Merchant grouping method and device, storage medium and electronic equipment

Abstract

The embodiment of the invention relates to the technical field of computers, and discloses a method and device for grouping merchants, a storage medium and electronic equipment, and the method comprises the steps: obtaining behavior sequence information corresponding to the merchants; obtaining time interval sequence information corresponding to the merchant according to a behavior time interval corresponding to every two adjacent historical behaviors in the behavior sequence information; respectively inputting the behavior sequence information and the time interval sequence information into a first embedding layer and a second embedding layer in a merchant representation information extraction model to obtain a corresponding first embedding vector and a corresponding second embedding vector; inputting the first embedded vector and the second embedded vector into a cross attention layer in the commercial tenant representation information extraction model for unsupervised learning, and obtaining a representation vector corresponding to the commercial tenant output by the cross attention layer in the converged commercial tenant representation information extraction model; and grouping the plurality of merchants according to the representation vectors corresponding to the plurality of merchants to obtain a plurality of merchant groups.

Description

Technical Field

[0001] The present invention relates to computer technology, and in particular to a method, device, storage medium and electronic device for merchant grouping. Background Art

[0002] Traditional risk decision-making usually relies heavily on historical cases and expert opinions, and after the punishment is issued, a large amount of subsequent manual operation and maintenance costs are required, resulting in the current situation where the decision-making system is huge, the operation and maintenance flexibility is low, and the manpower and time consumption is high. The answer to this situation may be to establish an automated risk decision-making system, and the premise of the establishment is to be able to first use historical information to reasonably group merchants, assign different sets of control actions to different groups, and select the most optimized decision from the control candidate sets of different groups. Therefore, the most optimized decision often depends first on the effect of merchant grouping, and the grouping that cannot well characterize the merchant behavior pattern will lead to failed decisions. Summary of the invention

[0003] The purpose of the embodiments of this specification is to provide a method, device, storage medium and electronic device for merchant grouping.

[0004] The embodiments of this specification provide a method for merchant clustering, which proposes a method for pre-unsupervised merchant clustering based on sequence similarity in a decision system, uses the merchant's historical sequence information to characterize the merchant's behavior pattern, improves the merchant's clustering effect, and further improves the decision-making ability of the decision-making system. By extracting the merchant's historical behavior sequence information and time series information to learn the cross-attention mechanism, the merchants are clustered by this unsupervised learning method, and the historical data can be effectively used to learn the merchant's behavior pattern to cluster the merchants to adapt to changes at different times, improve the clustering effect, and improve the decision-making effect of the subsequent decision-making system. The unsupervised learning using the merchant's historical behavior sequence does not rely on expert experience or label data. Instead, it mines patterns and structures directly from the data itself. Unsupervised learning can automatically discover natural clusters of merchants from a large number of behavior sequences. These clusters are based on the intrinsic similarities of the data rather than relying on experts' possibly outdated or rough judgments. It can capture the dynamic changes of merchant behavior in real time and discover subtle patterns that expert experience fails to cover, thereby providing a more sophisticated clustering effect. Compared with supervised learning, unsupervised learning is not restricted by existing labels and can adapt to the evolution of merchant behavior patterns. By learning the potential structure in the data rather than relying on fixed labels to build, it can better generalize to new and unseen data, thereby maintaining a high prediction accuracy when merchant behavior patterns change. The method includes: Obtaining behavior sequence information corresponding to a merchant, wherein the behavior sequence information includes multiple historical behaviors of the merchant; Obtaining time interval sequence information corresponding to the merchant according to the behavior time intervals corresponding to every two adjacent historical behaviors in the behavior sequence information; Inputting the behavior sequence information and the time interval sequence information into the first embedding layer and the second embedding layer of the merchant representation information extraction model respectively to obtain the corresponding first embedding vector and the second embedding vector; Inputting the first embedding vector and the second embedding vector into the cross-attention layer in the merchant representation information extraction model for unsupervised learning, and obtaining a representation vector corresponding to the merchant output by the cross-attention layer in the merchant representation information extraction model after convergence, wherein the learning goal of the unsupervised learning is to predict the next behavior through the preceding sequence in the behavior sequence information; The multiple merchants are grouped according to the representation vectors respectively corresponding to the multiple merchants to obtain multiple merchant groups.

[0005] Furthermore, obtaining the behavior sequence information corresponding to the merchant includes: The multiple historical behaviors of the merchant are sorted according to the corresponding behavior occurrence time to obtain the behavior sequence information corresponding to the merchant.

[0006] Furthermore, the ordering of the multiple historical behaviors of the merchant according to the corresponding behavior occurrence time to obtain the behavior sequence information corresponding to the merchant includes: Sorting the multiple historical behaviors of the merchant according to the corresponding behavior occurrence time to obtain the sorted multiple historical behaviors; For each of the sorted multiple historical behaviors, determining behavior coding information of the historical behavior according to the behavior type corresponding to the historical behavior; Behavior sequence information corresponding to the merchant is obtained, wherein the behavior sequence includes a plurality of sorted behavior coding information.

[0007] Further, the obtaining the time interval sequence information corresponding to the merchant according to the behavior time intervals corresponding to every two adjacent historical behaviors in the behavior sequence information includes: According to the behavior time interval corresponding to each two adjacent historical behaviors in the behavior sequence information, by unitizing the behavior time interval, a time interval unit corresponding to the behavior time interval is obtained; According to the time interval unit, the time interval sequence information corresponding to the merchant is obtained.

[0008] Furthermore, the loss function corresponding to the learning objective of the unsupervised learning is the cross entropy loss between the next behavior predicted by the preceding sequence in the behavior sequence information and the next historical behavior in the behavior sequence information that is located after the preceding sequence.

