing linear transformation processing on the event feature matrix based on a pre-trained event weight matrix to obtain an event query matrix, an event key matrix, and an event value matrix, determining a second shared key matrix corresponding to the set of to-be-processed sequences according to the event key matrix corresponding to the event feature matrix, the second shared key matrix being obtained based on fusing the event key matrix corresponding to each event sequence, and for each event sequence, calculating an event fusion feature matrix corresponding to the event sequence according to the event query matrix, the second shared key matrix, and the event value matrix, the event fusion feature matrix being used to describe the sequence representation of the event sequence.

[0006] In another aspect, one or more embodiments of the present specification provide a risk control method, comprising: obtaining a target event to be subjected to risk control, inputting the target event into a risk control model to obtain a risk detection result, and determining a risk control measure for the target event according to the risk detection result. The risk control model is constructed based on a risk control event fusion feature matrix corresponding to each risk control event sequence in a set of risk control event sequences, the risk control event fusion feature matrix being obtained by calculating a risk control event query matrix, a first shared key matrix, and a risk control event value matrix corresponding to the risk control event sequence, and the first shared key matrix being obtained based on fusing a risk control event key matrix corresponding to each risk control event sequence.

[0007] In another aspect, one or more embodiments of the present specification provide a risk control model modeling device, comprising: a first obtaining module configured to obtain a risk control event feature matrix corresponding to each risk control event sequence in a set of risk control event sequences. A first linear transformation processing module is configured to perform linear transformation processing on the risk control event feature matrix based on a pre-trained risk control event weight matrix to obtain a risk control event query matrix, a risk control event key matrix, and a risk control event value matrix. A first determining module is configured to determine a first shared key matrix corresponding to the set of risk control event sequences according to the risk control event key matrix corresponding to the risk control event feature matrix, the first shared key matrix being obtained based on fusing the risk control event key matrix corresponding to each risk control event sequence. A first calculating module is configured to, for each risk control event sequence, calculate a risk control event fusion feature matrix corresponding to the risk control event sequence according to the risk control event query matrix, the first shared key matrix, and the risk control event value matrix, the risk control event fusion feature matrix being used to describe the sequence representation of the risk control event sequence. A constructing module is configured to construct a risk control model for risk control based on the risk control event fusion feature matrix.

[0008] In still another aspect, one or more embodiments of the present specification provide a sequence representation extraction device, comprising: a second acquisition module configured to acquire an event feature matrix corresponding to each event sequence in a set of sequences to be processed. A second linear transformation processing module is configured to perform linear transformation processing on the event feature matrix based on a pre-trained event weight matrix to obtain an event query matrix, an event key matrix, and an event value matrix. A second determination module is configured to determine a second shared key matrix corresponding to the set of sequences to be processed according to the event key matrix corresponding to the event feature matrix, the second shared key matrix being obtained based on fusing the event key matrix corresponding to each event sequence. A second calculation module is configured to calculate, for each event sequence, an event fusion feature matrix corresponding to the event sequence according to the event query matrix, the second shared key matrix, and the event value matrix, the event fusion feature matrix being used to describe the sequence representation of the event sequence.

[0009] In still another aspect, one or more embodiments of the present specification provide a risk control device, comprising: a third acquisition module configured to acquire a target event to be subjected to risk control. A model processing module is configured to input the target event into a risk control model to obtain a risk detection result. A third determination module is configured to determine a risk control measure for the target event according to the risk detection result. The risk control model is constructed based on a risk control event fusion feature matrix corresponding to each risk control event sequence in a set of risk control event sequences, the risk control event fusion feature matrix being obtained by calculating a risk control event query matrix, a first shared key matrix, and a risk control event value matrix corresponding to the risk control event sequence, the first shared key matrix being obtained based on fusing a risk control event key matrix corresponding to each risk control event sequence.

[0010] In still another aspect, one or more embodiments of the present specification provide an electronic device, comprising a processor and a memory electrically connected with the processor, the memory storing a computer program, the processor being configured to call and execute the computer program from the memory to implement: obtaining an anti-fraud event feature matrix corresponding to each anti-fraud event sequence in an anti-fraud event sequence set, performing linear transformation processing on the anti-fraud event feature matrix based on a pre-trained anti-fraud event weight matrix to obtain an anti-fraud event query matrix, an anti-fraud event key matrix, and an anti-fraud event value matrix, determining a first shared key matrix corresponding to the anti-fraud event sequence set according to the anti-fraud event key matrix corresponding to the anti-fraud event feature matrix, the first shared key matrix being obtained based on fusing the anti-fraud event key matrix corresponding to each anti-fraud event sequence, for each anti-fraud event sequence, calculating an anti-fraud event fusion feature matrix corresponding to the anti-fraud event sequence according to the anti-fraud event query matrix, the first shared key matrix, and the anti-fraud event value matrix, the anti-fraud event fusion feature matrix being used to describe sequence representation of the anti-fraud event sequence, and constructing an anti-fraud model for risk control based on the anti-fraud event fusion feature matrix.

[0011] In still another aspect, one or more embodiments of the present specification provide an electronic device, comprising a processor and a memory electrically connected with the processor, the memory storing a computer program, the processor being configured to call and execute the computer program from the memory to implement: obtaining an anti-fraud event feature matrix corresponding to each anti-fraud event sequence in an anti-fraud event sequence set, performing linear transformation processing on the anti-fraud event feature matrix based on a pre-trained anti-fraud event weight matrix to obtain an anti-fraud event query matrix, an anti-fraud event key matrix, and an anti-fraud event value matrix, determining a first shared key matrix corresponding to the anti-fraud event sequence set according to the anti-fraud event key matrix corresponding to the anti-fraud event feature matrix, the first shared key matrix being obtained based on fusing the anti-fraud event key matrix corresponding to each anti-fraud event sequence, for each anti-fraud event sequence, calculating an anti-fraud event fusion feature matrix corresponding to the anti-fraud event sequence according to the anti-fraud event query matrix, the first shared key matrix, and the anti-fraud event value matrix, the anti-fraud event fusion feature matrix being used to describe sequence representation of the anti-fraud event sequence, and constructing an anti-fraud model for risk control based on the anti-fraud event fusion feature matrix.

[0012] In still another aspect, the one or more embodiments of the present specification provide an electronic device, comprising a processor and a memory electrically connected with the processor, the memory storing a computer program, and the processor being configured to invoke and execute the computer program from the memory to implement: obtaining a target event to be subjected to risk control, inputting the target event into a risk control model, obtaining a risk detection result, and determining a risk control measure for the target event according to the risk detection result. The risk control model is constructed based on a risk control event fusion feature matrix corresponding to each risk control event sequence in a risk control event sequence set, and the risk control event fusion feature matrix is obtained by calculating a risk control event query matrix, a first shared key matrix and a risk control event value matrix corresponding to the risk control event sequence.

[0013] In still another aspect, the one or more embodiments of the present specification provide an electronic device, comprising a processor and a memory electrically connected with the processor, the memory storing a computer program, and the processor being configured to invoke and execute the computer program from the memory to implement: obtaining a target event to be subjected to risk control, inputting the target event into a risk control model, obtaining a risk detection result, and determining a risk control measure for the target event according to the risk detection result. The risk control model is constructed based on a risk control event fusion feature matrix corresponding to each risk control event sequence in a risk control event sequence set, and the risk control event fusion feature matrix is obtained by calculating a risk control event query matrix, a first shared key matrix and a risk control event value matrix corresponding to the risk control event sequence.

[0014] In still another aspect, the one or more embodiments of the present specification provide an electronic device, comprising a processor and a memory electrically connected with the processor, the memory storing a computer program, and the processor being configured to invoke and execute the computer program from the memory to implement: obtaining a target event to be subjected to risk control, inputting the target event into a risk control model, obtaining a risk detection result, and determining a risk control measure for the target event according to the risk detection result. The risk control model is constructed based on a risk control event fusion feature matrix corresponding to each risk control event sequence in a risk control event sequence set, and the risk control event fusion feature matrix is obtained by calculating a risk control event query matrix, a first shared key matrix and a risk control event value matrix corresponding to the risk control event sequence.

[0015] In still another aspect, the embodiments of the present specification provide a storage medium for storing a computer program, which can be executed by a processor to implement the following process: obtaining a target event to be subjected to risk control, inputting the target event into a risk control model, obtaining a risk detection result, and determining a risk control measure for the target event according to the risk detection result. The risk control model is constructed based on a risk control event fusion feature matrix corresponding to each risk control event sequence in a risk control event sequence set, and the risk control event fusion feature matrix is obtained by calculating a risk control event query matrix, a first shared key matrix, and a risk control event value matrix corresponding to the risk control event sequence. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the one or more embodiments of the present specification or the prior art, the drawings needed in the embodiment or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments described in the one or more embodiments of the present specification, and other drawings can be obtained by those skilled in the art without creative labor.

