Method for dynamically generating mobile banking activity set page

Abstract

The invention relates to a method for dynamically generating a mobile banking activity set page, which comprises the following steps of: constructing a tree structure in which a root node represents an activity set page and a first-level node represents a service category, and dynamically generating child nodes according to marketing activity data, a DIEN model comprising an embedded layer, an interest extraction layer, an interest evolution layer and a prediction layer is used to predict the interestingness of the user to different marketing activities, and user interest prediction is realized; according to the method, the user geographic position, preference setting and interest prediction value are integrated, child node weights are calculated, a descending sorting algorithm is used for sorting the child node weights, activity set pages are dynamically arranged according to the sorting result, and personalized recommendation and efficient marketing are achieved. Through modular design and real-time dynamic updating, the system efficiency and the user experience are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and mainly to a method for dynamically generating a mobile banking activity collection page. Background Art

[0002] With the rapid development of mobile Internet, mobile banking has become an important channel for banking services. In order to attract customers and improve user experience, banks need to frequently launch various marketing activities and quickly adjust the display content according to market changes and customer needs. However, most mobile banking activities currently require a relatively cumbersome development, testing and launch process during the adjustment and launch of the activity page, which takes a long time, usually a month or even longer.

[0003] Most of the mobile banking page layouts in the prior art rely on static configuration, that is, each time the page layout or content is adjusted, it needs to be redeveloped and released. This method is not only time-consuming, but also has a great restriction on the flexibility and response speed of the business. For example, a Chinese patent with publication number CN116501434A discloses a method and system for automatic switching of mobile banking themes based on a decision tree, which belongs to the field of data processing technology in the financial industry. The method includes: parameter management of control parameters required for automatic switching of mobile banking themes through the server; the control parameters include resident area information, interface version, maintenance organization, parameter validity time and parameter maintainer; parameter management includes parameter query, parameter modification, parameter maintenance, parameter approval and control parameter historical version backup management; a mobile banking theme automatic switching model is established based on sample data through the theme modeling system of the server, and the mobile banking theme automatic switching model is sent to the client for mobile banking theme determination and then the mobile banking theme is automatically switched. The present invention can automatically select an adapted version based on regional information, device information, customer age and other information, and control and realize the automatic switching of personalized interface and version functions of the mobile banking client. However, the above patent mainly relies on server-side parameter management and a theme switching model established on the server side. The client passively receives the model to switch themes, which cannot ensure that the displayed page always meets the user's current needs. In addition, the system design and processing methods are relatively traditional, and the system efficiency improvement is limited. Therefore, there is an urgent need for a method for dynamically generating mobile banking activity collection pages that can effectively improve user participation and system efficiency. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, the present application provides a method for dynamically generating a mobile banking activity collection page.

[0005] The technical solution of this application is as follows:

[0006] A method for dynamically generating a mobile banking activity collection page, the method comprising:

[0007] Constructing a tree structure corresponding to the mobile banking activity collection page, the tree structure comprising a root node and a plurality of first-level nodes, wherein the root node represents a page corresponding to the mobile banking activity collection page, and the first-level node represents a business category;

[0008] Obtain the current marketing activity data of mobile banking and dynamically generate sub-nodes corresponding to different marketing activities;

[0009] Collecting user behavior data, geographic location and preference settings in mobile banking, building a user interest prediction model, using the behavior data to train the user interest prediction model, obtaining a trained user interest prediction model, and using the trained user interest prediction model to predict the user's interest in different marketing activities;

[0010] The weight of each sub-node is calculated based on the user's geographic location, preference settings and interest in different marketing activities, and the weights of the sub-nodes are sorted using a weight sorting algorithm to obtain the sorting results. The mobile banking activity collection page is dynamically laid out based on the sorting results.

[0011] As a preferred embodiment of the present invention, obtaining the current marketing activity data of mobile banking and dynamically generating sub-nodes corresponding to different marketing activities are specifically as follows:

[0012] A marketing campaign number, marketing campaign name, marketing campaign icon, marketing campaign link and marketing campaign sequence are obtained according to the marketing campaign data, wherein the marketing campaign number is a unique identifier, and a new child node is generated under the corresponding first-level node according to the business category to which the marketing campaign type belongs, and the child node includes a node number, a node name, a leaf node flag, a parent node number, a node type, a picture name, a jump link and a serial number, wherein: the node number corresponds to the marketing campaign number; the node name corresponds to the marketing campaign name; the leaf node flag is specifically used to determine whether the current node is a child node, if the current node is a child node, the leaf node flag is displayed as yes, if the current node is not a child node, the leaf node flag is displayed as no; the parent node number is the corresponding first-level node number; the node type corresponds to the marketing campaign type; the picture name corresponds to the marketing campaign icon; the jump link corresponds to the marketing campaign link; and the serial number corresponds to the marketing campaign sequence.