[0009] Furthermore, the loss function corresponding to the learning objective of the unsupervised learning is the sum of the cross entropy losses corresponding to each historical behavior from the second historical behavior in the behavior sequence information to the next historical behavior after the preceding sequence in the behavior sequence information.

[0010] Furthermore, the obtaining of the representation vector corresponding to the merchant output by the cross attention layer in the converged merchant representation information extraction model includes: Obtaining a target vector corresponding to the merchant output by a cross-attention layer in a converged merchant representation information extraction model; The target vector is input into the fully connected layer in the merchant representation information extraction model to obtain a representation vector of a preset size.

[0011] Furthermore, the grouping of the multiple merchants according to the representation vectors respectively corresponding to the multiple merchants to obtain multiple merchant groups includes: Randomly selecting a preset number of multiple target merchants from the multiple merchants, and using the representation vectors of the multiple target merchants as initial cluster centers; For each merchant among the multiple merchants, according to the distance between the merchant's characterization vector and each cluster center, the merchant is assigned to the cluster to which the nearest cluster center belongs; Recalculate the average value of the representation vectors of all merchants in each cluster and use the average value as the new cluster center corresponding to the cluster; For each merchant among the multiple merchants, the merchant is reallocated to the cluster to which the nearest cluster center belongs, and so on, until a preset iteration end condition is met.

[0012] Furthermore, the preset iteration end condition includes at least one of the following: The change of cluster center meets the preset conditions; The number of iterations has reached the preset threshold.

[0013] Furthermore, the method further comprises: The preset quantity is determined according to the preset merchant management and control granularity.

[0014] The embodiment of this specification also provides a device for merchant grouping, including: A behavior sequence acquisition module, used to obtain behavior sequence information corresponding to a merchant, wherein the behavior sequence information includes a plurality of historical behaviors of the merchant; A time interval sequence acquisition module, used to obtain the time interval sequence information corresponding to the merchant according to the behavior time intervals corresponding to every two adjacent historical behaviors in the behavior sequence information; An embedding vector acquisition module, used to input the behavior sequence information and the time interval sequence information into the first embedding layer and the second embedding layer of the merchant representation information extraction model respectively, to obtain the corresponding first embedding vector and second embedding vector; an unsupervised learning module, configured to input the first embedding vector and the second embedding vector into the cross-attention layer in the merchant representation information extraction model for unsupervised learning, and obtain a representation vector corresponding to the merchant output by the cross-attention layer in the merchant representation information extraction model after convergence, wherein the learning goal of the unsupervised learning is to predict the next behavior through the preceding sequence in the behavior sequence information; The merchant grouping module is used to group the multiple merchants according to the representation vectors respectively corresponding to the multiple merchants to obtain multiple merchant groups.

[0015] The embodiments of the present specification also provide a storage medium, wherein the storage medium stores a computer program, and the computer program is suitable for being loaded by a processor and executing the steps of the above method.

[0016] An embodiment of the present specification also provides an electronic device, comprising: a processor and a memory; wherein the memory stores a computer program, and the computer program is suitable for being loaded by the processor and executing the steps of the above method.

[0017] In the embodiments of this specification, a method for clustering merchants in a decision system based on sequence similarity is proposed, and the merchant's historical sequence information is used to characterize the merchant's behavior pattern, so as to improve the merchant's clustering effect, so as to further improve the decision-making ability of the decision-making system. By extracting the merchant's historical behavior sequence information and time series information to learn the cross-attention mechanism, the merchants are clustered by this unsupervised learning method, and the historical data can be effectively used to learn the merchant's behavior pattern to cluster the merchants to adapt to changes at different times, improve the clustering effect, and improve the decision-making effect of the subsequent decision-making system. The unsupervised learning using the merchant's historical behavior sequence does not rely on expert experience or label data, but on By mining patterns and structures directly from the data itself, unsupervised learning can automatically discover natural clustering of merchants from a large number of behavior sequences. These clusterings are based on the intrinsic similarities of the data rather than relying on experts’ possibly outdated or rough judgments. They can capture the dynamic changes in merchant behavior in real time and discover subtle patterns that expert experience fails to cover, thereby providing a more refined clustering effect. Compared with supervised learning, unsupervised learning is not restricted by existing labels and can adapt to the evolution of merchant behavior patterns. By learning the latent structure in the data rather than relying on fixed labels to build, it can better generalize to new, unseen data, thereby maintaining a high level of prediction accuracy when merchant behavior patterns change. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flowchart of a method for merchant grouping provided in an embodiment of this specification.

[0019] Figure 2 A schematic diagram of the structure of an exemplary merchant representation information extraction model provided in an embodiment of this specification.

[0020] Figure 3 A schematic diagram of the structure of a device for merchant grouping provided in an embodiment of this specification.

[0021] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of this specification. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of this specification more clear, the technical solutions of this specification will be clearly and completely described below in combination with the specific embodiments of this specification and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this specification.

[0023] See also Figure 1 , is a flow chart of a method for merchant grouping provided in an embodiment of this specification. In an embodiment of this specification, the method for merchant grouping is applied to a device for merchant grouping (hereinafter referred to as "merchant grouping device") or an electronic device equipped with a merchant grouping device. Figure 1 The process shown is described in detail, and the method for merchant grouping may specifically include the following steps: S102, obtaining behavior sequence information corresponding to the merchant, wherein the behavior sequence information includes a plurality of historical behaviors of the merchant.