[0017] Figure 1 is a schematic flow chart of a risk control model modeling method according to an embodiment of the present specification;

[0018] Figure 2 is a schematic diagram of the implementation principle of a risk control model modeling process according to an embodiment of the present specification;

[0019] Figure 3 is a schematic flow chart of a risk control model modeling method according to another embodiment of the present specification;

[0020] Figure 4 is a schematic flow chart of a sequence representation extraction method according to an embodiment of the present specification;

[0021] Figure 5 is a schematic flow chart of a risk control method according to an embodiment of the present specification;

[0022] Figure 6 is a schematic flow chart of a risk control method according to another embodiment of the present specification;

[0023] Figure 7 is a schematic block diagram of a risk control model modeling device according to an embodiment of the present specification;

[0024] Figure 8is a schematic block diagram of a sequence feature extraction device according to an embodiment of the present specification;

[0025] Figure 9 is a schematic block diagram of a risk control device according to an embodiment of the present specification;

[0026] Figure 10 is a schematic block diagram of an electronic device according to an embodiment of the present specification. DETAILED DESCRIPTION

[0027] One or more embodiments of the present specification provide a risk control model modeling, a risk control method and device to solve the problem of low accuracy of the extracted sequence features, resulting in low accuracy of the risk control model.

[0028] In order for those skilled in the art to better understand the technical solutions in one or more embodiments of the present specification, the technical solutions in one or more embodiments of the present specification will be described clearly and completely below in conjunction with the accompanying drawings of one or more embodiments of the present specification. Obviously, the described embodiments are only some of the embodiments of the present specification, not all. Based on one or more embodiments of the present specification, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of one or more embodiments of the present specification.

[0029] Currently, the risk control model often uses risk control event sequence modeling. For example, for transaction risk control events, the risk control model for judging whether a transaction is risky is constructed using the payment behavior sequence and the collection behavior sequence of the event initiator (such as the payer of the transaction event) in the recent period of time, and the collection behavior sequence and the payment behavior sequence of the event passive party (such as the payee of the transaction event) in the recent period of time. In the specific implementation process, sequence features can be extracted from the sequence set based on the attention mechanism, so as to construct the risk control model using the sequence features. In related technologies, multiple sequences in the sequence set are usually spliced into a long sequence, and then the sequence features are extracted based on the attention mechanism for the long sequence as a whole, as the sequence features of the sequence set. Assuming that the length of each sequence is n, the dimension is d, and the number of sequences in the set is m, since the algorithm complexity of the attention mechanism is O(n 2 *d), then after splicing m sequences, the algorithm complexity of extracting sequence features is O((n*m) 2d), obviously, the algorithm complexity of this sequence representation extraction manner is high, so that it is difficult to ensure the extraction efficiency of the sequence representation. Therefore, the skilled person extracts the sequence representation based on the attention mechanism for each sequence respectively, and finally connects multiple sequence representations together as the sequence representation of the sequence set, so as to ensure the extraction efficiency of the sequence representation. However, since this sequence representation extraction manner only connects at the last representation layer, the interaction information between different sequences is lost, so that it is difficult to ensure the accuracy of the extracted sequence representation. Based on the above problems, the embodiment of the present specification provides a risk control model modeling, sequence representation extraction, risk control method and device, which can ensure the accuracy of the extracted sequence representation under the condition of low algorithm complexity, so as to ensure the accuracy of the risk control model, and then realize accurate risk control of the to-be-risk-controlled event. The following will be described in detail.

[0030] Figure 1 is a schematic flow chart of a risk control model modeling method according to an embodiment of the present specification, Figure 1 shows the specific process of risk control model modeling, Figure 2 shows the implementation principle of risk control model modeling. As Figure 1 shown, the method can include:

[0031] S102, obtaining the risk control event feature matrix corresponding to each risk control event sequence in the risk control event sequence set.

[0032] Among them, the risk control event sequence set can include multiple risk control event sequences, and the multiple risk control event sequences are event sequences with interaction information, so that the sequence representation corresponding to the risk control event sequence set extracted through the subsequent steps can be used to construct a risk control model for controlling the risk of events containing the interaction information. For example, the multiple risk control event sequences can be event sequences with transaction interaction information, such as the payment behavior sequence of the transaction payee, the collection behavior sequence of the transaction payee, the collection behavior sequence of the transaction payer, the payment behavior sequence of the transaction payer, etc., so that the extracted sequence representation can be used to construct a risk control model for controlling the risk of events containing the transaction payee and / or the transaction payer.

[0033] Optionally, the risk control event feature matrix can be the vectorized representation of the corresponding risk control event sequence. For example, the risk control event sequence is [x 1 , x 2 , x 3 , x 4 ], then after vectorizing the risk control event sequence, the obtained risk control event feature matrix can be [a 1 , a 2 , a 3 , a 4 ], wherein a 1is a vectorized representation of x 1 is a vectorized representation of x 2 is a vectorized representation of x 2 is a vectorized representation of x 3 is a vectorized representation of x 3 is a vectorized representation of x 4 is a vectorized representation of x 4 is a vectorized representation of x. This embodiment facilitates matrix calculation in subsequent steps by converting the risk control event sequence into a corresponding vectorized representation.

[0034] S104, based on the pre-trained risk control event weight matrix, linearly transforms the risk control event feature matrix respectively to obtain a risk control event query matrix, a risk control event key matrix and a risk control event value matrix.

[0035] Optionally, the step of obtaining the risk control event feature matrix can be implemented by a self-attention model, and the risk control event weight matrix can be obtained by pre-training the self-attention model. The risk control event weight matrix can include a risk control event query weight matrix W q , a risk control event key weight matrix W k and a risk control event value weight matrix W v .

[0036] S106, according to the risk control event key matrix corresponding to the risk control event feature matrix, determine the first shared key matrix corresponding to the risk control event sequence set, the first shared key matrix is based on fusing the risk control event key matrix corresponding to each risk control event sequence.

[0037] S108, for each risk control event sequence, according to the risk control event query matrix, the first shared key matrix and the risk control event value matrix, calculate the risk control event fusion feature matrix corresponding to the risk control event sequence, the risk control event fusion feature matrix is used to describe the sequence representation of the risk control event sequence.

[0038] S110, based on the risk control event fusion feature matrix, construct a risk control model for risk control.

[0039] By adopting the technical solutions of one or more embodiments of the present specification, the key matrix and the query matrix are used to determine the dependency between each element in the event sequence, thereby combining the value matrix to determine the sequence representation corresponding to the event sequence, so that each element in the sequence representation contains the global information of the event sequence. Therefore, by obtaining the risk control event feature matrix corresponding to each risk control event sequence in the risk control event sequence set, based on the pre-trained risk control event weight matrix, the risk control event feature matrix is linearly transformed to obtain the risk control event query matrix, the risk control event key matrix and the risk control event value matrix, thereby fusing the risk control event key matrix corresponding to each risk control event sequence to obtain the first shared key matrix corresponding to the risk control event sequence set. Then, for each risk control event sequence, the risk control event fusion feature matrix corresponding to the risk control event sequence is calculated according to the risk control event query matrix, the first shared key matrix and the risk control event value matrix. The risk control event fusion feature matrix is used to describe the sequence representation of the risk control event sequence, so that based on the risk control event query matrix and the first shared key matrix, the dependency between each two elements in the risk control event sequence set can be determined, so that each element in the finally calculated risk control event fusion feature matrix contains the global information of the event sequence set. Compared with the way of determining the sequence representation of each risk control event sequence respectively, the technical solution can avoid the loss of interaction information between different event sequences in the process of determining the sequence representation, and ensure the accuracy of the extracted sequence representation. Moreover, since the algorithm complexity of the step of calculating the risk control event fusion feature matrix is high, compared with the way of splicing multiple risk control event sequences to determine the risk control event fusion feature matrix based on the spliced risk control event sequence, the technical solution adopts the way of calculating the risk control event fusion feature matrix of each risk control event sequence respectively, so that the sequence length participating in calculation each time is greatly reduced, the algorithm complexity is reduced, and the extraction efficiency of the sequence representation is improved. Further, the risk control model for risk control is constructed based on the risk control event fusion feature matrix. Since the sequence representation used to construct the risk control model can be accurately and efficiently extracted, the construction of the risk control model is accurately and efficiently realized.