[0013] As a preferred embodiment of the present invention, the user interest prediction model is constructed based on the DIEN model, including an embedding layer, an interest extraction layer, an interest evolution layer and a prediction layer, wherein the behavior data is input as input data into the embedding layer, the behavior data includes user ID, product ID and behavior type, and the embedding layer outputs the user embedding vector e user , product embedding vector e item and the behavior type embedding vector eaction The product embedding vector and the behavior type embedding vector are concatenated to form the behavior sequence embedding vector e behavior , expressed as:

[0014] e behavior =[e item ;e action ];

[0015] Embed the behavior sequence into vector e behavior As the input data of the interest extraction layer, it outputs the hidden state sequence;

[0016] The user embedding vector output by the embedding layer and the hidden state sequence output by the interest extraction layer are used as the input data of the interest evolution layer, and the interest vector v is output;

[0017] The interest vector v is used as the input data of the prediction layer, and the predicted value of the user's interest in different marketing activities is output.

[0018] As a preferred embodiment of the present invention, the interest extraction layer is a GRU structure, including an update gate and a reset gate, wherein:

[0019] The reset gate is used to control the impact of historical information on the current time step, which is expressed as:

[0020] r t =σ(W r h t-1 +U r e behavior +b r );

[0021] In the formula, r t is the output of the reset gate, σ is the sigmoid activation function, W r h is the weight matrix that affects the historical information in the reset gate; t-1 is the hidden state of the previous time step, U r b is the weight matrix that affects the current input information in the reset gate; b r is the bias vector for resetting the gate;

[0022] The update gate is used to control the contribution of the candidate hidden state of the current time step to the final hidden state, which is expressed as:

[0023] z t =σ(W z h t-1 +U z e behavior +b z );

[0024] In the formula, z t is the output of the update gate, Wz is the weight matrix that affects the historical information in the update gate; U z is the weight matrix that affects the current input information in the update gate; b z is the bias vector of the update gate;

[0025] Among them, the hidden state is expressed as:

[0026] h t =(1-z t )⊙h t-1 +z t ⊙h~ t ;

[0027] h~ t =tanh(W h (r t ⊙h t-1 )+U h e behavior +b h );

[0028] In the formula, h t is the hidden state of the current time step, ⊙ is the XOR operation, h~ t is the candidate hidden state of the current time step; tanh() is the hyperbolic tangent activation function; W h is the weight matrix that affects the historical information in the candidate hidden state; U h is the weight matrix that affects the current input information in the candidate hidden state.

[0029] As a preferred implementation manner of the present invention, the specific steps of the interest evolution layer outputting the interest vector are:

[0030] The user embedding vector is used as the query vector and the hidden state sequence is used as the key vector. The similarity score between the query vector and the key vector is calculated, which can be expressed as:

[0031]

[0032] In the formula, s i is the similarity score between the i-th element of the query vector and the i-th element of the key vector;

[0033] The attention weight is obtained by normalizing the similarity score through the softmax function, which is expressed as:

[0034]

[0035] In the formula, a i is the attention weight corresponding to the i-th hidden state, and j is the element index of the traversal similarity score;

[0036] The interest vector v is obtained by weighted summing of the key vectors through the attention weights, which can be expressed as:

[0037]

[0038] As a preferred embodiment of the present invention, the prediction layer includes an input layer, a fully connected layer and an output layer. The input layer uses the interest vector output by the interest evolution layer as input data. The fully connected layer has two layers, each with 128 neurons, and the activation function is ReLU, wherein the first layer is expressed as:

[0039] z'1 = ReLU(W1v+b1);

[0040] Where z'1 is the output of the first fully connected layer, W1 is the weight matrix of the first fully connected layer, and b1 is the bias vector of the first fully connected layer;

[0041] The second layer is expressed in the formula:

[0042] z'2 = ReLU(W2z1+b2);

[0043] Where z'2 is the output of the second fully connected layer, W2 is the weight matrix of the second fully connected layer, and b2 is the bias vector of the second fully connected layer;

[0044] The output layer is a linear layer, which outputs the predicted value of the user's interest in different marketing activities, expressed as:

[0045]

[0046] In the formula, is the output of the output layer, W3 is the weight matrix of the output layer, and b3 is the bias vector of the output layer.

