Intelligent branch distribution method, device and equipment for network, and storage medium
By extracting customer characteristics and constructing an intent matrix and a terminal resource matrix, the system intelligently allocates terminal resources from bank branches, solving the problem of inaccurate customer demand prediction and achieving efficient allocation of terminal resources and improved customer experience.
Patent Information
- Application Number
- CN202511042263.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-28
- Publication Date
- 2025-10-31
AI Technical Summary
In traditional bank branch services, inaccurate customer demand forecasting and unreasonable allocation of terminal resources lead to low service efficiency and poor customer experience.
By acquiring basic customer information and historical behavior information, features are extracted using convolutional neural networks, Transformers, and graph neural networks to construct intent matrices and terminal resource matrices, and then the most suitable terminal devices are recommended through matching.
This improved the customer service experience, reduced front desk inquiries, optimized the allocation of terminal resources, prevented customers from queuing twice, and increased service efficiency.
Smart Images
Figure CN120875422A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, device and storage medium for intelligent branch traffic diversion. Background Technology
[0002] In traditional bank branch service scenarios, customers typically go through a process of "front desk inquiry of needs - manual terminal allocation - customer transaction processing." While this model is simple and direct, it exposes several problems in practice, failing to meet the efficiency and experience requirements of modern financial services. Specifically, these include: front desk staff cannot anticipate customer needs and must repeatedly inquire with customers. Furthermore, due to limitations in customers' language expression accuracy and unfamiliarity with service categories, front desk staff must repeatedly inquire about and explain customer service requirements, resulting in significant workload spent on requirement classification and guidance, leading to overall low service efficiency at the branch. Whether manually allocating terminals or using a queuing system, the reliance on subjective experience or simple, unclassified cash / non-cash windows makes it difficult to adapt to dynamically changing terminal resources in real time, such as the terminal's cash withdrawal limit, available card types, equipment malfunction status, and queue length. If a customer needs to withdraw a large sum of cash but is directed to a terminal with insufficient cash, they will have to queue again, resulting in a poor customer experience. Summary of the Invention
[0003] To address the above problems, this application provides a method for intelligent branch traffic diversion, including the following:
[0004] Firstly, this application provides a method for intelligent branch traffic diversion, the method comprising:
[0005] Obtain basic customer information and historical behavior information, and obtain customer behavior characteristics based on the basic information and historical behavior information;
[0006] The customer behavior characteristics are input into a pre-trained prediction network to identify customer intent.
[0007] Construct a customer intent matrix for this visit based on the customer intent recognition results;
[0008] A terminal resource matrix is generated based on real-time changes in the network terminal resource information database.
[0009] The intent matrix and the terminal resource matrix are matched, and the terminal with the highest score in the matching results is recommended to the customer for business processing.
[0010] Optionally, obtaining customer behavior characteristics based on the basic information and historical behavior information includes:
[0011] Extract the customer's static characteristics, dynamic behavioral characteristics, and business transfer characteristics from the customer's basic information and historical behavior information;
[0012] Based on the customer's static characteristics, dynamic behavioral characteristics, and business transfer characteristics, a vector corresponding to the static feature vector, the dynamic behavioral characteristics, and the business transfer characteristics is obtained;
[0013] The customer behavior features are obtained by fusing and concatenating the vectors corresponding to the static feature vector, the dynamic behavior features, and the business transfer features.
[0014] Optionally, matching the intent matrix and the terminal resource matrix includes:
[0015] Based on the intent matrix and the terminal resource matrix, combined with preset penalty and reward factors, the matching score of each terminal is calculated.
[0016] Optionally, the static features include the customer's age, occupation, asset size, and income level; the dynamic behavioral features are the customer's business transaction records, including time, type, amount, and transaction result; and the business transfer features are the transfer probability from business A to business B and the business association graph.
[0017] Optionally, the process of obtaining a static feature vector based on the static features includes:
[0018] Static features are input into a convolutional neural network, and features are extracted from the static features through convolutional layers and pooling layers to obtain static feature vectors.
[0019] Optionally, the process of obtaining a static feature vector based on the dynamic behavioral features includes:
[0020] Dynamic behavioral features are input into the Transformer architecture, and its encoder part is used to encode the behavioral feature sequence.
[0021] A multi-head attention mechanism is used to assign learnable attention weights to different historical behaviors, and to identify the impact of key behaviors on intent.