[0040] In one embodiment, obtaining the risk control event feature matrix corresponding to each risk control event sequence in the risk control event sequence set (i.e., S102) can be performed by: respectively vectorizing each risk control event sequence in the risk control event sequence set to obtain the risk control event feature matrix corresponding to each risk control event sequence respectively.

[0041] Optionally, the vectorization processing of the risk control event sequence can be to multiply the risk control event sequence by a preset matrix to obtain the risk control event feature matrix corresponding to the risk control event sequence.

[0042] Taking a set of risk control event sequences containing m risk control event sequences as an example, for a risk control event sequence S of length n and dimension d... 1 S 2 ... S m ,like Figure 2 As shown, the risk control event feature matrix A can be obtained by multiplying the preset matrix W by the dot product. 1 A 2 A m Among them, A i =WS i , i∈[1,m].

[0043] Optionally, the step of obtaining the risk control event feature matrix can be implemented by a self-attention model, specifically by the Input Embedding layer in the self-attention model. Alternatively, the step of obtaining the risk control event feature matrix can be implemented by the data processing layer in the risk control model. This specification does not limit the implementation in this way. It should be noted that, in addition to the Input Embedding layer, the self-attention model structure may also include other structural layers for data processing, such as an Attention layer and a feature output layer. This specification does not limit the implementation in this way.

[0044] In the embodiments of this specification, each risk control event sequence in the risk control event sequence set is vectorized, thereby converting the risk control event sequence into a corresponding vectorized representation, which facilitates matrix calculation in subsequent steps.

[0045] In one embodiment, the risk control event weight matrix includes a risk control event query weight matrix, a risk control event key weight matrix, and a risk control event value weight matrix. Based on this, and using the pre-trained risk control event weight matrix, linear transformations are performed on the risk control event feature matrices to obtain the risk control event query matrix, risk control event key matrix, and risk control event value matrix (i.e., S104), which can be executed as follows:

[0046] By multiplying the risk control event feature matrix by the risk control event query weight matrix, we obtain the corresponding risk control event query matrix. By multiplying the risk control event feature matrix by the risk control event key weight matrix, we obtain the corresponding risk control event key matrix. Finally, by multiplying the risk control event feature matrix by the risk control event value weight matrix, we obtain the corresponding risk control event value matrix.

[0047] For example, targeting Figure 2 The risk control event feature matrix A shown below 1 A 2 A m By multiplying the risk control events separately, the weight matrix W is queried. q Risk control event key weight matrix Wk and the risk control event value weight matrix W v , the risk control event query matrix q i , the risk control event key matrix k i and the risk control event value matrix v i , wherein q i =W q A i , k i =W k A i and v i =W v A i , i∈[1, m].

[0048] In one embodiment, the determination of the first shared key matrix corresponding to the risk control event sequence set according to the risk control event key matrix corresponding to the risk control event feature matrix (i.e., S106) can be performed as follows: according to the risk control event key matrix corresponding to each risk control event feature matrix, the average value of each element in the risk control event key matrix is calculated to obtain an average value matrix, and the average value matrix is taken as the first shared key matrix corresponding to the risk control event sequence set.

[0049] wherein the average value of each element in the risk control event key matrix is calculated, that is, for the element at each position in the risk control event key matrix, the elements of the risk control event key matrix at this position in all risk control event sequence sets are summed and then averaged.

[0050] Suppose the first shared key matrix is represented as S, then for the example in , the element at the i-th row and the j-th column position of the first shared key matrix is Figure 2

[0051] In the embodiments of the present specification, by determining the first shared key matrix corresponding to the risk control event sequence set, the effect of fusing the key space of different risk control event sequences is achieved, so that the interaction information between multiple risk control event sequences in the risk control event sequence set is retained, providing a data basis for the subsequent step of calculating the risk control event fusion feature matrix corresponding to the risk control event sequence based on the first shared key matrix. It is a prerequisite condition that each element in the risk control event fusion feature matrix contains global information of the event sequence set, ensuring the accuracy of the extracted sequence representation.

[0052] In one embodiment, as shown in Figure 3 , for each risk control event sequence, the risk control event fusion feature matrix corresponding to the risk control event sequence is calculated according to the risk control event query matrix, the first shared key matrix and the risk control event value matrix (i.e., S108), which can be performed as follows S1082-S1086:

[0053] ​S1082, For each risk control event sequence, the similarity between the risk control event query matrix and the first shared key matrix is ​​calculated to obtain the risk control event attention weight corresponding to the risk control event sequence.

[0054] Optionally, commonly used similarity functions include dot product, concatenation, and perceptron. For... Figure 2 In the example above, if the similarity is calculated by applying a dot product to the risk control event query matrix and the first shared key matrix, then the risk control event attention weights corresponding to the obtained risk control event sequences are:

[0055] S1084 normalizes the attention weights for risk control events.

[0056] Continue Figure 2 The example in, due to The value increases with increasing dimension, therefore it is necessary to... Normalization is performed. Optionally, after obtaining the attention weights for risk control events, [further normalization is applied]. Then, it can be divided by The value of the risk control event attention weight is used to normalize the risk control event attention weight, resulting in the normalized risk control event attention weight. Where i∈[1,m], and d is the dimension of the risk control event sequence.

[0057] S1086, calculate the risk control event fusion feature matrix corresponding to the risk control event sequence based on the normalized risk control event attention weight and risk control event value matrix.

[0058] Continue Figure 2 The example in the text refers to the attention weights for risk control events after normalization. You can first perform the softmax (activation function) operation to obtain... Then, using Multiplying the risk control event value matrix by a dot matrix yields the risk control event fusion feature matrix corresponding to the risk control event sequence. Optionally, if Y... i The feature matrix representing risk control events is then...

[0059] In the embodiments described in this specification, it is equivalent to calculating the corresponding risk control event fusion feature matrix (i.e., sequence representation) for each of the m risk control event sequences of length n. Therefore, after adopting this technical solution, the algorithm complexity for extracting the sequence representation is O(m*n). 2 *d) Compared to the scheme of splicing m sequences and then extracting sequence representations, the algorithm complexity is greatly reduced, which helps to improve the efficiency of sequence representation extraction.

[0060] In an embodiment, the risk control model for risk control is constructed based on the risk control event fusion feature matrix (i.e., S110), which can be implemented as follows: inputting the calculated risk control event fusion feature matrix into the to-be-trained risk control model to train the model parameters in the to-be-trained risk control model, thereby obtaining the trained risk control model.

[0061] Figure 4 is a schematic flowchart of a sequence feature extraction method according to an embodiment of the present specification, as shown in Figure 4 The method can include the following steps:

[0062] S402, obtaining an event feature matrix corresponding to each event sequence in a to-be-processed sequence set.

[0063] The to-be-processed sequence set can include a plurality of event sequences, and the plurality of event sequences are event sequences with interaction information, so that the sequence features of the to-be-processed sequence set extracted through subsequent steps can provide guidance for solving related problems of events containing the interaction information, for example, can be used to construct a risk control model for risk control of events containing the interaction information.

[0064] S404, based on the pre-trained event weight matrix, linearly transforming the event feature matrix to obtain an event query matrix, an event key matrix, and an event value matrix.

[0065] Optionally, the step of obtaining the event feature matrix can be implemented by a self-attention model, and the event weight matrix can be obtained by pre-training the self-attention model. The event weight matrix can include an event query weight matrix W q , an event key weight matrix W k , and an event value weight matrix W v .

[0066] S406, determining a second shared key matrix corresponding to the to-be-processed sequence set according to the event key matrix corresponding to the event feature matrix, the second shared key matrix being obtained by fusing the event key matrix corresponding to each event sequence.

[0067] S408, for each event sequence, calculating an event fusion feature matrix corresponding to the event sequence according to the event query matrix, the second shared key matrix, and the event value matrix, the event fusion feature matrix being used to describe the sequence feature of the event sequence.