[0047] As a preferred embodiment of the present invention, the weight of each sub-node is calculated according to the user's geographical location, preference settings and interest in different marketing activities as follows:

[0048] Check if the user's geolocation matches the activity's geolocation. If so, the geolocation weight is 1; otherwise, it is 0.

[0049] Check if the user's preferences match the activity's category. If so, the location preference weight is 1; otherwise, it is 0.

[0050] The total weight of each child node to the user is calculated based on the geographic location weight, preference setting weight, and interest in different marketing activities, expressed as:

[0051] Total weight = geographic location weight * preference setting weight * user's interest in different marketing activities.

[0052] As a preferred embodiment of the present invention, the weight sorting algorithm is a descending sorting algorithm. After sorting, the child node with the highest total weight will be placed at the front, and the child node with the lowest total weight will be placed at the end to obtain a sorting result.

[0053] The present invention also provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a method for dynamically generating a mobile banking activity collection page as described in any embodiment of the present invention is implemented.

[0054] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method for dynamically generating a mobile banking activity collection page as described in any embodiment of the present invention is implemented.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] 1) The present invention provides a method for dynamically generating a mobile banking activity collection page, which accurately predicts user interests through user behavior data, geographic location and preference settings, realizes personalized recommendations, and improves user participation and satisfaction; and dynamically updates the activity collection page according to the user's real-time behavior and geographic location changes to ensure that the displayed activities always meet the user's current needs;

[0057] 2) The present invention provides a method for dynamically generating a mobile banking activity collection page. The modular design of the tree structure makes the addition and management of activities more flexible, which is convenient for system maintenance and expansion; the optimization of the user interest prediction model and the weight calculation algorithm improves the system's ability and efficiency in processing large amounts of data, ensuring page loading speed and user experience;

[0058] 3) The present invention provides a method for dynamically generating a mobile banking activity collection page, which uses a sorting algorithm to rank the activities that users are most interested in in the most prominent position, thereby increasing the click-through rate and conversion rate of the activities and effectively improving the marketing effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a method flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0060] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0061] The present invention provides the following technical solution: a method for dynamically generating a mobile banking activity collection page.

[0062] Embodiment 1:

[0063] This embodiment provides a method for dynamically generating a mobile banking activity collection page, the method comprising:

[0064] S1. Construct a tree structure corresponding to the mobile banking activity collection page, wherein the tree structure includes a root node and multiple first-level nodes, wherein the root node represents a page corresponding to the mobile banking activity collection page, and the first-level node represents a business category;

[0065] Preferably, the root node and the first-level node are fixed;

[0066] Each first-level node represents a business category. In this implementation, the first-level nodes include the following categories: loan services, deposit products, investment and financial management, credit card services, and life payments;

[0067] Through the above design, different marketing activities can be classified and organized according to business categories, which is convenient for subsequent dynamic generation and display;

[0068] S2. Obtain the current marketing activity data of mobile banking and dynamically generate sub-nodes corresponding to different marketing activities;

[0069] Acquire the marketing activity data currently being carried out from the mobile banking database or other data sources, and dynamically generate a sub-node corresponding to each marketing activity according to the marketing activity data;

[0070] In this embodiment, a marketing campaign number, a marketing campaign name, a marketing campaign icon, a marketing campaign link and a marketing campaign sequence are obtained according to the marketing campaign data. The marketing campaign link jump interface corresponds to the start and end time of the activity, the conditions for participation and the promotion method. The marketing campaign number is a unique identifier. According to the business category to which the marketing campaign type belongs, a new child node is generated under the corresponding first-level node. The child node includes a node number, a node name, a leaf node mark, a parent node number, a node type, a picture name, a jump link and a sequence number. Specifically: the node number corresponds to the marketing campaign number; the node name corresponds to the marketing campaign name; the leaf node mark is specifically used to determine whether the current node is a child node. If the current node is a child node, the leaf node mark is displayed as yes, and if the current node is not a child node, the leaf node mark is displayed as no; the parent node number is the corresponding first-level node number; the node type corresponds to the marketing campaign type; the picture name corresponds to the marketing campaign icon; the jump link corresponds to the marketing campaign link; and the sequence number corresponds to the marketing campaign sequence;

