Mobile banking activity recommendation method and device, equipment and storage medium

Abstract

The invention discloses an activity recommendation method and device for a mobile banking, equipment and a storage medium. The method comprises the steps of obtaining a user entity, an activity entity and a scene entity related to the mobile banking; constructing a knowledge graph based on the user entities, the activity entities and the scene entities; based on the knowledge graph, calculating three-party association strength of the user entity, the activity entity and the scene entity, and determining the three-party association strength as target association strength when the three-party association strength is greater than or equal to a preset association strength threshold; the target association strength is input into an activity recommendation model, an initial activity strategy is generated, and the activity recommendation model is obtained through pre-training in a federal learning mode; and optimizing the initial activity strategy based on a preset reinforcement learning model to generate a target activity strategy. In this way, personalized activity recommendation can be provided for the user, and user experience and recommendation pertinence are effectively improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method, apparatus, device, and storage medium for recommending activities in mobile banking. Background Art

[0002] In the mobile finance sector, mobile banking has become a core channel for users to access financial services. With the widespread adoption of smartphones and the rapid development of mobile internet technology, users are increasingly relying on mobile banking for a variety of financial operations, including account management, fund transfers, purchasing financial products, and loan applications. Through mobile banking, financial institutions can deliver the latest product information and promotions in real time, meeting users' diverse financial needs and improving user loyalty and satisfaction.

[0003] Currently, mobile banking promotion recommendations often use a "fixed window" model similar to traditional web pages, where developers pre-set the image link, title, and carousel order of the front-end promotion page through code. However, with the increasing diversification of user behavior and the continuous expansion of financial products, traditional promotion recommendation methods are no longer able to provide users with personalized and precise service needs, thus reducing the user experience. Summary of the Invention

[0004] The embodiments of the present application provide a method, apparatus, device, and storage medium for recommending activities for mobile banking, which can provide users with personalized activity recommendations, effectively improving user experience and the pertinence of recommendations.

[0005] In a first aspect, an embodiment of the present application provides a method for recommending activities in a mobile banking application, the method comprising:

[0006] Get user entities, activity entities, and scenario entities related to mobile banking;

[0007] Building a knowledge graph based on the user entity, activity entity and scenario entity;

[0008] Based on the knowledge graph, calculating the three-party association strength of the user entity, the activity entity, and the scene entity, and determining the three-party association strength as a target association strength when the three-party association strength is greater than or equal to a preset association strength threshold;

[0009] Inputting the target association strength into an activity recommendation model to generate an initial activity strategy, wherein the activity recommendation model is pre-trained through federated learning;

[0010] The initial activity strategy is optimized based on a preset reinforcement learning model to generate a target activity strategy.

[0011] Optionally, obtaining the user entity, activity entity, and scene entity includes:

[0012] Obtaining user location data, user behavior data, bank-side data, merchant-side data, and external scenario data related to the mobile banking;

[0013] The user entity is constructed based on the user location data and the user behavior data; the activity entity is constructed based on the bank-side data and the merchant-side data; and the scenario entity is constructed based on the external scenario data.

[0014] Optionally, the calculating, based on the knowledge graph, the tripartite association strength among the user entity, the activity entity, and the scenario entity includes:

[0015] Based on the knowledge graph, calculating the association strength between the user entity and the activity entity as a first association strength;

[0016] calculating the association strength between the scene entity and the activity entity as a second association strength;

[0017] multiplying the first association strength by the second association strength as a third association strength;

[0018] The tripartite association strength is obtained by performing a weighted sum calculation based on the first association strength, the second association strength, and the third association strength.

[0019] Optionally, the optimizing the initial activity strategy based on a preset reinforcement learning model to generate a target activity strategy includes:

[0020] Constructing a reward function of the reinforcement learning model based on the user behavior data;

[0021] Based on the reward function, the reinforcement learning model is iteratively trained to optimize the initial activity policy and generate a target activity policy.

[0022] Optionally, the activity recommendation model is trained through the following steps:

[0023] Acquire input data related to the mobile banking, the input data including user history profile data and merchant history activity participation data, the user history profile data including user history location data and user history behavior data, the user history behavior data including user history activity click volume and user history activity search volume;

[0024] Preprocessing the input data to obtain target input data;

[0025] Desensitizing the target input data to obtain target desensitized data;

[0026] The target desensitized data is input into the model to be trained, and cross-institutional collaborative training is performed based on multiple distributed clients of the mobile banking. After the training is completed, the activity recommendation model is obtained.

