A multi-strategy fusion recommendation method and device

Abstract

The application discloses a multi-strategy fusion recommendation method and device, relates to the field of big data analysis and processing, and comprises the following steps: collecting user click behavior data about services; processing the click behavior data and storing the processed click behavior data into a Hive data warehouse; determining different service recommendation strategies based on the Hive data warehouse; the service recommendation strategies comprise a top service recommendation strategy, a recent use service recommendation strategy, a label matching service recommendation strategy and a popular service recommendation strategy; and fusing the different service recommendation strategies to determine a fused service recommendation strategy list. The application meets the diversified and personalized service demands of different users for different services.

Description

Technical Field

[0001] This application relates to the field of big data analysis and processing, and in particular to a multi-strategy fusion recommendation method and device. Background Technology

[0002] With the deepening of digital transformation, a certain platform, as an important carrier of digital social security services, has been widely used in scenarios such as social security inquiries, benefit applications, and personal pensions. The National Government Affairs Platform provides users with convenient service access through mobile terminals (such as apps, H5 pages, and mini-programs), significantly improving the accessibility of public services. However, unlike high-frequency usage scenarios such as e-commerce and short videos, the National Government Affairs Platform has unique characteristics, mainly reflected in the following aspects: 1. Low user frequency: Users typically only use the national government affairs platform when they have specific needs (such as querying or applying for services), resulting in sparse user behavior data.

[0003] 2. Diverse and unevenly distributed services: The national government affairs platform offers a wide variety of services, covering both national and local services. Different users pay significant attention to and use different services at different frequencies.

[0004] 3. Data sparsity: Due to the sparse nature of user behavior data, traditional recommendation methods (such as collaborative filtering and deep learning) perform poorly in this scenario, making it difficult to meet the diverse and personalized service needs of users. Summary of the Invention

[0005] The purpose of this application is to provide a multi-strategy fusion recommendation method and device, which solves the problem of poor personalized recommendation performance caused by sparse user behavior data and a wide variety of services in national government platforms.

[0006] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a multi-strategy fusion recommendation method, including: In one implementation, tracking code is embedded in the front end of the web application; Based on the tracking code, user click behavior data related to the service is collected and recorded in real time to generate log records; the log records include user ID, service name, click event and insurance location code.

[0007] In one implementation, click behavior data from log records is collected through a cloud log service; The click behavior data is stored in the created log database; The click behavior data in the log library is split into two parts: the first part of the click behavior data is written to a message queue, and the second part of the click behavior data is written to the Hadoop Distributed File System. The click behavior data in the Hadoop Distributed File System is parsed, and the parsed click behavior data is written to the Hive data warehouse.

[0008] In one implementation, based on the Hive data warehouse, the operations staff selects n services for top recommendation and determines the top service recommendation strategy; Recommendations are made based on services that users have recently used frequently, and a recommendation strategy for recently used services is determined. Recommendations are made based on the tag matching degree between users and services, and a tag matching service recommendation strategy is determined. Recommendations are made based on the number of clicks for each service, and a strategy for recommending popular services is determined.

[0009] In one implementation, the operators select n services as top-recommended services and set the order of their placement. The recommendation results of the pinning order are output to the pinning service recommendation strategy table for pinning recommendation.

[0010] In one implementation, service tags are set, and a score is determined for each click on a service tag; Based on the click service tag score, determine the user tag score and the service tag score; the user tag score = user-actively set score + coefficient × click service tag score × cumulative number of clicks; the service tag score = coefficient × service tag score; The degree of tag matching between users and services is determined based on user tag scores and service tag scores; The user and service tag matching degree is output to the tag matching service recommendation strategy table for recommendation.

[0011] In one implementation, based on the Hive data warehouse, click behavior data related to the service over the past N days is extracted; Based on the click behavior data, the click volume of national services and local services is calculated respectively; Sort the click volume of national services and local services from highest to lowest, record the serial numbers, and determine the sorting sequence; the serial numbers are 1, 2, 3, ..., m, where m is the total number of serial numbers; According to the sorting sequence, the services with clicks corresponding to those with serial numbers less than a set serial number are output to the popular service recommendation strategy table for recommendation; the services are national services or local services.

[0012] In one implementation, the selected n top-rated services are added sequentially to the recently used service recommendation strategy table; Remove services that are duplicated in the recently used service recommendation strategy table from the top service recommendation strategy table to determine the first service table; Sort the first service table by click time from most recent to oldest, and select n services to add to the tag matching service recommendation strategy table; Remove services from the tag matching service recommendation strategy table that are duplicated in the top service recommendation strategy table and the recently used service recommendation strategy table to determine the second service table; Sort the second service table from high to low according to the matching degree, and select n services to add to the popular service recommendation strategy table; Remove services from the popular service recommendation strategy table that are duplicated with the top service recommendation strategy table, the recently used service recommendation strategy table, and the tag matching service recommendation strategy table to determine the third service table; The third service table is sorted from high to low popularity and added to the recommendation list to determine the integrated service recommendation strategy table.