[0022] Positional encoding technology is used to add positional information to the behavioral features of each time step to preserve the temporal structure of the sequence. A time decay weighting mechanism is introduced to assign different weights to behaviors at different time steps based on the interval between the operation time and the current time, so that the most recent behavior contributes more to the feature vector, and the temporal feature vector is output.
[0023] Optionally, the process of obtaining a static feature vector based on the business transfer characteristics includes:
[0024] By taking the customer's business transfer characteristics as input, a graph neural network is used to perform graph propagation on the time-series features, model the relationships between the time-series features, obtain the implicit associations between customers and between businesses, identify the transfer patterns between businesses, and output the transfer feature vector.
[0025] Secondly, this application provides a smart branch traffic diversion device, which includes:
[0026] The acquisition unit is used to acquire basic customer information and historical behavior information, and to obtain customer behavior characteristics based on the basic information and historical behavior information.
[0027] The identification unit is used to input the customer behavior features into a pre-trained prediction network to identify customer intent.
[0028] The matrix generation unit is used to construct the customer's intent matrix for this visit based on the customer intent recognition results; and to generate a terminal resource matrix based on real-time changes in the network terminal resource information database.
[0029] The matching unit is used to match the intent matrix and the terminal resource matrix, and recommend the terminal with the highest score in the matching results to the customer for business processing.
[0030] Optionally, the acquisition unit obtains customer behavior characteristics based on the basic information and historical behavior information, including:
[0031] Extract the customer's static characteristics, dynamic behavioral characteristics, and business transfer characteristics from the customer's basic information and historical behavior information;
[0032] Based on the customer's static characteristics, dynamic behavioral characteristics, and business transfer characteristics, a vector corresponding to the static feature vector, the dynamic behavioral characteristics, and the business transfer characteristics is obtained;
[0033] The customer behavior features are obtained by fusing and concatenating the vectors corresponding to the static feature vector, the dynamic behavior features, and the business transfer features.
[0034] Optionally, the matching unit matches the intent matrix and the terminal resource matrix by:
[0035] Based on the intent matrix and the terminal resource matrix, combined with preset penalty and reward factors, the matching score of each terminal is calculated.
[0036] Optionally, the static features include the customer's age, occupation, asset size, and income level; the dynamic behavioral features are the customer's business transaction records, including time, type, amount, and transaction result; and the business transfer features are the transfer probability from business A to business B and the business association graph.
[0037] Optionally, the process of obtaining a static feature vector based on the static features includes:
[0038] Static features are input into a convolutional neural network, and features are extracted from the static features through convolutional layers and pooling layers to obtain static feature vectors.
[0039] Optionally, the process of obtaining a static feature vector based on the dynamic behavioral features includes:
[0040] Dynamic behavioral features are input into the Transformer architecture, and its encoder part is used to encode the behavioral feature sequence.
[0041] A multi-head attention mechanism is used to assign learnable attention weights to different historical behaviors, and to identify the impact of key behaviors on intent.
[0042] Positional encoding technology is used to add positional information to the behavioral features of each time step to preserve the temporal structure of the sequence. A time decay weighting mechanism is introduced to assign different weights to behaviors at different time steps based on the interval between the operation time and the current time, so that the most recent behavior contributes more to the feature vector, and the temporal feature vector is output.
[0043] Optionally, the process of obtaining a static feature vector based on the business transfer characteristics includes:
[0044] By taking the customer's business transfer characteristics as input, a graph neural network is used to perform graph propagation on the time-series features, model the relationships between the time-series features, obtain the implicit associations between customers and between businesses, identify the transfer patterns between businesses, and output the transfer feature vector.
[0045] Thirdly, this application provides an apparatus comprising a memory and a processor, the memory for storing instructions or code, and the processor for executing the instructions or code to cause the apparatus to perform the intelligent branch traffic diversion method described in any of the implementations of the first aspect.
[0046] Fourthly, this application provides a computer-readable storage medium storing code, wherein when the code is executed, a device running the code implements the intelligent branch traffic diversion method described in any of the implementations of the first aspect.