[0027] Optionally, preprocessing the input data to obtain target input data includes:

[0028] Convert the user's historical location data into a region label including province, city, district, and business district to obtain the user's historical region label;

[0029] The user historical activity heat value is calculated based on the user historical activity click volume and the user historical activity search volume.

[0030] Optionally, the cross-institutional collaborative training based on multiple distributed clients includes:

[0031] The multiple distributed clients perform cross-institutional collaborative training on the model to be trained based on local data, wherein the multiple distributed clients include a bank side and a merchant side, the local data of the bank side is the user's historical regional labels and the user's historical activity heat values ​​that have been desensitized, and the local data of the merchant side is the merchant's historical activity participation data that has been desensitized.

[0032] In a second aspect, an embodiment of the present application provides an activity recommendation device for mobile banking, comprising:

[0033] Entity acquisition module, used to obtain user entities, activity entities and scenario entities related to mobile banking;

[0034] A graph construction module, configured to construct a knowledge graph based on the user entity, activity entity, and scenario entity;

[0035] an association strength calculation module, configured to calculate, based on the knowledge graph, the tripartite association strength of the user entity, the activity entity, and the scenario entity, and determine the tripartite association strength as a target association strength when the tripartite association strength is greater than or equal to a preset association strength threshold;

[0036] a strategy generation module, configured to input the target association strength into an activity recommendation model to generate an initial activity strategy, wherein the activity recommendation model is pre-trained through federated learning;

[0037] The strategy optimization module is used to optimize the initial activity strategy based on a preset reinforcement learning model to generate a target activity strategy.

[0038] In a third aspect, an embodiment of the present application provides an electronic device, the device comprising: a processor, a memory, and a system bus;

[0039] The processor and the memory are connected via the system bus;

[0040] The memory is used to store a program, wherein the program includes instructions, and when the instructions are executed by the processor, the processor executes any one of the implementation steps of the above-mentioned mobile banking activity recommendation method.

[0041] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which is used to store a computer program. When the computer program is executed by a terminal device, it implements any implementation step of the above-mentioned mobile banking activity recommendation method.

[0042] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0043] In an embodiment of the present application, user entities, activity entities, and scenario entities related to mobile banking are first obtained, and then a knowledge graph is constructed based on the obtained user entities, activity entities, and scenario entities. Based on the knowledge graph, the three-party association strength of the user entity, activity entity, and scenario entity is calculated, and when the three-party association strength is greater than or equal to the preset association strength threshold, the three-party association strength is determined to be the target association strength. Subsequently, the target association strength is input into the preset activity recommendation model to generate an initial activity strategy, and finally, the initial activity strategy is optimized based on the preset reinforcement learning model to generate a target activity recommendation strategy. It can be seen that this solution constructs the obtained user entity, activity entity, and scenario entity into a knowledge graph, which can mine the semantic association relationship between entities from multiple dimensions. On this basis, by calculating the three-party association strength of user entity, activity entity and scenario entity, when the three-party association strength is greater than or equal to the preset association strength threshold, the target association strength is determined, and the target association strength number is input into the preset activity recommendation model to generate a personalized activity strategy, and further combine the reinforcement learning mechanism to optimize the preliminary activity strategy to obtain the target activity strategy, thereby effectively improving the personalization and accuracy of mobile banking activity recommendations, and enhancing the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A flowchart of a mobile banking activity recommendation method provided in an embodiment of the present application;

[0045] Figure 2 This is a structural diagram of an activity recommendation device for mobile banking provided in an embodiment of the present application. DETAILED DESCRIPTION

[0046] Before starting with the embodiments of the present application, in order to facilitate understanding of the technical solution of the present application, the professional terms involved in the embodiments of the present application are first introduced in detail.

[0047] Federated learning: A distributed machine learning framework that enables multiple distributed clients to conduct cross-institutional collaborative training without sharing original data.

[0048] Knowledge graph: A structured data set that graphically displays entities and their relationships.

[0049] Reinforcement learning: A machine learning method that uses a reward function as feedback for a reinforcement learning model, continuously learning and optimizing activity recommendation strategies in actual application scenarios.