[0024] In some embodiments, the behavior sequence information corresponding to multiple merchants is obtained, and the behavior sequence information corresponding to each merchant includes the behavior information of multiple historical behaviors corresponding to the merchant, and the historical behaviors include but are not limited to payment behaviors, marketing behaviors, and other behaviors that the merchant has performed historically. It should be noted that the types of the above historical behaviors are only examples, not limitations. Those skilled in the art should understand that any type of historical behavior can be included in the protection scope of this specification. In some embodiments, the behavior information of historical behaviors includes but is not limited to any information related to historical behaviors, such as behavior content, behavior type, behavior occurrence time, behavior code, behavior participating users, etc., and this specification does not limit this. In some embodiments, multiple historical behaviors in the behavior sequence information are sorted according to a predetermined sorting rule, for example, multiple historical behaviors in the behavior sequence information are sorted according to the corresponding behavior occurrence time, for example, sorted in order from early to late according to the corresponding behavior occurrence time, or sorted in order from late to early according to the corresponding behavior occurrence time. It should be noted that the sorting rules of the above behavior sequence information are only examples, not limitations. Those skilled in the art should understand that the behavior sequence information of any sorting rules can be included in the protection scope of this specification.

[0025] S104, obtaining time interval sequence information corresponding to the merchant according to the behavior time intervals corresponding to every two adjacent historical behaviors in the behavior sequence information.

[0026] In some embodiments, for multiple historical behaviors in the behavior sequence information, the time interval between the behavior occurrence times corresponding to each two adjacent historical behaviors in the behavior sequence information (i.e., the behavior time interval) is obtained in sequence order from front to back, and the sequence composed of the multiple behavior time intervals obtained is used as the time interval sequence information corresponding to the merchant. For example, the behavior sequence information corresponding to a merchant is {behavior A, behavior B, behavior C, behavior D}, the behavior time interval corresponding to behavior A and behavior B is t1, the behavior time interval corresponding to behavior B and behavior C is t2, and the behavior time interval corresponding to behavior C and behavior D is t3, then the time interval sequence information corresponding to the merchant is {t1, t2, t3}. In some embodiments, since the length of the time interval sequence information corresponding to a merchant (i.e., the number of behavior time intervals contained in the time interval sequence information) is inconsistent with the length of the behavior sequence information corresponding to the merchant (i.e., the number of historical behaviors contained in the behavior sequence information), the length of the time interval sequence information is the length of the behavior sequence information minus one. In order to make the lengths of the two consistent, a preset value can be inserted at the first position in the time interval sequence information. This specification does not limit the preset value. For example, the preset value can be 0 or NAN. For example, the time interval sequence information can be {NAN, t1, t2, t3}.

[0027] S106, inputting the behavior sequence information and the time interval sequence information into a first embedding layer and a second embedding layer in a merchant representation information extraction model respectively to obtain a corresponding first embedding vector and a second embedding vector.

[0028] In some embodiments, the merchant characterization information extraction model is a neural network used to extract a characterization vector of a merchant's behavior pattern. The merchant characterization information extraction model includes a first embedding layer, a second embedding layer, and a cross-attention layer. The inputs of the first embedding layer and the second embedding layer are the inputs of the merchant characterization information extraction model. The output of the cross-attention layer is the output of the merchant characterization information extraction model. The output of the first embedding layer and the output of the second embedding layer are respectively input into the cross-attention layer, wherein the output of the cross-attention layer is a characterization vector. The characterization vector refers to a numerical vector that characterizes an object, that is, a characterization vector is used to characterize the merchant's behavior pattern.

[0029] In some embodiments, the embedding layer is used to convert high-dimensional discrete features into vector representations in a low-dimensional continuous space to facilitate the neural network model to learn these features. The embedding layer maps each discrete feature to a vector of a fixed length so that the distance between different features can reflect the semantic relationship between them, thereby helping the neural network to better understand and process these features.

[0030] In some embodiments, the behavior sequence information and the time interval sequence information are input into the merchant representation information extraction model, that is, the behavior sequence information is input into the first embedding layer in the merchant representation information extraction model, and the time interval sequence information is input into the second embedding layer in the merchant representation information extraction model. The first embedding layer will output a first embedding vector corresponding to the behavior sequence information, and the second embedding layer will output a second embedding vector corresponding to the time interval sequence information, wherein the first embedding vector includes a plurality of ordered vectors, each vector corresponds to a historical behavior in the behavior sequence information, and the order of a certain vector in the plurality of vectors is consistent with the order of the historical behavior corresponding to the vector in the behavior sequence information, that is, the first embedding vector is a vector sequence composed of a plurality of vectors, each vector corresponds to a historical behavior in the behavior sequence information, that is, the vector sequence corresponds to the behavior sequence information, and the second embedding vector also includes a plurality of ordered vectors, each vector corresponds to a behavior time interval in the time interval sequence information, and the order of a certain vector in the plurality of vectors is consistent with the order of the behavior time interval corresponding to the vector in the time interval sequence information, that is, the second embedding vector is also a vector sequence composed of a plurality of vectors, each vector corresponds to a behavior time interval in the time interval sequence information, that is, the vector sequence corresponds to the time interval sequence information.

[0031] S108, inputting the first embedding vector and the second embedding vector into the cross-attention layer in the merchant representation information extraction model for unsupervised learning, and obtaining the representation vector corresponding to the merchant output by the cross-attention layer in the merchant representation information extraction model after convergence, wherein the learning goal of the unsupervised learning is to predict the next behavior through the preceding sequence in the behavior sequence information.

[0032] In some embodiments, the first embedding vector and the second embedding vector are input into the cross attention layer in the merchant representation information extraction model for unsupervised learning. The cross attention layer is used to learn the correlation between the behavior sequence information and the time interval sequence information. The cross attention layer will pay attention to all elements in the other input sequence for each element in one input sequence, and calculate its representation based on the degree of attention. In this way, the two sequences can transmit information to each other, thereby achieving better representation.