[0013] In a second aspect, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the multi-strategy fusion recommendation method described in any one of the above.

[0014] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a multi-strategy fusion recommendation method and device. It collects user click behavior data related to services and logs it in real time, recording user click behavior to capture immediate interests even with low-frequency use, avoiding reliance on long-term historical data. The recorded click behavior data is collected and processed, and the processed click behavior data is stored in a Hive data warehouse. Based on the Hive data warehouse, different service recommendations are made according to different user needs scenarios. Service recommendations include top-ranked service recommendations, recently used service recommendations, tag-matching service recommendations, and popular service recommendations. In the context of sparse user behavior data in government services, this application proposes for the first time a personalized recommendation method that integrates four recommendation strategies: top-ranked recommendations, recently used recommendations, tag-matching recommendations, and popular recommendations. Through strategy priority ranking, deduplication mechanisms, and weight fusion algorithms, the accuracy, coverage, and fairness of the recommendation results are ensured. By fusing different service recommendations to determine the fused service recommendation list, this method solves the problem of poor personalized recommendation performance in national government platforms caused by sparse user behavior data and a wide variety of services, and can meet the diverse and personalized service needs of different users. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating a multi-strategy fusion recommendation method in one embodiment of this application.

[0017] Figure 2 This is a schematic diagram of the system structure provided in an embodiment of this application.

[0018] Figure 3 This is a schematic diagram illustrating the steps of a multi-strategy fusion recommendation method provided in an embodiment of this application.

[0019] Figure 4 This is a schematic diagram of a module provided in one embodiment of this application.

[0020] Figure 5 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] In existing technologies, the most similar implementation to this application is mainly based on user-defined service preference tags, recommending relevant services through tag matching. Specifically, after logging into the national government affairs platform application, users can select several preset service preference tags (such as "enterprise employees," "urban and rural residents," "flexible employment personnel," etc.). The system then filters and recommends matching services from the service library based on the user's selected tags. However, this solution has the following significant drawbacks: The tagging system is static and lacks dynamic optimization capabilities: the existing solution does not collect and analyze user behavior data (such as service click behavior) to achieve automatic tag optimization, resulting in the tagging system being unable to adapt to changes in user needs, and the recommendation accuracy and adaptability are low.

[0024] The recommendation strategy is simplistic and lacks adaptability to different scenarios: it relies solely on tag matching, lacks flexibility, and cannot take into account multiple dimensions of information such as recent user preferences, popular services, and operational priorities, resulting in limited coverage of recommendation results.

[0025] The recommended balance between local and national services has not been considered: Since the number of national services is usually much higher than that of local services, the existing scheme is prone to low exposure of local services, which affects user experience and service fairness.

[0026] The cold start problem is prominent: for users who have not selected service preference tags, the existing solution cannot provide effective recommendation results, affecting the availability of the system.

[0027] In view of the above-mentioned shortcomings of existing technical solutions, this application aims to solve the following core technical problems: The issue of recommendation accuracy in data-sparse scenarios: How to meet the diverse and personalized service needs of users and improve the relevance of recommendation results in scenarios where user behavior data is used but is sparse.

[0028] Recommendation fairness issue: After using user behavior data, how to balance the display opportunities of national services and local services in the recommendation results, so as to avoid local services being marginalized due to low click-through rates and improve the fairness of recommendations.

[0029] Cold start problem: How to ensure basic usability of recommendations and provide effective recommendation results when users have not selected service preference tags or have no user behavior data.

[0030] This application provides a multi-strategy fusion recommendation method and system, applicable to government service platforms such as national government service platforms, to achieve an intelligent service model of "services proactively finding people". Specifically, the objectives of this application include: Improve recommendation accuracy: By integrating multiple recommendation strategies (top recommendation, recently used recommendation, tag matching recommendation, popular recommendation), it can still dynamically provide highly relevant service recommendations even when user behavior data is sparse.

[0031] To ensure fairness in recommendations: Design a recommendation weight adjustment mechanism for national and local services to avoid local services being marginalized due to differences in click volume, thereby improving the exposure and user experience of local services.

[0032] Optimize user experience: In cases where users have not selected service preference tags or there is no user behavior data, provide fallback recommendations through top and popular recommendation strategies to ensure the continued availability of services.

[0033] Enhance system controllability and interpretability: Provide an operations management interface to support operators in flexibly adjusting recommendation strategies, tag scores, and weight coefficients, ensuring the traceability and interpretability of recommendation results.

[0034] like Figure 1 and Figure 3 As shown in the figure, this application provides a multi-strategy fusion recommendation method, the details of which are as follows.

[0035] S1: Collect user click behavior data related to the service.

[0036] S2: Process the click behavior data and store the processed click behavior data in the Hive data warehouse.

[0037] S3: Based on the Hive data warehouse, determine different service recommendation strategies; the service recommendation strategies include the top service recommendation strategy, the recently used service recommendation strategy, the tag matching service recommendation strategy, and the popular service recommendation strategy.