[0050] Sliding window algorithm: A technique for efficiently processing linear data structures such as strings and numbers. In data processing, a fixed-size window is set and the algorithm slides the window across the data sequence to process and analyze the data in segments.

[0051] Homomorphic Encryption (HE): is a public-key encryption algorithm that supports additive homomorphism, allowing specific mathematical operations to be performed on encrypted ciphertext. The decrypted results of these operations are consistent with the results of the same operations on the plaintext.

[0052] Secure Hash Algorithm 256-bit (SHA-256): is a cryptographic hash function whose core function is to generate a fixed-length 256-bit hash value through specific data operations on input data of any length.

[0053] Secure Multi-Party Computation (MPC): A cryptographic technique that allows multiple clients to perform collaborative computations without disclosing their input data.

[0054] As mentioned above, current mobile banking promotional recommendations still generally adopt a "fixed window" model similar to traditional web pages. The image links, titles, and carousel order of the front-end promotional page are all pre-set by developers through code. However, this promotional recommendation model fails to integrate with external scenarios. According to relevant data, a joint-stock bank's promotional conversion rate was 30% lower than the industry average for scenario-based recommendations due to its failure to implement scenario-based recommendations. Therefore, with the increasing diversification of user behavior and the continuous expansion of financial products, the traditional promotional recommendation model has become unable to provide users with personalized and precise service needs, thereby reducing the user experience.

[0055] Based on this, in order to solve the above problems, an embodiment of the present application provides an activity recommendation method for mobile banking, which obtains user entities, activity entities and scenario entities related to mobile banking, and constructs a knowledge graph based on the obtained user entities, activity entities and scenario entities. Then, based on the knowledge graph, the three-party association strength of the user entity, activity entity and scenario entity is calculated, and when the three-party association strength is greater than or equal to the preset association strength threshold, the three-party association strength is determined to be the target association strength. Subsequently, the target association strength is input into the preset activity recommendation model to generate an initial activity strategy. Finally, the initial activity strategy is optimized based on the preset reinforcement learning model to generate a target activity recommendation strategy.

[0056] As can be seen, this solution constructs the acquired user entities, activity entities, and scenario entities into a knowledge graph, fully exploring the semantic associations between activity entities and scenario entities. Furthermore, by calculating the tripartite association strength among user entities, activity entities, and scenario entities, a target association strength is determined when the tripartite association strength is greater than or equal to a preset association strength threshold. This target association strength is then input into a preset activity recommendation model to generate a personalized activity strategy. This initial activity strategy is further optimized using a reinforcement learning mechanism to obtain a target activity strategy, effectively improving the personalization and accuracy of mobile banking activity recommendations and enhancing the user experience.

[0057] It should be noted that the embodiments of the present application do not limit the execution entity of the mobile banking activity recommendation method. For example, the mobile banking activity recommendation method of the embodiments of the present application can be applied to information processing devices such as servers or terminal devices. The server can be a standalone server, a cluster server, or a cloud server. The terminal device can be an electronic device such as a smartphone, a computer, a personal digital assistant (PDA), or a tablet computer.

[0058] In order to make the purpose, technical solutions and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0059] Figure 1 This is a flow chart of a method for recommending mobile banking activities provided in an embodiment of the present application. Figure 1 As shown, the mobile banking activity recommendation method may include the following steps S101 to S105.

[0060] S101: Obtain user entities, activity entities, and scenario entities related to mobile banking.

[0061] In the embodiment of the present application, user location data, user behavior data, bank-side data, merchant-side data, and external scenario data related to mobile banking are obtained. The user location data is the user's current location, and the user behavior data includes user activity clicks and user activity searches. The user activity clicks are the cumulative number of times a user clicks on various activity entries in mobile banking, and the user activity searches are the cumulative number of times a user searches for related activity keywords in the mobile banking search bar.

[0062] The bank-side data includes data such as the activity name, activity period, and applicable population entered by bank staff; the merchant-side data includes activity information provided by the bank's cooperating merchants, which covers specific activity information such as shopping mall dining discounts and offline store discounts.

[0063] External scenario data includes structured data and unstructured data. Structured data can include standardized information such as weather data, holidays, dates and times, industry hot spots, and types of regional events. Unstructured data includes non-standardized content such as texts of hot topics on social media and full-text descriptions of news events.