[0033] In some embodiments, Cross Attention is an attention mechanism widely used in the field of deep learning, especially in natural language processing (NLP) and computer vision (CV). It allows the model to capture the dependencies between two different sequences when processing them. This mechanism enables the model to better understand the correlation between the two sequences, thereby performing well in multimodal tasks (such as image and text matching) or sequence-to-sequence tasks (such as machine translation).

[0034] In some embodiments, unsupervised learning is an important method in machine learning, which is different from supervised learning because unsupervised learning does not rely on predefined outputs or labels. The main goal of unsupervised learning is to discover patterns and structures in unlabeled data. In some embodiments, the learning goal of unsupervised learning is to predict the next behavior through the preamble sequence in the behavior sequence information, and the preamble sequence is a sequence consisting of all historical behaviors from the first historical behavior in the behavior sequence information to a certain historical behavior in the behavior sequence information except the first historical behavior. For example, if the behavior sequence information is {behavior A, behavior B, behavior C, behavior D}, then the learning goals of unsupervised learning include predicting the next behavior through the preamble sequence {behavior A}, predicting the next behavior through the preamble sequence {behavior A, behavior B}, and predicting the next behavior through the preamble sequence {behavior A, behavior B, behavior C}. In some embodiments, the loss function corresponding to unsupervised learning can be the loss between the next predicted behavior predicted by the preamble sequence and the next actual historical behavior in the behavior sequence information that is located after the preamble sequence (for example, cross entropy loss). This specification does not limit the type of loss function. For example, the behavior sequence information is {behavior A, behavior B, behavior C, behavior D}, and the next behavior is predicted by the preamble sequence {behavior A, behavior B}. The corresponding loss function is the cross entropy loss between the next predicted behavior predicted by the preamble sequence {behavior A, behavior B} and the next actual historical behavior in the behavior sequence information that is located after the preamble sequence, namely behavior C. In some embodiments, the loss function corresponding to unsupervised learning may also be the sum of the losses (e.g., cross entropy losses) between the next predicted behavior predicted by each subsequence in the preamble sequence corresponding to the same starting point as the preamble sequence and the next actual historical behavior in the behavior sequence information following the subsequence. This specification does not limit the type of loss function. For example, the behavior sequence information is {behavior A, behavior B, behavior C, behavior D}, and the next behavior is predicted by the preamble sequence {behavior A, behavior B, behavior C}. The corresponding loss function is the sum of the cross entropy losses between the next predicted behavior predicted by the preamble sequence {behavior A} and the next actual historical behavior following the preamble sequence in the behavior sequence information, namely, behavior B, the cross entropy losses between the next predicted behavior predicted by the preamble sequence {behavior A, behavior B} and the next actual historical behavior following the preamble sequence in the behavior sequence information, namely, behavior C, and the cross entropy losses between the next predicted behavior predicted by the preamble sequence {behavior A, behavior B} and the next actual historical behavior following the preamble sequence in the behavior sequence information, namely, behavior D.

[0035] In some embodiments, the unsupervised learning of the merchant characterization information extraction model may be unsupervised learning of only the parameters of the cross-attention layer, or may be unsupervised learning of the parameters of the first embedding layer, the second embedding layer, and the cross-attention layer. In some embodiments, the merchant characterization information extraction model is subjected to unsupervised learning so that the model converges. Model convergence refers to the process in which the model gradually learns the patterns and features of the data during the training process, and the adjustment of the parameters gradually tends to be stable. The representation vector corresponding to the merchant output by the cross-attention layer in the converged merchant characterization information extraction model is obtained, and the behavior pattern of the merchant is represented as a vector.

[0036] S110, grouping the multiple merchants according to the representation vectors respectively corresponding to the multiple merchants to obtain multiple merchant groups.

[0037] In some embodiments, after obtaining the characterization vectors corresponding to a plurality of merchants, the plurality of merchants are grouped according to the characterization vectors corresponding to each of the plurality of merchants to obtain a plurality of merchant groups, each of which includes at least one of the plurality of merchants, and each of the plurality of merchants is only classified into one of the plurality of merchant groups, and is not classified into a plurality of merchant groups at the same time. In some embodiments, the characterization vectors corresponding to all merchants in each merchant group satisfy a predetermined similarity condition, that is, the plurality of merchants whose corresponding characterization vectors satisfy a predetermined similarity condition will be classified into the same merchant group, that is, the plurality of merchants whose corresponding characterization vectors have a high similarity will be classified into the same merchant group, that is, the plurality of merchants whose corresponding behavior patterns have a high similarity will be classified into the same merchant group. In some embodiments, for each merchant group, the similarity between the characterization vectors corresponding to any two merchants in the merchant group is greater than or equal to a first similarity threshold; for two different merchant groups, the similarity between the characterization vector corresponding to any merchant in one merchant group and the characterization vector corresponding to any merchant in another merchant group is less than a second similarity threshold.

[0038] In some embodiments, the K-Means clustering algorithm (a commonly used clustering algorithm) can be used to group multiple merchants based on the characterization vectors corresponding to the multiple merchants. The K-Means clustering algorithm is mainly used to divide the data set into several clusters, so that the similarity of data points in the same cluster is high, while the similarity of data points between different clusters is low. The K-Means clustering algorithm belongs to unsupervised learning and does not require a pre-labeled data set. It should be noted that the above clustering algorithm is only for example, not for limitation. Those skilled in the art should understand that any clustering algorithm can be included in the protection scope of this specification.