[0047] This application provides a method for intelligent customer triage at service outlets. When executing the method, firstly, basic customer information and historical behavior information are acquired. Based on this information, customer behavior characteristics are obtained. These characteristics are then input into a pre-trained prediction network for customer intent recognition. Next, an intent matrix for the customer's current visit is constructed based on the intent recognition results. A terminal resource matrix is generated based on real-time data changes in the outlet's terminal resource information database. Finally, the intent matrix and the terminal resource matrix are matched, and the terminal with the highest score in the matching results is recommended to the customer for service processing. In this way, by combining the user's basic information and historical behavior information with real-time data from the outlet's terminal resource information database, it is possible to predict user intent and recommend suitable terminal devices for service processing based on the real-time status of outlet equipment. This effectively reduces front desk inquiries, accurately matches terminal resources to customer needs, and significantly improves the customer service experience. Attached Figure Description
[0048] To more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings used in the description of the embodiment or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0049] Figure 1 A flowchart illustrating an intelligent branch traffic diversion method provided in this application embodiment;
[0050] Figure 2 This is a schematic diagram of the structure of an intelligent branch diversion device provided in an embodiment of this application. Detailed Implementation
[0051] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.
[0052] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.
[0053] Figure 1 A flowchart illustrating a smart branch traffic diversion method provided in an embodiment of this application. Figure 1 As shown in the embodiments of this application, the intelligent branch traffic diversion method may include:
[0054] S101. Obtain the customer's basic information and historical behavior information, and obtain the customer's behavioral characteristics based on the basic information and historical behavior information.
[0055] Basic customer information can include their age, occupation, asset size, and income level, while historical behavioral information refers to records of their past business-related activities, such as transaction history, service usage, and consultation records. By comprehensively analyzing both basic customer information and historical behavior, we can more accurately predict customer intentions and provide more precise service solutions.
[0056] After obtaining the customer's basic information and historical behavior information, further feature extraction is performed. This includes extracting the customer's static features, dynamic behavioral features, and business transfer features from the basic information and historical behavior information. The static features include the customer's age, occupation, asset size, and income level; the dynamic behavioral features are the customer's business transaction records, including time, type, amount, and transaction result; and the business transfer features are the probability of transferring from business A to business B and the business association graph. In this way, by extracting static features, dynamic behavioral features, and business transfer features, multi-dimensional customer information can be comprehensively captured, including not only the customer's basic attributes but also the customer's behavioral patterns and business preferences, providing a rich data foundation for subsequent customer analysis and personalized services.
[0057] After preliminary feature extraction from customer information, vectors corresponding to the static feature vector, the dynamic behavior feature, and the business transfer feature are further obtained based on the customer's static features, dynamic behavior features, and business transfer features.
[0058] The following describes, with reference to specific embodiments, the process of obtaining the vector corresponding to the static feature vector, the dynamic behavior feature, and the business transfer feature based on the customer's static features, dynamic behavior features, and business transfer features in this application.
[0059] The process of obtaining a static feature vector based on the static features includes: inputting the static features into a convolutional neural network, extracting features from the static features through convolutional layers and pooling layers to obtain a static feature vector. This static feature vector is mainly used for basic customer classification, such as guiding high-asset customers to the VIP channel and elderly customers to the age-friendly window, etc. After static feature processing, a vector of static features is obtained, denoted as S.
[0060] The process of obtaining static feature vectors based on the dynamic behavioral features includes: inputting the dynamic behavioral features into a Transformer architecture, using its encoder part to encode the behavioral feature sequence, where the Transformer, through a self-attention mechanism, can directly model the dependency relationship between any two time steps, such as a customer opening an account 3 months ago (S1) → loan eligibility assessment 1 month ago (S2) → current loan application (S3). Then, a multi-head attention mechanism is used to assign learnable attention weights to different historical behaviors, identifying the impact of key behaviors on intent, such as anticipating a user's "cash withdrawal" based on "salary deposit". Then, positional encoding technology is used to add positional information to the behavioral features of each time step to preserve the temporal structure of the sequence, and a time decay weight mechanism is introduced to set the time decay weights. , where t represents the time interval between the operation and the current time, thus giving higher weight to the most recent business, and finally outputting the time-series feature vector T.
[0061] The process of obtaining a static feature vector based on the business transfer characteristics includes taking the customer's business transfer characteristics as input, that is, taking the customer's business operation sequence as input, using a graph neural network (GNN) to perform graph propagation on the temporal features, modeling the relationships between the temporal features, such as customer groups, business types, and behavioral transfer paths, to obtain the implicit associations between customers and between businesses, identify the transfer patterns between businesses, such as the transfer from "financial consulting" to "fund purchase", and because it captures the associations between different users and different businesses, the model has good generalization ability. This part of the network model outputs the transfer feature vector G.