[0064] Furthermore, embodiments of the present application construct user entities based on user location data and user behavior data, activity entities based on bank-side data and merchant-side data, and scenario entities based on external scenario data. For example, a user entity can extract characteristic information such as "age = 30 years old," "occupation = internet industry," and "consumption habits = high-frequency coffee consumption." An activity entity can extract core information such as "activity name = credit card installment," "activity period = 3 days," and "applicable population = 25 to 35 years old." A scenario entity can access external scenario data such as "rainy days," "Dragon Boat Festival," "weekdays at 12:00," "618 Shopping Festival," "National Day travel guide," and "Central Bank interest rate cut."

[0065] S102: Build a knowledge graph based on user entities, activity entities, and scenario entities.

[0066] The knowledge graph in the embodiment of the present application constructs a triple network with "user entity-activity entity-scenario entity" as the core, and performs structural modeling by defining the association relationship between the three types of entities. Specifically, an association relationship is established between the user entity and the activity entity, such as "user A" participates in "activity B"; an association relationship is formed between the scene entity and the activity entity, such as "scenario C" covers "activity B". Therefore, the above triple network uses entities as nodes and association relationships as edges to form a closed-loop knowledge structure in which "user-activity-scenario" are connected in series, providing clear semantic support for subsequent association strength calculations, etc.

[0067] S103: Based on the knowledge graph, the tripartite association strength of the user entity, the activity entity, and the scenario entity is calculated. When the tripartite association strength is greater than or equal to a preset association strength threshold, the tripartite association strength is determined to be the target association strength.

[0068] In this embodiment of the present application, based on the constructed knowledge graph, the three-way association strength is calculated through the following steps: First, the association strength between the user entity and the activity entity is calculated as the first association strength; second, the association strength between the scene entity and the activity entity is calculated as the second association strength; then, the product of the first association strength and the second association strength is calculated to obtain the third association strength; finally, the first association strength, the second association strength, and the third association strength are weighted and summed to obtain the three-way association strength.

[0069] In addition, if the three-party correlation strength is greater than or equal to a preset correlation strength threshold, for example, the preset correlation strength threshold is 0.7, the three-party correlation strength can be determined as the target correlation strength. It should be noted that the embodiment of the present application does not limit the specific value of the correlation strength threshold, which can be dynamically adjusted according to actual business needs.

[0070] For example, if the user entity is "28-year-old young mother, who often buys parent-child products", the scene entity is "rainy days, shopping malls and afternoon hours", and the activity entity is "parent-child clothing discounts in shopping malls", in this scenario, the first association strength calculated based on the user entity and the activity entity is 0.8; the second association strength calculated based on the scene entity and the activity entity is 0.8; the product of the first association strength and the second association strength is 0.64, that is, the third association strength is 0.64; the first association strength, the second association strength and the third association strength are summed up according to weighted coefficients 0.4, 0.4 and 0.2, and the final three-party association strength is 0.832. At this time, the obtained three-party association strength of 0.832 is greater than the preset association strength threshold of 0.7, so the three-party association strength of 0.832 is the target association strength.

[0071] It should be noted that if the three-way correlation strength is less than 0.7, the correlation strength between the activity entity and the user entity and the scenario entity is insufficient. For example, in the above scenario, if the activity entity is "Credit Card Rainy Day Parking Discount," its calculated three-way correlation strength is 0.51. Because it is lower than the correlation strength threshold of 0.7, it indicates that the correlation strength between the activity entity and the user entity and the scenario entity is insufficient. Therefore, the calculated three-way correlation strength of 0.51 cannot be used as the target correlation strength.

[0072] S104: Input the target association strength into the activity recommendation model to generate an initial activity strategy. The activity recommendation model is pre-trained through federated learning.

[0073] In the activity recommendation model of traditional mobile banking, the data on the bank side and the merchant side are isolated from each other, making it difficult to achieve the integrated push of "bank activities" and "merchant discounts". Taking a shopping mall's co-branded credit card activity as an example, since the data on the bank side and the merchant side were not coordinated, only 32% of the target users received bank activity push notifications, which ultimately led to a 45% decrease in activity participation rate compared to expectations. The activity recommendation model in the embodiment of the present application is pre-trained using a federated learning method, which effectively solves the data barriers between the bank side and the merchant side. It can achieve data collaboration under the premise of protecting the data privacy of both parties, and generate more accurate recommendation strategies by integrating the information of both parties, avoiding problems such as insufficient activity attractiveness caused by data isolation. The training steps of the activity recommendation model are as follows.