[0039] In the embodiments of this specification, a method for clustering merchants in a decision system based on sequence similarity is proposed, and the merchant's historical sequence information is used to characterize the merchant's behavior pattern, so as to improve the merchant's clustering effect, so as to further improve the decision-making ability of the decision-making system. By extracting the merchant's historical behavior sequence information and time series information to learn the cross-attention mechanism, the merchants are clustered by this unsupervised learning method, and the historical data can be effectively used to learn the merchant's behavior pattern to cluster the merchants to adapt to changes at different times, improve the clustering effect, and improve the decision-making effect of the subsequent decision-making system. The unsupervised learning using the merchant's historical behavior sequence does not rely on expert experience or label data, but on By mining patterns and structures directly from the data itself, unsupervised learning can automatically discover natural clustering of merchants from a large number of behavior sequences. These clusterings are based on the intrinsic similarities of the data rather than relying on experts’ possibly outdated or rough judgments. They can capture the dynamic changes in merchant behavior in real time and discover subtle patterns that expert experience fails to cover, thereby providing a more refined clustering effect. Compared with supervised learning, unsupervised learning is not restricted by existing labels and can adapt to the evolution of merchant behavior patterns. By learning the latent structure in the data rather than relying on fixed labels to build, it can better generalize to new, unseen data, thereby maintaining a high level of prediction accuracy when merchant behavior patterns change.

[0040] In some embodiments, the obtaining of the behavior sequence information corresponding to the merchant includes: sorting the multiple historical behaviors of the merchant according to the corresponding behavior occurrence time, and obtaining the behavior sequence information corresponding to the merchant. In some embodiments, the multiple historical behaviors of the merchant are sorted according to the corresponding behavior occurrence time (for example, the order of behavior occurrence time from early to late, or the order of behavior occurrence time from late to early), and the sorted multiple historical behaviors are obtained, and the behavior sequence information corresponding to the merchant is obtained according to the sorted multiple historical behaviors, for example, the behavior sequence information is a sequence composed of the behavior types of the multiple historical behaviors.

[0041] In some embodiments, the ordering of the multiple historical behaviors of the merchant according to the corresponding behavior occurrence time to obtain the behavior sequence information corresponding to the merchant includes: ordering the multiple historical behaviors of the merchant according to the corresponding behavior occurrence time to obtain the ordered multiple historical behaviors; for each of the ordered multiple historical behaviors, according to the behavior type corresponding to the historical behavior, determining the behavior coding information of the historical behavior; obtaining the behavior sequence information corresponding to the merchant, wherein the behavior sequence includes the ordered multiple behavior coding information. In some embodiments, after obtaining the ordered multiple historical behaviors, for each of the multiple historical behaviors, according to the behavior type corresponding to the historical behavior, and the mapping relationship between the preset behavior type and the behavior coding information (for example, the payment behavior corresponds to the code "1", the marketing behavior corresponds to the code "2", etc.), the behavior coding information mapped by the behavior type is used as the behavior coding information of the historical behavior. In some embodiments, according to the behavior coding information corresponding to the ordered multiple historical behaviors, a plurality of ordered behavior coding information is obtained, and the sequence composed of the ordered multiple behavior coding information is used as the behavior sequence information corresponding to the merchant.

[0042] In some embodiments, the obtaining of the time interval sequence information corresponding to the merchant according to the behavior time interval corresponding to each two adjacent historical behaviors in the behavior sequence information includes: obtaining the time interval unit corresponding to the behavior time interval by unitizing the behavior time interval according to the behavior time interval corresponding to each two adjacent historical behaviors in the behavior sequence information; and obtaining the time interval sequence information corresponding to the merchant according to the time interval unit. In some embodiments, after obtaining the behavior time interval corresponding to each two adjacent historical behaviors in the behavior sequence information, for each behavior time interval, the behavior time interval is unitized according to a preset unit to obtain the time interval unit corresponding to the behavior time interval. For example, if the behavior time intervals are 50 minutes and 100 minutes respectively, and 10 minutes is used as a unit, the corresponding time interval units obtained are 5 units and 10 units respectively. In some embodiments, a sequence consisting of multiple time interval units obtained is used as the time interval sequence information corresponding to the merchant. For example, the behavior sequence information corresponding to a merchant is {behavior A, behavior B, behavior C, behavior D}, the behavior time interval corresponding to behavior A and behavior B is 50 minutes, the behavior time interval corresponding to behavior B and behavior C is 100 minutes, and the behavior time interval corresponding to behavior C and behavior D is 200 minutes. Taking 10 minutes as a unit, the time interval sequence information corresponding to the merchant is {5 units, 10 units, 20 units}.

[0043] In some embodiments, the loss function corresponding to the learning objective of the unsupervised learning is the cross entropy loss between the next behavior predicted by the preamble sequence in the behavior sequence information and the next historical behavior in the behavior sequence information located after the preamble sequence. In some embodiments, the loss function corresponding to unsupervised learning can be the cross entropy loss between the next behavior predicted by the preamble sequence (i.e., the predicted behavior) and the next historical behavior in the behavior sequence information located after the preamble sequence (i.e., the actual behavior). For example, the behavior sequence information is {behavior A, behavior B, behavior C, behavior D}, and the next behavior is predicted by the preamble sequence {behavior A, behavior B}. The corresponding loss function is the cross entropy loss between the next predicted behavior predicted by the preamble sequence {behavior A, behavior B} and the next actual historical behavior located after the preamble sequence, i.e., behavior C.