[0071] Preferably, in order to ensure the consistency and integrity of the tree structure, the root node and the first-level node also include the elements included in the child nodes, that is, the root node and the first-level node also include the node number, node name, leaf node mark, parent node number, node type, picture name, jump link and serial number, the difference is that, among the elements included in the root node and the first-level node, the node number of the root node is 0, and there is no parent node number, picture name and jump link, and the serial number is 0; the node number of the first-level node increases from 1, the node name corresponds to the business category name, the parent node number is the root node number 0, the picture name corresponds to the business category icon, there is no jump link, and the serial number increases from 1;

[0072] S3. Collecting the user's behavior data, geographic location and preference settings in mobile banking, including browsing history and transaction data; constructing a user interest prediction model, using the behavior data to train the user interest prediction model, obtaining a trained user interest prediction model, and using the trained user interest prediction model to predict the user's interest in different marketing activities;

[0073] S31, the behavior data includes the user's browsing history, transaction records, click behavior and dwell time, and the user ID, product ID, behavior type and timestamp are obtained according to the user behavior data; the geographic location includes the user's current location or frequently visited places, which helps to understand the user's geographic preferences; the preference settings include the personalized preferences set by the user in the mobile banking, including the business type of interest, risk preference and preferential activity subscription;

[0074] S32, user interest prediction model;

[0075] The user interest prediction model is constructed based on the DIEN model, including an embedding layer, an interest extraction layer, an interest evolution layer and a prediction layer, wherein:

[0076] S321, embedding layer;

[0077] The embedding layer includes multiple embedding matrices, each embedding matrix corresponds to a discrete feature. In this embodiment, the embedding layer includes three embedding matrices with an embedding dimension of 64. The discrete features include user ID, product ID and behavior type.

[0078] The discrete features are input into the embedding layer and the user embedding vector e is output. user , product embedding vector e item and the behavior type embedding vector e action The product embedding vector and the behavior type embedding vector are concatenated to form the behavior sequence embedding vector e behavior , expressed as:

[0079] e behavior =[eitem ;e action ];

[0080] S322, interest extraction layer;

[0081] Embed the behavior sequence into vector e behavior As the input data of the interest extraction layer, the interest extraction layer is a GRU structure, including an update gate and a reset gate, where:

[0082] The reset gate is used to control the impact of historical information on the current time step, which is expressed as:

[0083] r t =σ(W r h t-1 +U r e behavior +b r );

[0084] In the formula, r t is the output of the reset gate, σ is the sigmoid activation function, W r h is the weight matrix that affects the historical information in the reset gate; t-1 is the hidden state of the previous time step, U r b is the weight matrix that affects the current input information in the reset gate; b r is the bias vector for resetting the gate;

[0085] The update gate is used to control the contribution of the candidate hidden state of the current time step to the final hidden state, which is expressed as:

[0086] z t =σ(W z h t-1 +U z e behavior +b z );

[0087] In the formula, z t is the output of the update gate, W z is the weight matrix that affects the historical information in the update gate; U z is the weight matrix that affects the current input information in the update gate; b z is the bias vector of the update gate;

[0088] Among them, the hidden state is expressed as:

[0089] h t =(1-z t )⊙h t-1 +z t ⊙h~ t ;

[0090]

[0091] In the formula, h t is the hidden state of the current time step, ⊙ is the XOR operation, is the candidate hidden state of the current time step; tanh() is the hyperbolic tangent activation function; W h is the weight matrix that affects the historical information in the candidate hidden state; U h is the weight matrix that affects the current input information in the candidate hidden state;

[0092] S323, interest evolution layer;

[0093] The user embedding vector output by the embedding layer and the hidden state sequence output by the interest extraction layer are used as the input data of the interest evolution layer. The user embedding vector is used as the query vector, and the hidden state sequence is used as the key vector. The similarity score between the query vector and the key vector is calculated, which can be expressed as:

[0094]

[0095] In the formula, s i is the similarity score between the i-th element of the query vector and the i-th element of the key vector;

[0096] The attention weight is obtained by normalizing the similarity score through the softmax function, which is expressed as:

[0097]

[0098] In the formula, a i is the attention weight corresponding to the i-th hidden state, and j is the element index of the traversal similarity score;

[0099] The interest vector v is obtained by weighted summing of the key vectors through the attention weights, which can be expressed as:

[0100]

[0101] S324, prediction layer;

[0102] In this embodiment, the prediction layer includes an input layer, a fully connected layer and an output layer. The input layer uses the interest vector output by the interest evolution layer as input data. The fully connected layer has two layers, each with 128 neurons, and the activation function is ReLU, wherein the first layer is expressed as:

[0103] z'1 = ReLU(W1v+b1);

[0104] Where z'1 is the output of the first fully connected layer, W1 is the weight matrix of the first fully connected layer, and b1 is the bias vector of the first fully connected layer;

[0105] The second layer is expressed in the formula:

[0106] z'2 = ReLU(W2z1+b2);

[0107] Where z'2 is the output of the second fully connected layer, W2 is the weight matrix of the second fully connected layer, and b2 is the bias vector of the second fully connected layer;

[0108] The output layer is a linear layer, which outputs the predicted value of the user's interest in different marketing activities, expressed as:

[0109]

[0110] In the formula, is the output of the output layer, W3 is the weight matrix of the output layer, and b3 is the bias vector of the output layer;

[0111] S4. Calculate the weight of each sub-node based on the user's geographic location, preference settings, and interest in different marketing activities, specifically:

[0112] Check if the user's geolocation matches the activity's geolocation. If so, the geolocation weight is 1; otherwise, it is 0.

[0113] Check if the user's preferences match the activity's category. If so, the location preference weight is 1; otherwise, it is 0.

[0114] The total weight of each child node to the user is calculated based on the geographic location weight, preference setting weight, and interest in different marketing activities, expressed as:

[0115] Total weight = geographic location weight * preference setting weight * user's interest in different marketing activities;

[0116] S5. Sort the weights of the child nodes by a weight sorting algorithm. The weight sorting algorithm described in this embodiment is a descending sorting algorithm. After sorting, the child node with the highest total weight will be placed at the front, and the child node with the lowest total weight will be placed at the end, to obtain a sorting result. The mobile banking activity collection page is dynamically laid out according to the sorting result. The child node with the front ranking result has a larger proportion in the mobile banking activity collection page and is displayed at a front position.

[0117] Embodiment 2:

[0118] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, a method for dynamically generating a mobile banking activity collection page as described in Embodiment 1 is implemented.

[0119] Embodiment 3:

[0120] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method for dynamically generating a mobile banking activity collection page as described in Embodiment 1 of the present invention is implemented.

[0121] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for dynamically generating a mobile banking activity collection page, characterized in that: The method comprises: Constructing a tree structure corresponding to the mobile banking activity collection page, the tree structure comprising a root node and a plurality of first-level nodes, wherein the root node represents a page corresponding to the mobile banking activity collection page, and the first-level node represents a business category; Obtain the current marketing activity data of mobile banking and dynamically generate sub-nodes corresponding to different marketing activities; Collecting user behavior data, geographic location and preference settings in mobile banking, building a user interest prediction model, using the behavior data to train the user interest prediction model, obtaining a trained user interest prediction model, and using the trained user interest prediction model to predict the user's interest in different marketing activities; The weight of each sub-node is calculated based on the user's geographic location, preference settings and interest in different marketing activities, and the weights of the sub-nodes are sorted using a weight sorting algorithm to obtain the sorting results. The mobile banking activity collection page is dynamically laid out based on the sorting results.

2. A method for dynamically generating a mobile banking activity collection page according to claim 1, characterized in that: Get the current marketing activity data of mobile banking and dynamically generate sub-nodes corresponding to different marketing activities: A marketing campaign number, marketing campaign name, marketing campaign icon, marketing campaign link and marketing campaign sequence are obtained according to the marketing campaign data, wherein the marketing campaign number is a unique identifier, and a new child node is generated under the corresponding first-level node according to the business category to which the marketing campaign type belongs, and the child node includes a node number, a node name, a leaf node flag, a parent node number, a node type, a picture name, a jump link and a serial number, wherein: the node number corresponds to the marketing campaign number; the node name corresponds to the marketing campaign name; the leaf node flag is specifically used to determine whether the current node is a child node, if the current node is a child node, the leaf node flag is displayed as yes, if the current node is not a child node, the leaf node flag is displayed as no; the parent node number is the corresponding first-level node number; the node type corresponds to the marketing campaign type; the picture name corresponds to the marketing campaign icon; the jump link corresponds to the marketing campaign link; and the serial number corresponds to the marketing campaign sequence.