[0068] By adopting the technical solutions of one or more embodiments of the present specification, the key matrix and the query matrix are used to determine the dependency between each element in the event sequence, thereby combining the value matrix to determine the sequence representation corresponding to the event sequence, so that each element in the sequence representation contains the global information of the event sequence. Therefore, by obtaining the event feature matrix corresponding to each event sequence in the to-be-processed sequence set, performing linear transformation on the event feature matrix based on the pre-trained event weight matrix, obtaining the event query matrix, the event key matrix and the event value matrix, and then fusing the event key matrix corresponding to each event sequence to obtain the second shared key matrix corresponding to the to-be-processed sequence set, the event fusion feature matrix corresponding to the event sequence is calculated according to the event query matrix, the second shared key matrix and the event value matrix for each event sequence. The event fusion feature matrix is used to describe the sequence representation of the event sequence, so that the dependency between each two elements in the to-be-processed sequence set can be determined based on the event query matrix and the second shared key matrix, so that each element in the finally calculated event fusion feature matrix contains the global information of the to-be-processed sequence set. Compared with the way of determining the sequence representation of each event sequence respectively, the technical solution can avoid the loss of interaction information between different event sequences in the process of determining the sequence representation, and ensures the accuracy of the extracted sequence representation. Moreover, since the algorithm complexity of the step of calculating the event fusion feature matrix is high, compared with the way of splicing multiple event sequences to determine the event fusion feature matrix based on the spliced event sequences, the technical solution adopts the way of calculating the event fusion feature matrix of each event sequence respectively, so that the sequence length participating in the calculation each time is greatly reduced, the algorithm complexity is reduced, and the extraction efficiency of the sequence representation is improved.

[0069] It should be noted that the sequence representation extraction method described in S402-S408 is applicable to various scenarios, such as intelligent question answering scenarios, risk control scenarios, intelligent voice scenarios, etc. The embodiments of the present application do not limit this. For example, Figures 1 to 3 , that is, the application of the sequence representation extraction method provided in the present embodiment in the risk control scenario, in order to avoid repetition, Figure 4 In the embodiments shown, the specific implementation of the related steps is similar to that of the embodiments shown in Figures 1 to 3 .

[0070] In one embodiment, obtaining the event feature matrix corresponding to each event sequence in the to-be-processed sequence set (i.e., S402) can be performed by: performing vectorization processing on each event sequence in the to-be-processed sequence set respectively to obtain the event feature matrix corresponding to each event sequence respectively.

[0071] Optionally, the event sequence is vectorized, which can be multiplying the event sequence by a preset matrix, so as to obtain an event feature matrix corresponding to the event sequence. The step of obtaining the event feature matrix can be implemented by a self-attention model, and specifically can be implemented by an Input Embedding layer in the self-attention model, which is not limited in the embodiments of the present specification. It should be noted that, in addition to the Input Embedding layer, the model structure of the self-attention model can also include other structure layers for data processing, such as an Attention layer, a feature output layer, etc., which are not limited in the embodiments of the present specification.

[0072] In the embodiments of the present specification, each event sequence in the set of to-be-processed sequences is vectorized, so as to convert the event sequence into a corresponding vectorized representation, facilitating matrix calculation in subsequent steps.

[0073] In one embodiment, the event weight matrix includes an event query weight matrix W q , an event key weight matrix W k , and an event value weight matrix W v . Based on the pre-trained event weight matrix, the event feature matrix is linearly transformed to obtain an event query matrix, an event key matrix, and an event value matrix (i.e., S404), which can be implemented as follows: multiplying the event feature matrix by the event query weight matrix to obtain the event query matrix, multiplying the event feature matrix by the event key weight matrix to obtain the event key matrix, and multiplying the event feature matrix by the event value weight matrix to obtain the event value matrix.

[0074] In one embodiment, the second shared key matrix corresponding to the set of to-be-processed sequences is determined according to the event key matrix corresponding to the event feature matrix (i.e., S406), which can be implemented as follows: calculating the average value of each element in the event key matrix according to the event key matrix corresponding to each event feature matrix to obtain an average value matrix, so as to take the average value matrix as the second shared key matrix corresponding to the set of to-be-processed sequences.

[0075] In one embodiment, the average value of each element in the event key matrix is calculated, that is, for each element at a position in the event key matrix, the average value is obtained by summing all elements of the set of to-be-processed sequences at the position and then taking the average.

[0076] In the embodiments of the present specification, the key space of different event sequences is fused by determining the second shared key matrix corresponding to the to-be-processed sequence set, so that the interaction information between multiple event sequences in the to-be-processed sequence set is retained, which provides a data basis for subsequent steps of calculating the event fusion feature matrix corresponding to the event sequence based on the second shared key matrix, is a prerequisite for each element in the event fusion feature matrix to contain global information of the event sequence set, and ensures the accuracy of the extracted sequence representation.

[0077] In one embodiment, for each event sequence, the event fusion feature matrix corresponding to the event sequence is calculated according to the event query matrix, the second shared key matrix and the event value matrix (i.e., S408), which can be performed as steps A1-A3 as follows:

[0078] Step A1, for each event sequence, similarity calculation is performed on the event query matrix and the second shared key matrix to obtain the event attention weight corresponding to the event sequence.

[0079] Optionally, commonly used similarity functions include dot product, concatenation, perception, etc. Any similarity function can be used to perform similarity calculation on the event query matrix and the second shared key matrix.

[0080] Step A2, the event attention weight is normalized.

[0081] Step A3, the event fusion feature matrix corresponding to the event sequence is calculated according to the normalized event attention weight and the event value matrix.

[0082] The specific implementation process of steps A2-A3 is similar to S1084-S1086, which will not be described here. In the embodiments of the present specification, by using the method of calculating the event fusion feature matrix of each event sequence respectively, the sequence length participating in calculation each time is greatly reduced, the algorithm complexity is reduced, and thus the extraction efficiency of the sequence representation is improved.

[0083] Figure 5 is a schematic flowchart of a risk control method according to an embodiment of the present specification, as shown in Figure 5 The method can include:

[0084] S502, obtaining a target event to be controlled.

[0085] Optionally, the target event can be a transaction event, a website access event, or the like, which is prone to risks and can cause loss of personal property after the risk occurs. If the target event is a transaction event, then, Figure 5The risk control method shown can be executed by a transaction platform or an authorized third party, and the target event can be obtained through the transaction platform. If the target event is a website access event, then, Figure 5 The risk control method shown can be executed by a browser, a search engine, or an authorized third party, and the target event can be obtained through the browser or the search engine.

[0086] S504, input the target event into the risk control model to obtain a risk detection result.

[0087] The risk control model is constructed based on a risk control event fusion feature matrix corresponding to each risk control event sequence in the risk control event sequence set. The risk control event fusion feature matrix is obtained by calculating a risk control event query matrix, a first shared key matrix, and a risk control event value matrix corresponding to the risk control event sequence. The first shared key matrix is obtained by fusing a risk control event key matrix corresponding to each risk control event sequence. Figures 1 to 3 The risk control model modeling method is similar, and will not be described here. In this embodiment, the risk control model can be constructed based on a sequence feature corresponding to a risk control event sequence set that has interactive information with the target event. For example, the target event is a transaction event, and assume that the target event is user A paying 50 yuan to user B. The risk control event sequence set that has interactive information with the target event can include a sequence of user A's payment and receipt behavior, a sequence of user B's payment and receipt behavior, a sequence of user A's payment to user C, a sequence of user C's payment and receipt behavior, etc. The same payee or payer is the interactive information between the target event and the risk control event sequence set.

[0088] S506, according to the risk detection result, determine the risk control measures for the target event.

[0089] Alternatively, the risk detection result can include that the target event is a risk event, or that the target event is not a risk event. Correspondingly, for a target event that is a risk event, the risk control measures can be to reject the user's current operation request, send risk prompt information to the user, etc., to effectively prevent the user from implementing this risk event; for a target event that is not a risk event, the risk control measures can be to normally respond to the user's current operation request, to avoid affecting the normal execution of the target event.

[0090] By adopting the technical solutions of one or more embodiments of the present specification, the target event to be controlled is obtained, the target event is input into the risk control model, the risk detection result is obtained, and then the risk control measures for the target event are determined according to the risk detection result, so as to determine whether the target event is a risk event, and then execute the risk control measures accordingly, thereby achieving precise risk control of the target event.

[0091] Figure 6 is a schematic flow chart of a risk control method according to another embodiment of the present specification. In this embodiment, the risk control method is applied to a transaction risk control scenario, for example, a scenario of controlling the risk of a transaction. Wherein, the risk control model is constructed based on the sequence representation corresponding to the sequence of payment and receipt behaviors of company X, and the target event is a payment and receipt event related to company X. As shown in Figure 6 the method can comprise:

[0092] S601, obtaining a payment and receipt event related to company X.

[0093] Optionally, Figure 6 the risk control method shown in can be executed by a transaction platform or an authorized third party, and the payment and receipt event related to company X can be obtained through the transaction platform used when performing the payment and receipt operation. Wherein, the payment and receipt event related to company X includes a payment event related to company X and / or a payment event related to company X.