[0044] In some embodiments, the loss function corresponding to the learning objective of the unsupervised learning is the sum of the cross entropy losses corresponding to each historical behavior from the second historical behavior in the behavior sequence information to the next historical behavior after the preceding sequence in the behavior sequence information. In some embodiments, the loss function corresponding to unsupervised learning can also be the sum of the cross-entropy losses between the next behavior predicted by each subsequence in the preamble sequence corresponding to the same starting point as the preamble sequence (i.e., the predicted behavior) and the next historical behavior (i.e., the actual behavior) after the subsequence in the behavior sequence information. For example, the behavior sequence information is {behavior A, behavior B, behavior C, behavior D}, and the next behavior is predicted by the preamble sequence {behavior A, behavior B, behavior C}. The corresponding loss function is the sum of the cross-entropy losses between the next predicted behavior predicted by the preamble sequence {behavior A} and the next actual historical behavior after the preamble sequence in the behavior sequence information, i.e., behavior B, the cross-entropy loss between the next predicted behavior predicted by the preamble sequence {behavior A, behavior B} and the next historical behavior after the preamble sequence in the behavior sequence information, i.e., behavior C, and the cross-entropy loss between the next predicted behavior predicted by the preamble sequence {behavior A, behavior B} and the next historical behavior after the preamble sequence in the behavior sequence information, i.e., behavior D.

[0045] In some embodiments, the obtaining of the representation vector corresponding to the merchant output by the cross-attention layer in the merchant representation information extraction model after convergence includes: obtaining the target vector corresponding to the merchant output by the cross-attention layer in the merchant representation information extraction model after convergence; inputting the target vector into the fully connected layer in the merchant representation information extraction model to obtain a representation vector of a preset size. In some embodiments, the merchant representation information extraction model also includes a fully connected layer. At this time, the output of the fully connected layer is used as the output of the merchant representation information extraction model, and the output of the cross-attention layer is no longer used as the output of the merchant representation information extraction model. The output of the cross-attention layer will be input into the fully connected layer. In some embodiments, the target vector output by the cross-attention layer will be input into the fully connected layer. The function of the fully connected layer is to customize the size of the representation vector so that the fully connected layer outputs a representation vector of a preset size.

[0046] In some embodiments, the multiple merchants are grouped according to the characterization vectors respectively corresponding to the multiple merchants to obtain multiple merchant groups, including: randomly selecting a preset number of multiple target merchants from the multiple merchants, and using the characterization vectors of the multiple target merchants as initial cluster centers; for each merchant among the multiple merchants, according to the distance from the characterization vector of the merchant to each cluster center, the merchant is assigned to the cluster to which the nearest cluster center belongs; the average value of the characterization vectors of all merchants in each cluster is recalculated, and the average value is used as the new cluster center corresponding to the cluster; for each merchant among the multiple merchants, the merchant is reassigned to the cluster to which the nearest cluster center belongs, and so on, until the preset iteration end condition is met. In some embodiments, a preset number of target merchants are first randomly selected from multiple merchants, and the characterization vector of each target merchant is used as an initial cluster center (a cluster center belongs to one cluster, and one cluster contains one cluster center). For each merchant among the multiple merchants, according to the distance from the merchant's characterization vector to each cluster center (Cosine Distance can be used for calculation, and this specification does not limit the calculation method of vector distance), the merchant is assigned to the cluster to which the nearest cluster center belongs, and then the center of each cluster is recalculated, that is, the average value of the characterization vectors of all merchants in each cluster is used as the new cluster center corresponding to the cluster. For each merchant among the multiple merchants, according to the distance from the merchant's characterization vector to each new cluster center, the merchant is reassigned to the cluster to which the nearest new cluster center belongs, and then the center of each cluster is recalculated again, and so on, until the preset iteration end condition is met, and each current latest cluster is taken as a merchant group. When all merchants in the cluster are included in the merchant group, that is, all merchants in the cluster belong to the merchant group.

[0047] In some embodiments, the preset iteration termination condition includes at least one of the following: the change of cluster centers meets the preset condition; the number of iterations reaches the preset number threshold. In some embodiments, the preset iteration termination condition may be that the number of iterations reaches the preset number threshold, or it may be that the change of cluster centers meets the preset condition. For example, the preset condition may be that the current latest cluster centers after this iteration have not changed significantly compared with the cluster centers after the previous iteration. For example, for each cluster, the distance between the previous latest cluster center corresponding to the cluster after this iteration and the cluster center corresponding to the cluster after the previous iteration is obtained, and the average value of the cluster center distances corresponding to each cluster is compared with the preset threshold. If the average value is less than or equal to the preset threshold, it can be determined that the current latest cluster centers after this iteration have not changed significantly compared with the cluster centers after the previous iteration. If the average value is greater than the preset threshold, it can be determined that the current latest cluster centers after this iteration have changed significantly compared with the cluster centers after the previous iteration.

[0048] In some embodiments, the method further includes: determining the preset number according to the preset merchant control granularity. In some embodiments, the number of target merchants randomly selected from the multiple merchants as the initial cluster center (i.e., the preset number) depends on the preset merchant control granularity. The larger the granularity, i.e., the coarser the control and the fewer the groups, the smaller the preset number. The smaller the granularity, i.e., the finer the control and the more the groups, the larger the preset number. In some embodiments, a mapping relationship between the merchant control granularity and the number of merchants is pre-configured, and the number of merchants mapped by the preset merchant control granularity can be used as the preset number. In some embodiments, the preset merchant control granularity can also be input into a preset functional relationship, and the output of the functional relationship can be used as the preset number.

[0049] Figure 2 This is a schematic diagram of the structure of an example merchant representation information extraction model provided in the embodiments of this specification.

[0050] like Figure 2 As shown, the merchant representation information extraction model includes a first embedding layer, a second embedding layer and a cross-attention layer. The behavior sequence information corresponding to the merchant is input into the first embedding layer to obtain a first embedding vector output by the first embedding layer. The time interval sequence information corresponding to the merchant is input into the second embedding layer to obtain a second embedding vector output by the second embedding layer. The first embedding vector and the second embedding vector are input into the cross-attention layer to obtain a representation vector corresponding to the merchant output by the cross-attention layer.