[0062] The customer behavior features are obtained by fusing and concatenating the vectors corresponding to the static feature vector, the dynamic behavior features, and the business transfer features.
[0063] After obtaining the static feature vector S, the temporal feature vector T, and the transition feature vector G, they are further fused and spliced to obtain the fused customer behavior feature N, N=S⊕T⊕G.
[0064] S102. Input the customer behavior features into a pre-trained prediction network to identify customer intent.
[0065] The pre-trained prediction network is a classifier that can classify the fused features and output a customer intent map.
[0066] The pre-trained prediction network, during training, uses a loss function comprising contrastive learning loss, feature reconstruction loss, and classification loss. Contrastive learning loss ensures that behavioral and transfer features remain distinguishable after fusion; its loss function is given by Equation 1-1, where N is the number of samples. For real labels, Let T be the Euclidean distance between the feature vector T and the transition feature vector G, and m be the boundary value of the contrastive loss.
[0067] (1-1);
[0068] The feature reconstruction loss mainly requires that the fused features can reconstruct the original behavioral features and the transition features separately. Its loss function is given by Equation 1-2, where These are the static feature vector, the temporal feature vector, and the transition feature vector, respectively.
[0069] (1-2);
[0070] The classification loss uses cross-entropy loss to perform human classification on the fused features, thereby achieving customer behavior prediction. Its loss function is given by Equation 1-3, where... This represents the probability distribution output by the classifier.
[0071] (1-3);
[0072] The final total loss function is given by equation 1-4, where These are the weighting coefficients for each type of loss.
[0073] (1-4);
[0074] During training, customer behavior information is divided into training data and test data by setting a time window. The training data is used to train the model, while the test data is used to validate the model. In actual use, user feedback can be used to judge whether the current customer behavior prediction is accurate, thereby fine-tuning the model. Finally, a well-trained prediction network is obtained, which serves as the customer behavior prediction model.
[0075] S103. Construct a customer intent matrix for this visit based on the customer intent recognition results.
[0076] After obtaining the user's current business intent based on the pre-trained prediction network, the business intent is converted into a corresponding intent matrix, which is an n-dimensional vector. This includes the cash amount, type of service, whether VIP priority is required, and whether facial recognition verification is required.
[0077] S104. Generate a terminal resource matrix based on real-time changes in the network terminal resource information database.
[0078] Branch terminal resources refer to the functional attributes and service capabilities of various terminal devices deployed in a branch, including ATMs, smart teller machines, teller machines, and self-service terminals. These include service types, cash balances, number of cards that can be issued, hardware support types, current queue times, and VIP priority levels. Customers require corresponding terminal resource support when conducting business; for example, large cash withdrawals require sufficient cash balances and a normally operational branch, while facial recognition transactions require camera hardware support.
[0079] In this application, the attributes of each terminal are represented as an n-dimensional vector, forming a matrix. Where m is the number of terminals and n is the attribute dimension, which includes business type, cash balance, number of cards that can be opened, hardware support type, current queuing time, VIP priority level, etc.
[0080] S105. Match the intent matrix and the terminal resource matrix, and recommend the terminal with the highest score in the matching result to the customer for business processing.
[0081] Based on the intent matrix and the terminal resource matrix, combined with preset penalty and reward factors, the matching score of each terminal is calculated.
[0082] Specifically, matching scores are set for each dimension, based on whether terminal resources block customers from processing their business and whether the processing experience is good. A penalty factor of less than 1 is set for blocking customers' business or providing a poor customer experience, while a reward factor of more than 1 is set for accurately matching customer needs and ensuring smooth customer business processing.
[0083] The matching score for each device is calculated using cosine similarity, and the score is given by Equation 1-5:
[0084] (1-5);
[0085] Where M m×1 D represents the matching score for each device. i and C i These represent the values of the i-th dimension in the terminal resource matrix and the customer intent matrix, respectively. m is the number of terminals, and n is the attribute dimension. Finally, the matching scores of m terminals are obtained. The terminals are sorted according to the matching scores, and customers are guided to the terminal with the highest score to conduct business.