[0094] S602, inputting the payment and receipt event related to company X into the risk control model to obtain a risk detection result.

[0095] Optionally, the risk detection result can include that one or more payment and receipt events related to company X are risk events, or one or more payment and receipt events related to company X are not risk events.

[0096] S603, determining a risk control measure for the payment and receipt event related to company X according to the risk detection result.

[0097] Optionally, for the payment and receipt event related to company X which is a risk event, the risk control measure can be to reject the current operation request of the user, to send risk prompt information to the user, etc., so as to effectively prevent the user from implementing this risk event; for the payment and receipt event related to company X which is not a risk event, the risk control measure can be to normally respond to the current operation request of the user, so as to avoid affecting the normal execution of the payment and receipt event related to company X.

[0098] By adopting the technical solutions of one or more embodiments of the present specification, by obtaining an event to be controlled, inputting the event into the risk control model to obtain a risk detection result, and determining a risk control measure for the event according to the risk detection result, the effect of determining whether the event to be controlled is a risk event is realized, so as to execute the risk control measure targetedly, which is beneficial to realize accurate risk control.

[0099] In view of the foregoing, particular embodiments of the subject matter have been described. Other embodiments are within the scope of the following claims. In some cases, the actions recited in the claims can be performed in a different order and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown or sequential order in order to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous.

[0100] The risk control model modeling method, the sequence representation extraction method, and the risk control method provided in the above are for one or more embodiments of the present specification. Based on the same idea, one or more embodiments of the present specification also provide a risk control model modeling device, a sequence representation extraction device, and a risk control device.

[0101] Figure 7 is a schematic block diagram of a risk control model modeling device according to an embodiment of the present specification. Please refer to Figure 7 The risk control model modeling device can include:

[0102] The first acquisition module 710 is configured to acquire a risk control event feature matrix corresponding to each risk control event sequence in a risk control event sequence set.

[0103] The first linear transformation processing module 720 is configured to perform linear transformation processing on the risk control event feature matrix based on a pre-trained risk control event weight matrix to obtain a risk control event query matrix, a risk control event key matrix, and a risk control event value matrix.

[0104] The first determination module 730 is configured to determine a first shared key matrix corresponding to the risk control event sequence set according to the risk control event key matrix corresponding to the risk control event feature matrix. The first shared key matrix is obtained by fusing the risk control event key matrix corresponding to each risk control event sequence.

[0105] The first calculation module 740 is configured to calculate, for each risk control event sequence, a risk control event fusion feature matrix corresponding to the risk control event sequence according to the risk control event query matrix, the first shared key matrix, and the risk control event value matrix. The risk control event fusion feature matrix is used to describe the sequence representation of the risk control event sequence.

[0106] The construction module 750 is configured to construct a risk control model for risk control based on the risk control event fusion feature matrix.

[0107] In an embodiment, the first determination module 730 includes:

[0108] The first calculation unit is configured to calculate the average value of each element in the risk control event key matrix according to the risk control event key matrix corresponding to each risk control event feature matrix to obtain an average value matrix.

[0109] The first determination unit is configured to determine the average value matrix as a first shared key matrix corresponding to the set of risk control event sequences.

[0110] In one embodiment, the first calculation module 740 includes:

[0111] The first similarity calculation unit is configured to perform similarity calculation on the risk control event query matrix and the first shared key matrix for each risk control event sequence, to obtain a risk control event attention weight corresponding to the risk control event sequence.

[0112] The first normalization processing unit is configured to perform normalization processing on the risk control event attention weight.

[0113] The second calculation unit is configured to calculate a risk control event fusion feature matrix corresponding to the risk control event sequence according to the normalized risk control event attention weight and the risk control event value matrix.

[0114] In one embodiment, the first acquisition module 710 includes:

[0115] The vectorization processing unit is configured to perform vectorization processing on each risk control event sequence in the set of risk control event sequences respectively, to obtain a risk control event feature matrix corresponding to each risk control event sequence respectively.

[0116] According to the technical scheme of one or more embodiments of the present specification, the key matrix and the query matrix are used to determine the dependency between each element in the event sequence, and the value matrix is combined to determine the sequence representation corresponding to the event sequence, so that each element in the sequence representation contains global information of the event sequence. Therefore, by obtaining the risk control event feature matrix corresponding to each risk control event sequence in the risk control event sequence set, based on the pre-trained risk control event weight matrix, the risk control event feature matrix is linearly transformed to obtain the risk control event query matrix, the risk control event key matrix and the risk control event value matrix, thereby fusing the risk control event key matrix corresponding to each risk control event sequence to obtain the first shared key matrix corresponding to the risk control event sequence set. Then, for each risk control event sequence, the risk control event fusion feature matrix corresponding to the risk control event sequence is calculated according to the risk control event query matrix, the first shared key matrix and the risk control event value matrix. The risk control event fusion feature matrix is used to describe the sequence representation of the risk control event sequence, so that the dependency between each two elements in the risk control event sequence set can be determined based on the risk control event query matrix and the first shared key matrix, thereby ensuring that each element in the finally calculated risk control event fusion feature matrix contains global information of the event sequence set. Compared with the method of determining the sequence representation of each risk control event sequence respectively, the technical scheme can avoid the loss of interaction information between different event sequences in the process of determining the sequence representation, and ensure the accuracy of the extracted sequence representation. Moreover, since the algorithm complexity of the step of calculating the risk control event fusion feature matrix is high, compared with the method of splicing multiple risk control event sequences to determine the risk control event fusion feature matrix based on the spliced risk control event sequence, the technical scheme uses the method of calculating the risk control event fusion feature matrix of each risk control event sequence respectively, so that the sequence length participating in the calculation each time is greatly reduced, the algorithm complexity is reduced, and the extraction efficiency of the sequence representation is improved. Further, the risk control model for risk control is constructed based on the risk control event fusion feature matrix. Since the sequence representation used to construct the risk control model can be accurately and efficiently extracted, the construction of the risk control model is accurately and efficiently realized.

[0117] Those skilled in the art should understand that the above risk control model modeling device can be used to implement the risk control model modeling method described above, and the detailed description should be similar to the description in the method part above. To avoid tediousness, it is not described here.

[0118] Figure 8 is a schematic block diagram of a sequence representation extraction device according to an embodiment of the present specification. Please refer to Figure 8 , the sequence representation extraction device can include:

[0119] The second acquisition module 810 is configured to acquire an event feature matrix corresponding to each event sequence in the sequence set to be processed.

[0120] The second linear transformation processing module 820 is configured to perform linear transformation processing on the event feature matrix based on the pre-trained event weight matrix to obtain an event query matrix, an event key matrix, and an event value matrix.

[0121] The second determination module 830 is configured to determine a second shared key matrix corresponding to the sequence set to be processed according to the event key matrix corresponding to the event feature matrix; the second shared key matrix is obtained by fusing the event key matrix corresponding to each event sequence.

[0122] The second calculation module 840 is configured to calculate, for each event sequence, an event fusion feature matrix corresponding to the event sequence according to the event query matrix, the second shared key matrix, and the event value matrix; the event fusion feature matrix is used to describe sequence representation of the event sequence.

[0123] In an embodiment, the second determination module 830 includes:

[0124] The third calculation unit is configured to calculate, according to the event key matrix corresponding to each event feature matrix, an average value of each element in the event key matrix to obtain an average value matrix.

[0125] The second determination unit is configured to take the average value matrix as the second shared key matrix corresponding to the sequence set to be processed.

[0126] In an embodiment, the second calculation module 840 includes:

[0127] The second similarity calculation unit is configured to perform similarity calculation on the event query matrix and the second shared key matrix for each event sequence to obtain an event attention weight corresponding to the event sequence.

[0128] The second normalization processing unit is configured to perform normalization processing on the event attention weight.

[0129] The fourth calculation unit is configured to calculate, according to the event attention weight after normalization processing and the event value matrix, the event fusion feature matrix corresponding to the event sequence.