[0051] Figure 3A schematic diagram of a device for merchant grouping provided in an embodiment of this specification, the device for merchant grouping (hereinafter referred to as "merchant grouping device 1") can be implemented as all or part of an electronic device through software, hardware or a combination of both. According to some embodiments, the merchant grouping device 1 includes a behavior sequence acquisition module 11, a time interval sequence acquisition module 12, an embedding vector acquisition module 13, an unsupervised learning module 14 and a merchant grouping module 15.

[0052] The behavior sequence obtaining module 11 is used to obtain behavior sequence information corresponding to a merchant, wherein the behavior sequence information includes a plurality of historical behaviors of the merchant.

[0053] The time interval sequence obtaining module 12 is used to obtain the time interval sequence information corresponding to the merchant according to the behavior time interval corresponding to every two adjacent historical behaviors in the behavior sequence information.

[0054] The embedding vector obtaining module 13 is used to input the behavior sequence information and the time interval sequence information into the first embedding layer and the second embedding layer in the merchant representation information extraction model respectively to obtain the corresponding first embedding vector and second embedding vector.

[0055] The unsupervised learning module 14 is used to input the first embedding vector and the second embedding vector into the cross-attention layer in the merchant representation information extraction model for unsupervised learning, and obtain the representation vector corresponding to the merchant output by the cross-attention layer in the merchant representation information extraction model after convergence, wherein the learning goal of the unsupervised learning is to predict the next behavior through the preceding sequence in the behavior sequence information.

[0056] The merchant grouping module 15 is used to group the multiple merchants according to the characterization vectors respectively corresponding to the multiple merchants to obtain multiple merchant groups.

[0057] In some embodiments, obtaining the behavior sequence information corresponding to the merchant includes: The multiple historical behaviors of the merchant are sorted according to the corresponding behavior occurrence time to obtain the behavior sequence information corresponding to the merchant.

[0058] In some embodiments, the ordering of the multiple historical behaviors of the merchant according to the corresponding behavior occurrence time to obtain the behavior sequence information corresponding to the merchant includes: Sorting the multiple historical behaviors of the merchant according to the corresponding behavior occurrence time to obtain the sorted multiple historical behaviors; For each of the sorted multiple historical behaviors, determining behavior coding information of the historical behavior according to a behavior type corresponding to the historical behavior; Behavior sequence information corresponding to the merchant is obtained, wherein the behavior sequence includes a plurality of sorted behavior coding information.

[0059] In some embodiments, obtaining the time interval sequence information corresponding to the merchant according to the behavior time intervals corresponding to each two adjacent historical behaviors in the behavior sequence information includes: According to the behavior time interval corresponding to each two adjacent historical behaviors in the behavior sequence information, by unitizing the behavior time interval, a time interval unit corresponding to the behavior time interval is obtained; According to the time interval unit, the time interval sequence information corresponding to the merchant is obtained.

[0060] In some embodiments, the loss function corresponding to the learning objective of the unsupervised learning is the cross entropy loss between the next behavior predicted by the preceding sequence in the behavior sequence information and the next historical behavior corresponding to the preceding sequence in the behavior sequence information.

[0061] In some embodiments, the loss function corresponding to the learning objective of the unsupervised learning is the sum of the cross entropy losses corresponding to each historical behavior from the second historical behavior in the behavior sequence information to the next historical behavior corresponding to the preceding sequence in the behavior sequence information.

[0062] In some embodiments, the step of obtaining the representation vector corresponding to the merchant output by the cross attention layer in the converged merchant representation information extraction model includes: Obtaining a target vector corresponding to the merchant output by a cross-attention layer in a converged merchant representation information extraction model; The target vector is input into the fully connected layer in the merchant representation information extraction model to obtain a representation vector of a preset size.

[0063] In some embodiments, grouping the multiple merchants according to the representation vectors respectively corresponding to the multiple merchants to obtain multiple merchant groups includes: Randomly selecting a preset number of multiple target merchants from the multiple merchants, and using the representation vectors of the multiple target merchants as initial cluster centers; For each merchant among the multiple merchants, according to the distance between the merchant's characterization vector and each cluster center, the merchant is assigned to the cluster to which the nearest cluster center belongs; Recalculate the average value of the representation vectors of all merchants in each cluster and use the average value as the new cluster center corresponding to the cluster; For each merchant among the multiple merchants, the merchant is reallocated to the cluster to which the nearest cluster center belongs, and so on, until a preset iteration end condition is met.

[0064] In some embodiments, the preset iteration end condition includes at least one of the following: The change of cluster center meets the preset conditions; The number of iterations has reached the preset threshold.

[0065] In some embodiments, the merchant grouping device 1 is further used for: The preset quantity is determined according to the preset merchant management and control granularity.

[0066] The above device embodiments correspond to the method embodiments. For specific descriptions, please refer to the description of the method embodiments, which will not be repeated here. The device embodiments are obtained based on the corresponding method embodiments and have the same technical effects as the corresponding method embodiments. For specific descriptions, please refer to the corresponding method embodiments.

[0067] The embodiment of the present specification also provides a computer storage medium, which can store multiple instructions, and the instructions are suitable for being loaded by a processor to execute the method of the embodiment of the present specification.

[0068] The embodiments of the present specification also provide a computer program product, which stores at least one instruction, and the at least one instruction is loaded by the processor to execute the method of the embodiments of the present specification.

[0069] The embodiments of this specification also provide Figure 4 The structural diagram of the electronic device shown in FIG. Figure 4 At the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the method of the above-mentioned embodiment of this specification.

[0070] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, 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.

[0071] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0072] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0073] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0074] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0075] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0076] This specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0077] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0078] The above description is only an embodiment of the present specification and is not intended to limit the present specification. For those skilled in the art, the present specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification shall be included in the scope of the claims of the present specification.