[0074] First, obtain input data related to mobile banking, where the input data includes user historical profile data and merchant historical activity participation data. User historical profile data includes user historical location data and user historical behavior data, and user historical behavior data also includes user historical activity click volume and user historical activity search volume.

[0075] It should be noted that in addition to user historical location data and user historical behavior data, user historical portrait data also covers user historical personal information, such as basic information such as user gender, age, and ID number. These data together constitute complete user historical portrait data.

[0076] Then, after obtaining the input data related to mobile banking, the user's historical location data in the user portrait data needs to be converted into a regional tag including province, city, district and business district to obtain the user's historical regional tag. The embodiment of the present application can obtain the user's real-time latitude and longitude through the Global Positioning System (GPS) or Beidou positioning, and then call the map application programming interface (API) interface to convert the user's historical location data into a precise regional tag such as "Guomao Business District, Chaoyang District, City A". If the conversion process fails, it will automatically switch to the base station closest to the user's historical location for positioning supplement to ensure the effective generation of the regional tag.

[0077] The traditional mobile banking activity recommendation model cannot respond to changes in popularity in real time. For example, a bank launched a "coffee instant discount" activity, and the number of clicks increased by 200% in the afternoon of the second day after the activity went online. However, because the code was not updated in time, it was still displayed in the original order on the third day, resulting in a 65% reduction in the exposure of the activity during the peak period of popularity, missing out on approximately 23,000 potential clicks. However, the embodiment of the present application can use a sliding window algorithm to first count the number of user historical activity clicks and user historical activity search volume in the user behavior data, and then based on the statistical results, multiply the user historical activity click volume by a weight coefficient of 0.7, and multiply the user historical activity search volume by a weight coefficient of 0.3. Finally, these two multiplication results are added together to obtain the user historical activity popularity value.

[0078] Specifically, the size of the sliding window can be set to 7 days, and the number of clicks on user historical activities that exceed the window will be reduced according to a decay coefficient of 0.5. For example, a certain activity was set up by the bank 14 days ago. The total number of clicks on user historical activities in these 14 days is 1960. Among them, the number of clicks on user historical activities in the first 7 days is 980, the number of clicks on user historical activities on the 8th day is 200, the number of clicks on user historical activities on the 9th day is 180, the number of clicks on user historical activities on the 10th day is 160, the number of clicks on user historical activities on the 11th day is 140, the number of clicks on user historical activities on the 12th day is 120, the number of clicks on user historical activities on the 13th day is 100, and the number of clicks on the 14th day is 100. The number of clicks on the user historical activity in the day is 80 times, and the number of searches for the user historical activity in these 14 days is 300 times, then the user historical activity heat value H = 0.7×(980+200×0.5+180×0.25+160×0.125+140×0.0625+120×0.03125+100×0.015625+80×0.0078125)+0.3×300≈901.78, that is, the user historical activity heat of the activity in these 14 days is 901.78.

[0079] Furthermore, after obtaining the user's historical regional label and the user's historical activity heat value, combined with the merchant's historical activity participation data, it is also necessary to consider the user's historical personal information part in the user's historical portrait data, and use these together as the target input data. Subsequently, the target input data is desensitized. Specifically, the embodiment of the present application can encrypt the sensitive information that may be involved in the target input data through SHA-256, and fuzzify the key identification information. In addition, the embodiment of the present application can use the HE algorithm to perform data encryption processing on the user's historical activity heat value to ensure that the encrypted data can still perform addition and multiplication operations, thereby obtaining the target desensitized data.

[0080] In the embodiment of the present application, the target desensitized data is input into the model to be trained, and cross-institutional collaborative training is performed based on multiple distributed clients of mobile banking. The multiple distributed clients include a bank side and a merchant side. Therefore, the bank side and the merchant side each perform cross-institutional collaborative training on the model to be trained based on local data. The local data of the bank side is the desensitized user's historical regional labels and user's historical activity heat values, while the local data of the merchant side is the desensitized merchant's historical activity participation data.

[0081] It should be noted that when the bank and merchant sides train the to-be-trained model based on local data, the number of iterations can be set to 10, the learning rate can be set to 0.01, and the corresponding model gradient is generated for each iteration. In addition, to prevent the leakage of the generated model gradient information, the embodiment of the present application can use MPC to encrypt the model gradient and use obfuscation circuit technology to encode the model gradient parameters to prevent the leakage of gradient information.