[0130] According to the technical scheme of one or more embodiments of the present specification, the key matrix and the query matrix are used to determine the dependency between each element in the event sequence, and the value matrix is combined to determine the sequence representation corresponding to the event sequence, so that each element in the sequence representation contains global information of the event sequence. Therefore, by obtaining the event feature matrix corresponding to each event sequence in the to-be-processed sequence set, performing linear transformation on the event feature matrix based on the pre-trained event weight matrix, obtaining the event query matrix, the event key matrix and the event value matrix, and then fusing the event key matrix corresponding to each event sequence to obtain a second shared key matrix corresponding to the to-be-processed sequence set, for each event sequence, the event fusion feature matrix corresponding to the event sequence is calculated according to the event query matrix, the second shared key matrix and the event value matrix. The event fusion feature matrix is used to describe the sequence representation of the event sequence, so that the dependency between each two elements in the to-be-processed sequence set can be determined based on the event query matrix and the second shared key matrix, so that each element in the finally calculated event fusion feature matrix contains global information of the to-be-processed sequence set. Compared with the method of determining the sequence representation of each event sequence respectively, the technical scheme can avoid the loss of interaction information between different event sequences in the process of determining the sequence representation, and ensures the accuracy of the extracted sequence representation. Moreover, since the algorithm complexity of the step of calculating the event fusion feature matrix is high, compared with the method of splicing multiple event sequences to determine the event fusion feature matrix based on the spliced event sequence, the technical scheme uses the method of calculating the event fusion feature matrix of each event sequence respectively, so that the sequence length participating in the calculation each time is greatly reduced, the algorithm complexity is reduced, and the extraction efficiency of the sequence representation is improved.

[0131] Those skilled in the art should understand that the above sequence representation extraction device can be used to implement the sequence representation extraction method described above, and the detailed description thereof is similar to the method part described above. To avoid tediousness, it will not be described here.

[0132] Figure 9 is a schematic block diagram of a risk control device according to an embodiment of the present specification. Please refer to Figure 9 The risk control device can include:

[0133] The third acquisition module 910 is configured to acquire a target event to be subjected to risk control.

[0134] The model processing module 920 is configured to input the target event into a risk control model to obtain a risk detection result.

[0135] The third determination module 930 is configured to determine a risk control measure for the target event according to the risk detection result.

[0136] The risk control model is constructed based on a risk control event fusion feature matrix corresponding to each risk control event sequence in the set of risk control event sequences, and the risk control event fusion feature matrix is obtained by calculating a risk control event query matrix, a first shared key matrix, and a risk control event value matrix corresponding to the risk control event sequence.

[0137] The technical solutions of one or more embodiments of the present specification are adopted, the target event to be controlled is obtained, the target event is input into the risk control model, and a risk detection result is obtained, so as to determine the risk control measures for the target event according to the risk detection result, thereby realizing the effect of determining whether the target event is a risk event, and then executing the risk control measures in a targeted manner, which is beneficial to realize the accurate risk control of the target event.

[0138] Those skilled in the art should understand that the above risk control device can be used to realize the risk control method described above, and the detailed description should be similar to the method part described above. To avoid tediousness, it is not described here.

[0139] Based on the same idea, one or more embodiments of the present specification also provide an electronic device, as shown in Figure 10 The electronic device can have a large difference due to different configurations or performances, and can include one or more processors 1001 and memories 1002, and one or more storage applications or data can be stored in the memories 1002. The memory 1002 can be temporary storage or persistent storage. The application stored in the memory 1002 can include one or more modules (not shown in the figure), and each module can include a series of computer executable instructions in the electronic device. Further, the processor 1001 can be configured to communicate with the memory 1002 and execute a series of computer executable instructions in the memory 1002 on the electronic device. The electronic device can also include one or more power supplies 1003, one or more wired or wireless network interfaces 1004, one or more input and output interfaces 1005, and one or more keyboards 1006.

[0140] In particular, the electronic device includes a memory and one or more programs, wherein one or more programs are stored in the memory, and one or more programs can include one or more modules, and each module can include a series of computer executable instructions in the electronic device, and the one or more processors are configured to execute the one or more programs including the following computer executable instructions:

[0141] obtain an anti-risk event feature matrix corresponding to each anti-risk event sequence in the anti-risk event sequence set;

[0142] Based on the anti-risk event weight matrix obtained by pre-training, linear transformation is performed on the anti-risk event feature matrix respectively to obtain an anti-risk event query matrix, an anti-risk event key matrix and an anti-risk event value matrix;

[0143] According to the anti-risk event key matrix corresponding to the anti-risk event feature matrix, a first shared key matrix corresponding to the anti-risk event sequence set is determined; the first shared key matrix is obtained based on fusing the anti-risk event key matrix corresponding to each anti-risk event sequence;

[0144] For each anti-risk event sequence, according to the anti-risk event query matrix, the first shared key matrix and the anti-risk event value matrix, an anti-risk event fusion feature matrix corresponding to the anti-risk event sequence is calculated, and the anti-risk event fusion feature matrix is used to describe the sequence representation of the anti-risk event sequence;

[0145] Based on the anti-risk event fusion feature matrix, an anti-risk model for risk control is constructed.

[0146] In addition, in this embodiment, the electronic device includes a memory and one or more programs, wherein one or more programs are stored in the memory, and one or more programs can include one or more modules, and each module can include a series of computer executable instructions in the electronic device, and is configured to be executed by one or more processors The one or more programs include computer executable instructions for:

[0147] obtain an event feature matrix corresponding to each event sequence in the to-be-processed sequence set;

[0148] Based on the event weight matrix obtained by pre-training, linear transformation is performed on the event feature matrix respectively to obtain an event query matrix, an event key matrix and an event value matrix;

[0149] According to the event key matrix corresponding to the event feature matrix, a second shared key matrix corresponding to the to-be-processed sequence set is determined; the second shared key matrix is obtained based on fusing the event key matrix corresponding to each event sequence;

[0150] For each event sequence, according to the event query matrix, the second shared key matrix and the event value matrix, an event fusion feature matrix corresponding to the event sequence is calculated, and the event fusion feature matrix is used to describe the sequence representation of the event sequence.

[0151] In addition, in particular in the present embodiment, the electronic device comprises a memory, and one or more programs, wherein the one or more programs are stored in the memory, and the one or more programs can comprise one or more modules, and each module can comprise a series of computer executable instructions in the electronic device, and the one or more programs configured to be executed by the one or more processors include the following computer executable instructions:

[0152] Obtaining a target event to be subjected to risk control;

[0153] Inputting the target event into the risk control model to obtain a risk detection result;

[0154] According to the risk detection result, determining a risk control measure for the target event;

[0155] The risk control model is constructed based on a risk control event fusion feature matrix corresponding to each risk control event sequence in the risk control event sequence set, and the risk control event fusion feature matrix is obtained by calculating a risk control event query matrix, a first shared key matrix and a risk control event value matrix corresponding to the risk control event sequence, and the first shared key matrix is obtained by fusing a risk control event key matrix corresponding to each risk control event sequence.

[0156] The one or more embodiments of the present specification also propose a storage medium storing one or more computer programs, the one or more computer programs comprising instructions capable of causing an electronic device comprising a plurality of application programs to execute the processes of the above-mentioned risk control model modeling method embodiments when the instructions are executed by the electronic device, and specifically for executing:

[0157] Obtaining a risk control event feature matrix corresponding to each risk control event sequence in the risk control event sequence set;

[0158] Based on the pre-trained risk control event weight matrix, linear transformation processing is performed on the risk control event feature matrix respectively to obtain a risk control event query matrix, a risk control event key matrix and a risk control event value matrix;

[0159] According to the risk control event key matrix corresponding to the risk control event feature matrix, a first shared key matrix corresponding to the risk control event sequence set is determined; the first shared key matrix is obtained by fusing the risk control event key matrix corresponding to each risk control event sequence;

[0160] For each risk control event sequence, according to the risk control event query matrix, the first shared key matrix and the risk control event value matrix, a risk control event fusion feature matrix corresponding to the risk control event sequence is calculated, and the risk control event fusion feature matrix is used to describe the sequence representation of the risk control event sequence;

[0161] The risk control model for risk control is constructed based on a risk control event fusion feature matrix.

[0162] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments. In particular, for the above-mentioned one method embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant part can be referred to the part of the method embodiment.

[0163] In another embodiment of the specification, the storage medium can further store one or more computer programs, which include instructions that, when executed by an electronic device comprising a plurality of application programs, can enable the electronic device to perform the processes of the above-mentioned sequence feature extraction method embodiment, and specifically for performing:

[0164] Obtaining an event feature matrix corresponding to each event sequence in the to-be-processed sequence set;

[0165] Based on the pre-trained event weight matrix, linear transformation is performed on the event feature matrix respectively to obtain an event query matrix, an event key matrix, and an event value matrix;

[0166] According to the event key matrix corresponding to the event feature matrix, a second shared key matrix corresponding to the to-be-processed sequence set is determined; the second shared key matrix is obtained based on fusion of the event key matrix corresponding to each event sequence.