Claims

1. A method for merchant grouping, comprising: Obtaining behavior sequence information corresponding to a merchant, wherein the behavior sequence information includes multiple historical behaviors of the merchant; Obtaining time interval sequence information corresponding to the merchant according to the behavior time intervals corresponding to every two adjacent historical behaviors in the behavior sequence information; Inputting the behavior sequence information and the time interval sequence information into the first embedding layer and the second embedding layer of the merchant representation information extraction model respectively to obtain the corresponding first embedding vector and the second embedding vector; Inputting the first embedding vector and the second embedding vector into the cross-attention layer in the merchant representation information extraction model for unsupervised learning, and obtaining a representation vector corresponding to the merchant output by the cross-attention layer in the merchant representation information extraction model after convergence, wherein the learning goal of the unsupervised learning is to predict the next behavior through the preceding sequence in the behavior sequence information; The multiple merchants are grouped according to the representation vectors respectively corresponding to the multiple merchants to obtain multiple merchant groups.

2. According to the method of claim 1, the step of obtaining the behavior sequence information corresponding to the merchant comprises: The multiple historical behaviors of the merchant are sorted according to the corresponding behavior occurrence time to obtain the behavior sequence information corresponding to the merchant.

3. According to the method of claim 2, the step of sorting the multiple historical behaviors of the merchant according to the corresponding behavior occurrence time to obtain the behavior sequence information corresponding to the merchant comprises: Sorting the multiple historical behaviors of the merchant according to the corresponding behavior occurrence time to obtain the sorted multiple historical behaviors; For each of the sorted multiple historical behaviors, determining behavior coding information of the historical behavior according to the behavior type corresponding to the historical behavior; Behavior sequence information corresponding to the merchant is obtained, wherein the behavior sequence includes a plurality of sorted behavior coding information.

4. The method according to claim 1, wherein obtaining the time interval sequence information corresponding to the merchant according to the behavior time intervals corresponding to each two adjacent historical behaviors in the behavior sequence information comprises: According to the behavior time interval corresponding to each two adjacent historical behaviors in the behavior sequence information, by unitizing the behavior time interval, a time interval unit corresponding to the behavior time interval is obtained; According to the time interval unit, the time interval sequence information corresponding to the merchant is obtained.

5. According to the method described in claim 1, the loss function corresponding to the learning objective of the unsupervised learning is the cross entropy loss between the next behavior predicted by the preceding sequence in the behavior sequence information and the next historical behavior in the behavior sequence information that is located after the preceding sequence.

6. According to the method described in claim 1, the loss function corresponding to the learning objective of the unsupervised learning is the sum of the cross entropy losses corresponding to each historical behavior from the second historical behavior in the behavior sequence information to the next historical behavior after the preceding sequence in the behavior sequence information.

7. According to the method of claim 1, the step of obtaining the representation vector corresponding to the merchant output by the cross attention layer in the converged merchant representation information extraction model comprises: Obtaining a target vector corresponding to the merchant output by a cross-attention layer in a converged merchant representation information extraction model; The target vector is input into the fully connected layer in the merchant representation information extraction model to obtain a representation vector of a preset size.

8. The method according to claim 1, wherein the step of grouping the multiple merchants according to the representation vectors respectively corresponding to the multiple merchants to obtain multiple merchant groups comprises: Randomly selecting a preset number of multiple target merchants from the multiple merchants, and using the representation vectors of the multiple target merchants as initial cluster centers; For each merchant among the multiple merchants, according to the distance between the merchant's characterization vector and each cluster center, the merchant is assigned to the cluster to which the nearest cluster center belongs; Recalculate the average value of the representation vectors of all merchants in each cluster and use the average value as the new cluster center corresponding to the cluster; For each merchant among the multiple merchants, the merchant is reallocated to the cluster to which the nearest cluster center belongs, and so on, until a preset iteration end condition is met.

9. According to the method of claim 8, the preset iteration end condition comprises at least one of the following: The change of cluster center meets the preset conditions; The number of iterations has reached the preset threshold.

10. The method according to claim 8 or 9, further comprising: The preset quantity is determined according to the preset merchant management and control granularity.

11. A device for merchant grouping, comprising: A behavior sequence acquisition module, used to obtain behavior sequence information corresponding to a merchant, wherein the behavior sequence information includes a plurality of historical behaviors of the merchant; A time interval sequence acquisition module, used to obtain the time interval sequence information corresponding to the merchant according to the behavior time intervals corresponding to every two adjacent historical behaviors in the behavior sequence information; An embedding vector acquisition module, used to input the behavior sequence information and the time interval sequence information into the first embedding layer and the second embedding layer of the merchant representation information extraction model respectively, to obtain the corresponding first embedding vector and second embedding vector; an unsupervised learning module, configured to input the first embedding vector and the second embedding vector into a cross-attention layer in the merchant representation information extraction model for unsupervised learning, and obtain a representation vector corresponding to the merchant output by the cross-attention layer in the merchant representation information extraction model after convergence, wherein the learning goal of the unsupervised learning is to predict the next behavior through the preceding sequence in the behavior sequence information; The merchant grouping module is used to group the multiple merchants according to the representation vectors respectively corresponding to the multiple merchants to obtain multiple merchant groups.

12. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 10 are implemented.

13. An electronic device, characterized in that: include: A processor and a memory; wherein the memory stores a computer program, and the computer program is suitable for being loaded by the processor and executing the steps of the method as claimed in any one of claims 1 to 10.

14. A computer program product having at least one instruction stored thereon, characterized in that: When the at least one instruction is executed by the processor, the steps of the method described in any one of claims 1 to 10 are implemented.

Images