[0086] This application provides an intelligent customer flow management method for store locations. By using a customer behavior prediction model (a pre-trained prediction network), it directly outputs customer visit intent, reducing the workload of front-line staff and effectively alleviating queuing during peak hours. Through Transformer to capture dynamic sequence patterns, static features to supplement user profiles, and graph neural networks to strengthen business transfer relationship learning, it adaptively identifies the importance of different behaviors and strengthens inference on the logical relationships of historical behaviors, especially short-term behaviors, avoiding reliance solely on statistical rules, thereby achieving effective prediction of customer business intent.
[0087] In customer flow management scenarios, this application establishes a terminal resource monitoring mechanism to match customer business needs with terminal resources based on real-time terminal resource distribution, thereby rationally allocating the most suitable terminal equipment to customers, avoiding secondary queuing, and effectively improving the quality of customer service at the outlets.
[0088] The above are some specific implementations of a network intelligent traffic diversion method provided in the embodiments of this application. Based on this, this application also provides a corresponding device. The device provided in the embodiments of this application will be described below from the perspective of functional modularization.
[0089] Figure 2 This is a schematic diagram of the structure of a smart branch traffic diversion device provided in an embodiment of this application. (Combined with...) Figure 2 As shown in the figure, the intelligent branch diversion device 200 provided in this application embodiment includes:
[0090] The acquisition unit 210 is used to acquire basic information and historical behavior information of customers, and to obtain customer behavior characteristics based on the basic information and historical behavior information.
[0091] The identification unit 220 is used to input the customer behavior features into a pre-trained prediction network to identify customer intent.
[0092] The matrix generation unit 230 is used to construct the customer's intent matrix for this visit based on the customer intent recognition results; and to generate a terminal resource matrix based on the real-time changing data in the network terminal resource information database.
[0093] The matching unit 240 is used to match the intent matrix and the terminal resource matrix, and recommend the terminal with the highest score in the matching result to the customer for business processing.
[0094] In one implementation of this application, the acquisition unit obtains customer behavior characteristics based on the basic information and historical behavior information, including:
[0095] Extract the customer's static characteristics, dynamic behavioral characteristics, and business transfer characteristics from the customer's basic information and historical behavior information;
[0096] Based on the customer's static characteristics, dynamic behavioral characteristics, and business transfer characteristics, a vector corresponding to the static feature vector, the dynamic behavioral characteristics, and the business transfer characteristics is obtained;
[0097] The customer behavior features are obtained by fusing and concatenating the vectors corresponding to the static feature vector, the dynamic behavior features, and the business transfer features.
[0098] In one implementation of this application embodiment, the matching unit matches the intent matrix and the terminal resource matrix by:
[0099] Based on the intent matrix and the terminal resource matrix, combined with preset penalty and reward factors, the matching score of each terminal is calculated.
[0100] In one implementation of this application, the static features include the customer's age, occupation, asset size, and income level; the dynamic behavioral features are the customer's business transaction records, including time, type, amount, and transaction result; and the business transfer features are the transfer probability from business A to business B and the business association graph.
[0101] In one implementation of this application, the process of obtaining a static feature vector based on the static features includes:
[0102] Static features are input into a convolutional neural network, and features are extracted from the static features through convolutional layers and pooling layers to obtain static feature vectors.
[0103] In one implementation of this application, the process of obtaining a static feature vector based on the dynamic behavioral features includes:
[0104] Dynamic behavioral features are input into the Transformer architecture, and its encoder part is used to encode the behavioral feature sequence.
[0105] A multi-head attention mechanism is used to assign learnable attention weights to different historical behaviors, and to identify the impact of key behaviors on intent.
[0106] Positional encoding technology is used to add positional information to the behavioral features of each time step to preserve the temporal structure of the sequence. A time decay weighting mechanism is introduced to assign different weights to behaviors at different time steps based on the interval between the operation time and the current time, so that the most recent behavior contributes more to the feature vector, and the temporal feature vector is output.
[0107] In one implementation of this application, the process of obtaining a static feature vector based on the service transfer characteristics includes:
[0108] By taking the customer's business transfer characteristics as input, a graph neural network is used to perform graph propagation on the time-series features, model the relationships between the time-series features, obtain the implicit associations between customers and between businesses, identify the transfer patterns between businesses, and output the transfer feature vector.
[0109] This application also provides corresponding devices and computer storage media for implementing the solutions provided in this application.
[0110] The device includes a memory and a processor. The memory stores instructions or code, and the processor executes the instructions or code to cause the device to perform the method described in any embodiment of this application.
[0111] The computer storage medium stores code, and when the code is run, the device running the code implements the method described in any embodiment of this application.