[0167] For each event sequence, according to the event query matrix, the second shared key matrix, and the event value matrix, an event fusion feature matrix corresponding to the event sequence is calculated, and the event fusion feature matrix is used to describe the sequence feature of the event sequence.

[0168] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments. In particular, for the above-mentioned one method embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant part can be referred to the part of the method embodiment.

[0169] In another embodiment of the specification, the storage medium can further store one or more computer programs, which include instructions that, when executed by an electronic device comprising a plurality of application programs, can enable the electronic device to perform the processes of the above-mentioned risk control method embodiment, and specifically for performing:

[0170] Obtaining a target event to be controlled;

[0171] Input the target event into the risk control model to obtain a risk detection result;

[0172] According to the risk detection result, determine a risk control measure for the target event;

[0173] The risk control model is constructed based on a risk control event fusion feature matrix corresponding to each risk control event sequence in the set of risk control event sequences, and the risk control event fusion feature matrix is obtained by calculating a risk control event query matrix, a first shared key matrix, and a risk control event value matrix corresponding to the risk control event sequence, and the first shared key matrix is obtained by fusing a risk control event key matrix corresponding to each risk control event sequence.

[0174] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments. In particular, for the above-mentioned storage medium embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the related parts can be referred to the part of the method embodiment.

[0175] The system, device, module or unit described in the above embodiments can be specifically implemented by a computer chip or entity, or by a product with certain functions. A typical implementation device is a computer. Specifically, the computer may, for example, be a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0176] For the convenience of description, the above device is described as various units divided by functions. Of course, the functions of each unit can be implemented in the same or more software and / or hardware when implementing one or more embodiments of the specification.

[0177] Those skilled in the art should understand that one or more embodiments of the specification can be provided as a method, a system, or a computer program product. Therefore, one or more embodiments of the specification can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, one or more embodiments of the specification can be in the form of a computer program product implemented on one or more computer usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0178] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks.

[0179] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks.

[0180] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 Figure 1 The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks.

[0181] In one typical configuration, the computing device includes one or more processors (CPU's), input / output interfaces, network interfaces, and memory.

[0182] The memory can include non-persistent memory and / or persistent memory, such as flash memory, read-only memory (ROM), and / or volatile or non-volatile random access memory (RAM), among others. The memory is an example of computer-readable media.

[0183] Computer-readable media includes permanent and non-permanent, movable and non-movable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer-readable media does not include transitory media such as modulated data signals and carriers.

[0184] It should also be noted that the terms "comprising", "containing", or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, method, article or apparatus that comprises a list of elements does not only include those elements, but also includes other elements not explicitly listed or inherent to such a process, method, article or apparatus. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.

[0185] One or more embodiments of the present specification can be described in the general context of computer-executable instructions being executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform particular tasks or implement particular abstract data types. The present specification can also be practiced in a distributed computing environment, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0186] Each embodiment in the present specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other, and each embodiment focuses on the difference from other embodiments. In particular, for system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiment.

[0187] The above merely provides one or more embodiments of the present specification, and is not intended to limit the present application. One of ordinary skill in the art can make various modifications and changes to the one or more embodiments of the present specification. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the one or more embodiments of the present specification shall be included in the scope of the claims of the one or more embodiments of the present specification.

Claims

1. A risk control model modeling method, comprising: obtaining, by using a data processing layer of a risk control model, a risk control event feature matrix corresponding to each risk control event sequence in a risk control event sequence set in which transaction interaction information exists, the plurality of risk control event sequences including a payment behavior sequence of a transaction payee, a payment behavior sequence of a transaction payee, a payment behavior sequence of a transaction payee, and a payment behavior sequence of a transaction payee; performing linear transformation processing on the risk control event feature matrix based on a pre-trained risk control event weight matrix to obtain a risk control event query matrix, a risk control event key matrix, and a risk control event value matrix; determining a first shared key matrix corresponding to the risk control event sequence set according to the risk control event key matrix corresponding to the risk control event feature matrix, the first shared key matrix being obtained based on fusing the risk control event key matrix corresponding to each risk control event sequence; for each risk control event sequence, calculating a risk control event fusion feature matrix corresponding to the risk control event sequence according to the risk control event query matrix, the first shared key matrix, and the risk control event value matrix, the risk control event fusion feature matrix being used to describe sequence representation of the risk control event sequence; constructing a risk control model for risk control of transaction type events containing transaction interaction information based on the risk control event fusion feature matrix; wherein the constructing a risk control model for risk control of transaction type events containing transaction interaction information based on the risk control event fusion feature matrix comprises: inputting the risk control event fusion feature matrix into a to-be-trained risk control model to train model parameters in the to-be-trained risk control model, thereby obtaining a trained risk control model; wherein the determining a first shared key matrix corresponding to the risk control event sequence set according to the risk control event key matrix corresponding to the risk control event feature matrix comprises: calculating an average value matrix by calculating an average value of each element in the risk control event key matrix according to the risk control event key matrix corresponding to each risk control event feature matrix; taking the average value matrix as the first shared key matrix corresponding to the risk control event sequence set.

2. The method of claim 1, wherein the calculating a risk control event fusion feature matrix corresponding to each risk control event sequence according to the risk control event query matrix, the first shared key matrix, and the risk control event value matrix comprises: performing similarity calculation on the risk control event query matrix and the first shared key matrix for each risk control event sequence to obtain a risk control event attention weight corresponding to the risk control event sequence; performing normalization processing on the risk control event attention weight; calculating a risk control event fusion feature matrix corresponding to the risk control event sequence according to the normalized risk control event attention weight and the risk control event value matrix.

3. The method of claim 1, wherein the obtaining a risk control event feature matrix corresponding to each risk control event sequence in a risk control event sequence set in which transaction interaction information exists comprises: vectorize each risk control event sequence with transaction interaction information in the risk control event sequence set respectively to obtain the risk control event feature matrix corresponding to each risk control event sequence respectively.

4. A sequence representation extraction method, comprising: obtaining an event feature matrix corresponding to each event sequence with transaction interaction information in a to-be-processed sequence set by using a data processing layer of a risk control model; performing linear transformation processing on the event feature matrix based on a pre-trained event weight matrix to obtain an event query matrix, an event key matrix, and an event value matrix; determining a second shared key matrix corresponding to the to-be-processed sequence set according to the event key matrix corresponding to the event feature matrix; the second shared key matrix is obtained based on fusion of the event key matrix corresponding to each event sequence; for each event sequence, calculating an event fusion feature matrix corresponding to the event sequence according to the event query matrix, the second shared key matrix, and the event value matrix, the event fusion feature matrix being used to describe the sequence representation of the event sequence, and the event fusion feature matrix being used to construct a risk control model for risk control of transaction events containing transaction interaction information; wherein, the second shared key matrix corresponding to the to-be-processed sequence set is determined according to the event key matrix corresponding to the event feature matrix, comprising: calculating the average value of each element in the event key matrix according to the event key matrix corresponding to each event feature matrix to obtain an average value matrix; the average value matrix is used as the second shared key matrix corresponding to the to-be-processed sequence set.

5. The method of claim 4, wherein the event fusion feature matrix corresponding to each event sequence is calculated according to the event query matrix, the second shared key matrix, and the event value matrix, comprising: for each event sequence, performing similarity calculation on the event query matrix and the second shared key matrix to obtain an event attention weight corresponding to the event sequence; normalizing the event attention weight; calculating the event fusion feature matrix corresponding to the event sequence according to the normalized event attention weight and the event value matrix.

6. A risk control method, comprising: obtaining a target event to be subjected to risk control; inputting the target event into a risk control model for risk control of transaction events containing transaction interaction information to obtain a risk detection result; determining a risk control measure for the target event according to the risk detection result; The risk control model is constructed based on a risk control event fusion feature matrix corresponding to each risk control event sequence with transaction interaction information in the risk control event sequence set, the risk control event fusion feature matrix is obtained by calculating a risk control event query matrix, a first shared key matrix and a risk control event value matrix corresponding to the risk control event sequence, the first shared key matrix is obtained based on fusing a risk control event key matrix corresponding to each risk control event sequence, the plurality of risk control event sequences include a payment behavior sequence of a transaction payee, a payment behavior sequence of a transaction payee, a payment behavior sequence of a transaction payer, and a payment behavior sequence of a transaction payer, the risk control model is obtained by inputting the risk control event fusion feature matrix into a to-be-trained risk control model to train model parameters in the to-be-trained risk control model, the first shared key matrix corresponding to the risk control event sequence set is determined according to an average value matrix, and the average value matrix is determined according to an average value of each element in the risk control event key matrix corresponding to each risk control event feature matrix.