[0082] Furthermore, after using MPC to encrypt the model gradient, the bank and merchant sides must upload the encrypted model gradient to the central server. The central server calculates the average encrypted model gradient of the encrypted model gradients of both parties in the ciphertext state, and still uses the HE algorithm to implement additive homomorphic encryption in the calculation, so that the encrypted model gradient aggregation can be completed without decryption.

[0083] Next, after the encrypted model gradients are aggregated on the bank and merchant sides, the central server returns the aggregated encrypted gradients to each bank and merchant side. Each participating side uses the local private key to decrypt the gradients and obtain the updated model parameters. Through the federated learning model of "data remains unchanged but model moves", the model training is finally completed, and the activity recommendation model is obtained after the training.

[0084] S105: Optimize the initial activity strategy based on the preset reinforcement learning model to generate a target activity strategy.

[0085] The user behavior data in the embodiment of the present application also includes the user activity click-through rate, user activity participation rate, and user activity invalid display rate. Generally speaking, the user activity click-through rate is the number of activity clicks divided by the total number of activity exposures, the user activity participation rate is the number of activity participations divided by the number of activity clicks, and the user activity invalid display rate is the number of exposures that were not clicked by the user divided by the total number of activity exposures.

[0086] Therefore, in the embodiment of the present application, a reward function in the reinforcement learning model can be constructed based on the user activity click-through rate, user activity participation rate, and activity invalid display rate in the user behavior data. Specifically, the reward function R = 0.5 × user activity click-through rate + 0.3 × user activity participation rate - 0.2 × user activity invalid display rate. Based on the constructed reward function, the preset reinforcement learning model is iteratively trained to optimize the initial activity strategy and ultimately generate the target activity strategy.

[0087] Furthermore, the embodiment of the present application can provide personalized recommendations to users based on target activity strategies in the activity display module on the mobile banking homepage, and rotate activities that users are interested in according to activity heat value, timeliness weight and target association strength, where the timeliness weight is the remaining days of the activity / the initial days of the activity, which is used to quantify the time urgency of the activity.

[0088] In addition, the embodiments of the present application can also monitor user behavior to trigger a dynamic recommendation mechanism. For example, when a user stays on an activity page card for more than a preset time, graphic details will automatically pop up to clearly display the activity rules and the user's exclusive discount code; or, for limited-time activities, the system will automatically trigger a full-screen pop-up prompt, which contains an activity countdown, exclusive discount information and a one-click participation button to strengthen the user's urge to participate in the activity; at the same time, the activity heat value and timeliness weight will be refreshed in real time every 10 minutes, and the display order will be dynamically adjusted to ensure that the recommended content always fits the user's current interests and changes in activity timeliness.

[0089] Based on the relevant contents of the above steps S101 to S105, it can be known that in an embodiment of the present application, by obtaining user entities, activity entities and scenario entities related to mobile banking, a knowledge graph is constructed based on the obtained user entities, activity entities and scenario entities. Based on the constructed knowledge graph, the three-party association strength of the user entity, activity entity and scenario entity is calculated, and when the three-party association strength is greater than or equal to the preset association strength threshold, the three-party association strength is determined to be the target association strength. Subsequently, the target association strength is input into the preset activity recommendation model to generate an initial activity strategy. Finally, the initial activity strategy is optimized based on the preset reinforcement learning model to generate a target activity recommendation strategy.

[0090] As can be seen, this solution constructs the acquired user entities, activity entities, and scenario entities into a knowledge graph, enabling the multi-dimensional mining of semantic associations between entities. Based on this, by calculating the tripartite association strength among user, activity, and scenario entities, a target association strength is determined when the tripartite association strength is greater than or equal to a preset association strength threshold. This target association strength is then input into a preset activity recommendation model to generate a personalized activity strategy. This initial activity strategy is further optimized using a reinforcement learning mechanism to obtain a target activity strategy, effectively improving the personalization and accuracy of mobile banking activity recommendations and enhancing the user experience.