3. A method for dynamically generating a mobile banking activity collection page according to claim 2, characterized in that: The user interest prediction model is built based on the DIEN model, including an embedding layer, an interest extraction layer, an interest evolution layer, and a prediction layer. The behavior data is input into the embedding layer as input data. The behavior data includes user ID, product ID, and behavior type. The embedding layer outputs the user embedding vector e user , product embedding vector e item and the behavior type embedding vector e action The product embedding vector and the behavior type embedding vector are concatenated to form the behavior sequence embedding vector e behavior , expressed as: And behavior =[and item ;And action ]; Embed the behavior sequence into vector e behavior As the input data of the interest extraction layer, it outputs the hidden state sequence; The user embedding vector output by the embedding layer and the hidden state sequence output by the interest extraction layer are used as the input data of the interest evolution layer, and the interest vector v is output; The interest vector v is used as the input data of the prediction layer, and the predicted value of the user's interest in different marketing activities is output.

4. A method for dynamically generating a mobile banking activity collection page according to claim 3, characterized in that: The interest extraction layer is a GRU structure, including an update gate and a reset gate, where: The reset gate is used to control the impact of historical information on the current time step, which is expressed as: r t σ(W r h t-1 +U r e behavior +b r )4 In the formula, r t is the output of the reset gate, σ is the sigmoid activation function, W r h is the weight matrix that affects the historical information in the reset gate; t-1 is the hidden state of the previous time step, U r b is the weight matrix that affects the current input information in the reset gate; b r is the bias vector for resetting the gate; The update gate is used to control the contribution of the candidate hidden state of the current time step to the final hidden state, which is expressed as: z t =σ(W z h t-1 +U z e behavior +b z ); In the formula, z t is the output of the update gate, W z is the weight matrix that affects the historical information in the update gate; U z is the weight matrix that affects the current input information in the update gate; b z is the bias vector of the update gate; Among them, the hidden state is expressed as: In the formula, h t is the hidden state of the current time step, ⊙ is the XOR operation, is the candidate hidden state of the current time step; tanh() is the hyperbolic tangent activation function; W h is the weight matrix that affects the historical information in the candidate hidden state; U h is the weight matrix that affects the current input information in the candidate hidden state.

5. A method for dynamically generating a mobile banking activity collection page according to claim 3, characterized in that: The specific steps of the interest evolution layer outputting the interest vector are: The user embedding vector is used as the query vector and the hidden state sequence is used as the key vector. The similarity score between the query vector and the key vector is calculated, which can be expressed as: In the formula, s i is the similarity score between the i-th element of the query vector and the i-th element of the key vector; The attention weight is obtained by normalizing the similarity score through the softmax function, which is expressed as: In the formula, a i is the attention weight corresponding to the i-th hidden state, and j is the element index of the traversal similarity score; The interest vector v is obtained by weighted summing of the key vectors through the attention weights, which can be expressed as:

6. A method for dynamically generating a mobile banking activity collection page according to claim 3, characterized in that: The prediction layer includes an input layer, a fully connected layer and an output layer. The input layer uses the interest vector output by the interest evolution layer as input data. The fully connected layer has two layers, each with 128 neurons, and the activation function is ReLU, wherein the first layer is expressed as: z'1 = ReLU(W1v+b1); Where z'1 is the output of the first fully connected layer, W1 is the weight matrix of the first fully connected layer, and b1 is the bias vector of the first fully connected layer; The second layer is expressed in the formula: z'2 = ReLU(W2z1+b2); Where z'2 is the output of the second fully connected layer, W2 is the weight matrix of the second fully connected layer, and b2 is the bias vector of the second fully connected layer; The output layer is a linear layer, which outputs the predicted value of the user's interest in different marketing activities, expressed as: In the formula, is the output of the output layer, W3 is the weight matrix of the output layer, and b3 is the bias vector of the output layer.

7. A method for dynamically generating a mobile banking activity collection page according to claim 3, characterized in that: The weight of each sub-node is calculated based on the user's geographic location, preferences, and interest in different marketing activities: Check if the user's geolocation matches the activity's geolocation. If so, the geolocation weight is 1; otherwise, it is 0. Check if the user's preferences match the activity's category. If so, the location preference is weighted as 1. Otherwise, 0; The total weight of each child node to the user is calculated based on the geographic location weight, preference setting weight, and interest in different marketing activities, expressed as: Total weight = geographic location weight * preference setting weight * user's interest in different marketing activities.

8. A method for dynamically generating a mobile banking activity collection page according to claim 7, characterized in that: The weight sorting algorithm is a descending sorting algorithm. After sorting, the child node with the highest total weight will be placed at the front, and the child node with the lowest total weight will be placed at the end to obtain the sorting result.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method for dynamically generating a mobile banking activity collection page as described in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, a method for dynamically generating a mobile banking activity collection page as described in any one of claims 1 to 8 is implemented.

Images