7. A risk control model modeling device, comprising: a first acquisition module configured to acquire, by using a data processing layer of a risk control model, a risk control event feature matrix corresponding to each risk control event sequence with transaction interaction information in a risk control event sequence set; a first linear transformation processing module configured to perform linear transformation processing on the risk control event feature matrix based on a pre-trained risk control event weight matrix to obtain a risk control event query matrix, a risk control event key matrix and a risk control event value matrix, wherein the plurality of risk control event sequences include a payment behavior sequence of a transaction payee, a payment behavior sequence of a transaction payee, a payment behavior sequence of a transaction payer, and a payment behavior sequence of a transaction payer; a first determination module configured to determine a first shared key matrix corresponding to the risk control event sequence set according to a risk control event key matrix corresponding to the risk control event feature matrix, wherein the first shared key matrix is obtained based on fusing a risk control event key matrix corresponding to each risk control event sequence; a first calculation module configured to calculate, for each risk control event sequence, a risk control event fusion feature matrix corresponding to the risk control event sequence according to the risk control event query matrix, the first shared key matrix and the risk control event value matrix, wherein the risk control event fusion feature matrix is used to describe sequence characteristics of the risk control event sequence; a construction module configured to construct a risk control model for risk control of a transaction event containing transaction interaction information based on the risk control event fusion feature matrix; The construction module is configured to: input the risk control event fusion feature matrix into a to-be-trained risk control model to train model parameters in the to-be-trained risk control model, thereby obtaining a trained risk control model; The first determination module is configured to: calculate an average value matrix by calculating an average value of each element in the risk control event key matrix corresponding to each risk control event feature matrix; use the average value matrix as the first shared key matrix corresponding to the risk control event sequence set.

8. A sequence representation extraction apparatus, comprising: a second obtaining module configured to obtain, by using a data processing layer of a risk control model, an event feature matrix corresponding to each event sequence with transaction interaction information in a sequence set to be processed; a second linear transformation processing module configured to perform linear transformation processing on the event feature matrix respectively based on a pre-trained event weight matrix to obtain an event query matrix, an event key matrix, and an event value matrix; a second determining module configured to determine a second shared key matrix corresponding to the sequence set to be processed according to the event key matrix corresponding to the event feature matrix; the second shared key matrix is obtained based on fusion of the event key matrix corresponding to each event sequence; a second calculating module configured to calculate, for each event sequence, an event fusion feature matrix corresponding to the event sequence according to the event query matrix, the second shared key matrix, and the event value matrix, the event fusion feature matrix being used to describe sequence representation of the event sequence, and the event fusion feature matrix being used to construct the risk control model for risk control of transaction events containing transaction interaction information; wherein the second determining module is configured to: calculate an average value matrix by calculating an average value of each element in the event key matrix according to the event key matrix corresponding to each event feature matrix; and use the average value matrix as the second shared key matrix corresponding to the sequence set to be processed.

9. A risk control apparatus, comprising: a third obtaining module configured to obtain a target event to be subjected to risk control; a model processing module configured to input the target event into a risk control model for risk control of transaction events containing transaction interaction information to obtain a risk detection result; a third determining module configured to determine a risk control measure for the target event according to the risk detection result; wherein the risk control model is constructed based on a risk control event fusion feature matrix corresponding to each risk control event sequence with transaction interaction information in a risk control event sequence set, the risk control event fusion feature matrix being obtained by calculating a risk control event query matrix, a first shared key matrix, and a risk control event value matrix corresponding to the risk control event sequence, the first shared key matrix being obtained based on fusion of a risk control event key matrix corresponding to each risk control event sequence, the plurality of risk control event sequences including a transaction payor payment behavior sequence, a transaction payor collection behavior sequence, a transaction payee collection behavior sequence, and a transaction payee payment behavior sequence, the risk control model being obtained by inputting the risk control event fusion feature matrix into a risk control model to be trained to train model parameters in the risk control model to be trained, and the first shared key matrix corresponding to the risk control event sequence set being determined according to an average value matrix, the average value matrix being determined according to an average value of each element in the risk control event key matrix calculated according to the risk control event key matrix corresponding to each risk control event feature matrix.

10. An electronic device, comprising: a processor; and a memory arranged to store computer executable instructions that, when executed, can cause the processor to: ​ The data processing layer of the risk control model is used to obtain a risk control event feature matrix corresponding to each risk control event sequence with transaction interaction information in the risk control event sequence set, and the plurality of risk control event sequences include a payment behavior sequence of a transaction payee, a payment behavior sequence of a transaction payee, a payment behavior sequence of a transaction payee, and a payment behavior sequence of a transaction payee. Based on the pre-trained risk control event weight matrix, linear transformation is performed on the risk control event feature matrix respectively to obtain a risk control event query matrix, a risk control event key matrix, and a risk control event value matrix. According to the risk control event key matrix corresponding to the risk control event feature matrix, a first shared key matrix corresponding to the risk control event sequence set is determined, and the first shared key matrix is obtained based on fusion of the risk control event key matrix corresponding to each risk control event sequence. For each risk control event sequence, the risk control event query matrix, the first shared key matrix, and the risk control event value matrix are used to calculate a risk control event fusion feature matrix corresponding to the risk control event sequence, and the risk control event fusion feature matrix is used to describe the sequence representation of the risk control event sequence. Based on the risk control event fusion feature matrix, a risk control model for risk control of transaction events containing transaction interaction information is constructed. The risk control model for risk control of transaction events containing transaction interaction information is constructed based on the risk control event fusion feature matrix, including: The risk control event fusion feature matrix is input into a to-be-trained risk control model to train the model parameters in the to-be-trained risk control model, so as to obtain a trained risk control model. The first shared key matrix corresponding to the risk control event sequence set is determined according to the risk control event key matrix corresponding to the risk control event feature matrix, including: According to the risk control event key matrix corresponding to each risk control event feature matrix, the average value of each element in the risk control event key matrix is calculated to obtain an average value matrix. The average value matrix is used as the first shared key matrix corresponding to the risk control event sequence set.

11. An electronic device comprising: a processor; and a memory arranged to store computer executable instructions that, when executed, can cause the processor to: obtain, using a data processing layer of a risk control model, an event feature matrix corresponding to each event sequence with transaction interaction information in a to-be-processed sequence set; perform linear transformation on the event feature matrix based on a pre-trained event weight matrix to obtain an event query matrix, an event key matrix, and an event value matrix; determine a second shared key matrix corresponding to the to-be-processed sequence set according to the event key matrix corresponding to the event feature matrix; the second shared key matrix is obtained based on fusion of the event key matrix corresponding to each event sequence; and ​ For each event sequence, an event fusion feature matrix corresponding to the event sequence is calculated according to the event query matrix, the second shared key matrix and the event value matrix, the event fusion feature matrix being used to describe sequence representation of the event sequence, and the event fusion feature matrix being used to construct a risk control model for risk control of transaction events containing transaction interaction information. The second shared key matrix corresponding to the to-be-processed sequence set is determined according to the event key matrix corresponding to the event feature matrix, and includes: An average value matrix is obtained by calculating an average value of each element in the event key matrix according to the event key matrix corresponding to each event feature matrix. The average value matrix is taken as the second shared key matrix corresponding to the to-be-processed sequence set.

12. An electronic device, comprising: a processor; and a memory arranged to store computer executable instructions that, when executed, can cause the processor to: obtain a target event to be subjected to risk control; input the target event into a risk control model for risk control of transaction events containing transaction interaction information to obtain a risk detection result; determine a risk control measure for the target event according to the risk detection result; The risk control model is constructed based on a risk control event fusion feature matrix corresponding to each risk control event sequence in a risk control event sequence set, the risk control event fusion feature matrix being obtained by calculating a risk control event query matrix, a first shared key matrix and a risk control event value matrix corresponding to the risk control event sequence, the first shared key matrix being obtained by fusing risk control event key matrices corresponding to each risk control event sequence, the plurality of risk control event sequences including a payment behavior sequence of a transaction payee, a collection behavior sequence of a transaction payee, a collection behavior sequence of a transaction payer, and a payment behavior sequence of a transaction payer, the risk control model being obtained by inputting the risk control event fusion feature matrix into a to-be-trained risk control model to train model parameters in the to-be-trained risk control model, and the first shared key matrix corresponding to the risk control event sequence set being determined according to an average value matrix, the average value matrix being determined by calculating an average value of each element in the risk control event key matrix corresponding to each risk control event feature matrix. ​

Images