[0091] It should be noted that the user's historical personal information (including but not limited to basic information such as user location data, name, age, and ID number) and data (including but not limited to user location data and user behavior data, as well as data used for analysis, stored data, displayed data, 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 relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0092] Figure 2 This is a schematic diagram of the structure of a mobile banking activity recommendation device provided in an embodiment of the present application. Figure 2 As shown, the mobile banking activity recommendation device 200 provided in the embodiment of the present application may include:

[0093] Entity acquisition module 201, used to acquire user entities, activity entities and scenario entities related to mobile banking;

[0094] A graph construction module 202, configured to construct a knowledge graph based on the user entity, activity entity, and scenario entity;

[0095] An association strength calculation module 203 is configured to calculate the three-party association strength among the user entity, the activity entity, and the scenario entity based on the knowledge graph, and determine the three-party association strength as a target association strength when the three-party association strength is greater than or equal to a preset association strength threshold;

[0096] A strategy generation module 204 is configured to input the target association strength into an activity recommendation model to generate an initial activity strategy, wherein the activity recommendation model is pre-trained by federated learning;

[0097] The strategy optimization module 205 is used to optimize the initial activity strategy based on a preset reinforcement learning model to generate a target activity strategy.

[0098] Optionally, the entity acquisition module 201 is specifically configured to:

[0099] Obtaining user location data, user behavior data, bank-side data, merchant-side data, and external scenario data related to the mobile banking;

[0100] The user entity is constructed based on the user location data and the user behavior data; the activity entity is constructed based on the bank-side data and the merchant-side data; and the scenario entity is constructed based on the external scenario data.

[0101] Optionally, the association strength calculation module 203 is specifically configured to:

[0102] Based on the knowledge graph, calculating the association strength between the user entity and the activity entity as a first association strength;

[0103] calculating the association strength between the scene entity and the activity entity as a second association strength;

[0104] multiplying the first association strength by the second association strength as a third association strength;

[0105] The tripartite association strength is obtained by performing a weighted sum calculation based on the first association strength, the second association strength, and the third association strength.

[0106] Optionally, the strategy optimization module 205 is specifically configured to:

[0107] Constructing a reward function of the reinforcement learning model based on the user behavior data;

[0108] Based on the reward function, the reinforcement learning model is iteratively trained to optimize the initial activity policy and generate a target activity policy.

[0109] Optionally, the policy generation module 204 may include:

[0110] A training data acquisition module is used to acquire input data related to the mobile banking, wherein the input data includes user historical profile data and merchant historical activity participation data, wherein the user historical profile data includes user historical location data and user historical behavior data, and wherein the user historical behavior data includes user historical activity click volume and user historical activity search volume;

[0111] A data preprocessing module, used to preprocess the input data to obtain target input data;

[0112] A desensitization processing module, used to perform desensitization processing on the target input data to obtain target desensitized data;

[0113] The collaborative training module is used to input the target desensitized data into the model to be trained, and perform cross-institutional collaborative training based on multiple distributed clients of the mobile banking. After the training is completed, the activity recommendation model is obtained.

[0114] Optionally, the data preprocessing module is specifically configured to:

[0115] Convert the user's historical location data into a region label including province, city, district, and business district to obtain the user's historical region label;

[0116] The user historical activity heat value is calculated based on the user historical activity click volume and the user historical activity search volume.

[0117] Optionally, the collaborative training module is specifically used to:

[0118] The multiple distributed clients perform cross-institutional collaborative training on the model to be trained based on local data, wherein the multiple distributed clients include a bank side and a merchant side, the local data of the bank side is the user's historical regional labels and the user's historical activity heat values ​​that have been desensitized, and the local data of the merchant side is the merchant's historical activity participation data that has been desensitized.

[0119] Furthermore, an embodiment of the present application also provides an electronic device, comprising: a processor, a memory, and a system bus;

[0120] The processor and the memory are connected via the system bus;

[0121] The memory is used to store one or more programs, and the one or more programs include instructions. When the instructions are executed by the processor, the processor executes any one of the implementation steps of the above-mentioned mobile banking activity recommendation method.

[0122] Furthermore, an embodiment of the present application also provides a computer-readable storage medium, which is used to store a computer program. When the computer program is executed by a terminal device, it implements any implementation step in the above-mentioned mobile banking activity recommendation method.

[0123] It can be seen from the description of the above implementation methods that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product can be stored in a storage medium, such as ROM / RAM, a disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network communication device such as a media gateway, etc.) to execute the methods described in each embodiment of the present application or certain parts of the embodiments. It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the various embodiments can be referred to each other.