[0112] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that all or part of the steps in the methods of the above embodiments can be implemented by means of software plus a general-purpose hardware platform. Based on this understanding, the technical solution of this application can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as a read-only memory (ROM) / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, a server, or a network communication device such as a router) to execute the methods described in various embodiments or some parts of the embodiments of this application.
[0113] It is understood that in the specific embodiments of this application, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved need to obtain user permission or consent when the above embodiments of this application are applied to specific products or technologies, and the collection, use and processing of related data need to comply with the relevant laws, regulations and standards of relevant countries and regions.
[0114] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0115] It should also be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for the device and apparatus embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments. The device and apparatus embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components indicated as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the solution in this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.
[0116] The above description is merely one specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A method for intelligent traffic diversion at service outlets, characterized in that, The method includes: Obtain basic customer information and historical behavior information, and obtain customer behavior characteristics based on the basic information and historical behavior information; The customer behavior characteristics are input into a pre-trained prediction network to identify customer intent. Construct a customer intent matrix for this visit based on the customer intent recognition results; A terminal resource matrix is generated based on real-time changes in the network terminal resource information database. The intent matrix and the terminal resource matrix are matched, and the terminal with the highest score in the matching results is recommended to the customer for business processing.
2. The method according to claim 1, characterized in that, The customer behavior characteristics obtained based on the aforementioned basic information and historical behavior information include: Extract the customer's static characteristics, dynamic behavioral characteristics, and business transfer characteristics from the customer's basic information and historical behavior information; Based on the customer's static characteristics, dynamic behavioral characteristics, and business transfer characteristics, a vector corresponding to the static feature vector, the dynamic behavioral characteristics, and the business transfer characteristics is obtained; The customer behavior features are obtained by fusing and concatenating the vectors corresponding to the static feature vector, the dynamic behavior features, and the business transfer features.
3. The method according to claim 1, characterized in that, The matching of the intent matrix and the terminal resource matrix includes: Based on the intent matrix and the terminal resource matrix, combined with preset penalty and reward factors, the matching score of each terminal is calculated.
4. The method according to claim 2, characterized in that, The static features include the customer's age, occupation, asset size, and income level; the dynamic behavioral features are the customer's business transaction records, including time, type, amount, and transaction result; the business transfer features are the transfer probability from business A to business B and the business association graph.
5. The method according to claim 2, characterized in that, The process of obtaining the static feature vector based on the static features includes: Static features are input into a convolutional neural network, and features are extracted from the static features through convolutional layers and pooling layers to obtain static feature vectors.
6. The method according to claim 2, characterized in that, The process of obtaining a static feature vector based on the dynamic behavioral features includes: Dynamic behavioral features are input into the Transformer architecture, and its encoder part is used to encode the behavioral feature sequence. A multi-head attention mechanism is used to assign learnable attention weights to different historical behaviors, and to identify the impact of key behaviors on intentions. Positional encoding technology is used to add positional information to the behavioral features of each time step to preserve the temporal structure of the sequence. A time decay weighting mechanism is introduced to assign different weights to behaviors at different time steps based on the interval between the operation time and the current time, so that the most recent behavior contributes more to the feature vector, and the temporal feature vector is output.
7. The method according to claim 2, characterized in that, The process of obtaining a static feature vector based on the aforementioned business transfer characteristics includes: By taking the customer's business transfer characteristics as input, a graph neural network is used to perform graph propagation on the time-series features, model the relationships between the time-series features, obtain the implicit associations between customers and between businesses, identify the transfer patterns between businesses, and output the transfer feature vector.
8. A smart branch traffic diversion device, characterized in that, The device includes: The acquisition unit is used to acquire basic customer information and historical behavior information, and to obtain customer behavior characteristics based on the basic information and historical behavior information. The identification unit is used to input the customer behavior features into a pre-trained prediction network to identify customer intent. The matrix generation unit is used to construct the customer's intent matrix for this visit based on the customer intent recognition results; and to generate a terminal resource matrix based on real-time changes in the network terminal resource information database. The matching unit is used to match the intent matrix and the terminal resource matrix, and recommend the terminal with the highest score in the matching results to the customer for business processing.
9. A computing device, characterized in that, The computing device includes: a memory and a processor; The memory is used to store computer programs; The processor is configured to implement the method as described in any one of claims 1 to 7 when executing the computer program.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.