[0124] As for the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0125] It should also be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0126] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for recommending activities in mobile banking, characterized in that: The method comprises: Get user entities, activity entities, and scenario entities related to mobile banking; Building a knowledge graph based on the user entity, activity entity and scenario entity; Based on the knowledge graph, calculating the three-party association strength of the user entity, the activity entity, and the scene entity, and determining the three-party association strength as a target association strength when the three-party association strength is greater than or equal to a preset association strength threshold; Inputting the target association strength into an activity recommendation model to generate an initial activity strategy, wherein the activity recommendation model is pre-trained through federated learning; The initial activity strategy is optimized based on a preset reinforcement learning model to generate a target activity strategy.

2. The method according to claim 1, characterized in that The obtaining of user entities, activity entities and scene entities includes: Obtaining user location data, user behavior data, bank-side data, merchant-side data, and external scenario data related to the mobile banking; The user entity is constructed based on the user location data and the user behavior data; the activity entity is constructed based on the bank-side data and the merchant-side data; and the scenario entity is constructed based on the external scenario data.

3. The method according to claim 1, characterized in that The calculating, based on the knowledge graph, the tripartite association strength among the user entity, the activity entity, and the scenario entity includes: Based on the knowledge graph, calculating the association strength between the user entity and the activity entity as a first association strength; calculating the association strength between the scene entity and the activity entity as a second association strength; multiplying the first association strength by the second association strength as a third association strength; The tripartite association strength is obtained by performing a weighted sum calculation based on the first association strength, the second association strength, and the third association strength.

4. The method according to claim 1, wherein The initial activity strategy is optimized based on a preset reinforcement learning model to generate a target activity strategy, including: Constructing a reward function of the reinforcement learning model based on the user behavior data; Based on the reward function, the reinforcement learning model is iteratively trained to optimize the initial activity policy and generate a target activity policy.

5. The method according to claim 1, wherein The activity recommendation model is trained through the following steps: Acquire input data related to the mobile banking, the input data including user history profile data and merchant history activity participation data, the user history profile data including user history location data and user history behavior data, the user history behavior data including user history activity click volume and user history activity search volume; Preprocessing the input data to obtain target input data; Desensitizing the target input data to obtain target desensitized data; The target desensitized data is input into the model to be trained, and cross-institutional collaborative training is performed based on multiple distributed clients of the mobile banking. After the training is completed, the activity recommendation model is obtained.

6. The method according to claim 5, characterized in that The preprocessing of the input data to obtain target input data includes: Convert the user's historical location data into a region label including province, city, district, and business district to obtain the user's historical region label; The user historical activity heat value is calculated based on the user historical activity click volume and the user historical activity search volume.

7. The method according to claim 5, characterized in that The cross-institutional collaborative training based on multiple distributed clients includes: The multiple distributed clients perform cross-institutional collaborative training on the model to be trained based on local data, wherein the multiple distributed clients include a bank side and a merchant side, the local data of the bank side is the user's historical regional labels and the user's historical activity heat values ​​that have been desensitized, and the local data of the merchant side is the merchant's historical activity participation data that has been desensitized.

8. A mobile banking activity recommendation device, characterized in that: include: Entity acquisition module, used to obtain user entities, activity entities and scenario entities related to mobile banking; A graph construction module, configured to construct a knowledge graph based on the user entity, activity entity, and scenario entity; an association strength calculation module, configured to calculate, based on the knowledge graph, the tripartite association strength of the user entity, the activity entity, and the scenario entity, and determine the tripartite association strength as a target association strength when the tripartite association strength is greater than or equal to a preset association strength threshold; a strategy generation module, configured to input the target association strength into an activity recommendation model to generate an initial activity strategy, wherein the activity recommendation model is pre-trained through federated learning; The strategy optimization module is used to optimize the initial activity strategy based on a preset reinforcement learning model to generate a target activity strategy.

9. An electronic device, characterized in that: The device includes: a processor, a memory, and a system bus; The processor and the memory are connected via the system bus; The memory is used to store a program, wherein the program includes instructions, and when the instructions are executed by the processor, the processor executes the steps of the mobile banking activity recommendation method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium is used to store a computer program, and when the computer program is executed by a terminal device, the steps of the mobile banking activity recommendation method according to any one of claims 1 to 7 are implemented.

Images