[0038] S4: Integrate different service recommendation strategies and determine the list of integrated service recommendation strategies.

[0039] Furthermore, in an exemplary embodiment, S1 can be replaced by the following steps.

[0040] S11: Embed tracking code in the front end of web applications.

[0041] S12: Based on the tracking code, collect user click behavior data about the service and record it in real time to generate log records; the log records include user ID, service name, click event and insurance location code.

[0042] Embed tracking code in the front end of national government affairs platform applications (including apps, H5 pages, mini-programs, etc.) to record user service clicks in real time. Log content includes, but is not limited to, the following fields: 1. User ID: A unique identifier for a user.

[0043] 2. Service Name: The specific service that the user clicks.

[0044] 3. Click time: The timestamp of the user's click action.

[0045] 4. Insurance Location Code: The insurance location to which the user belongs.

[0046] Log data is sent to the log server in real time through the front-end tracking module and is written to the server disk in a rolling manner.

[0047] Furthermore, in an exemplary embodiment, S2 can be replaced by the following steps.

[0048] S21: Collect click behavior data from log records through cloud log service.

[0049] S22: Store the click behavior data in the created log library.

[0050] S23: The click behavior data in the log database is split into two parts: the first part of the click behavior data is written to the message queue, and the second part of the click behavior data is written to the Hadoop Distributed File System.

[0051] S24: Parse the click behavior data in the Hadoop Distributed File System and write the parsed click behavior data to the Hive data warehouse.

[0052] The National Government Affairs Platform's personalized service recommendation system is based on a big data platform and employs distributed log collection and storage technology to collect and process user behavior data in real time. Specific implementation includes: 1. Real-time data collection: Collect behavior logs written to disk on each server through cloud log service, and create a log library to store raw log data.

[0053] 2. Data splitting and storage: The collected log data is split and processed using the data acquisition module. One part of the data is written to a message queue for real-time computing, and the other part is written to HDFS (Hadoop Distributed File System) for offline analysis.

[0054] 3. Data warehouse storage: The behavior logs in HDFS are parsed and written to the Hive data warehouse to support the calculation of offline recommendation strategies.

[0055] Furthermore, in an exemplary embodiment, S3 can be replaced by the following steps.

[0056] S31: Based on the Hive data warehouse, the operations staff selects n services for top recommendation and determines the top service recommendation strategy.

[0057] S32: Make recommendations based on services that users have used frequently recently, and determine the recommendation strategy for recently used services.

[0058] S33: Make recommendations based on the tag matching degree between users and services, and determine the tag matching service recommendation strategy.

[0059] S34: Make recommendations based on the number of clicks for services and determine the recommendation strategy for popular services.

[0060] Furthermore, S31 specifically includes: S311: The operations staff selects n services as top-recommended services and sets the order of their placement.

[0061] S312: Output the recommendation results of the pinning order to the pinning service recommendation strategy table for pinning recommendation.

[0062] Top recommendation: Function: Allows operators to manually select services to be featured at the top, for promoting key or policy-related services.

[0063] Furthermore, S32 specifically includes: S321: Extract user click behavior data related to the service from the Hive data warehouse over the past N days.

[0064] S322: Based on the click behavior data, retain the last record of each type of service recently clicked by the user.

[0065] S323: Sort the last record in each type of service according to the user's click time from most recent to oldest, and generate a click sequence.

[0066] S324: Determine the recommendation results based on the generated click sequence.

[0067] S325: Output the recommendation results to the recently used service recommendation strategy table for recommendation.

[0068] Recommended for recent use: Function: Recommends services based on users' recent high-frequency use, improving user operation efficiency.

[0069] Implementation method: 1. The system extracts user service click behavior data for the past N days from the real-time data warehouse. Record fields include service name, click time, etc.

[0070] 2. For services that are clicked repeatedly, only the latest record will be retained.

[0071] 3. Sort services by click time from most recent to oldest, and select the top-ranked services (N services by default) as the recommended results.

[0072] Furthermore, S33 specifically includes: S331: Set service tags and determine the score for clicking service tags.

[0073] S332: Determine the user tag score and the service tag score based on the click service tag score; the user tag score = user-actively set score + coefficient × click service tag score × cumulative number of clicks; the service tag score = coefficient × service tag score.

[0074] S333: Determine the tag matching degree between users and services based on user tag scores and service tag scores.

[0075] S334: Output the user and service tag matching degree to the tag matching service recommendation strategy table for recommendation.

[0076] Tag matching recommendations: Function: Make recommendations based on the matching degree of tags between users and services, and explore users' deeper service preferences.

[0077] Implementation method: Service Tag Settings: The system provides an operations management interface where operators can manually set tags and their representative scores (ranging from 0 to N) for each service. A service can be associated with multiple tags; for example, the service "Enterprise Employee Pension Insurance Payment Certificate Inquiry" can be associated with tags such as "Enterprise Employee" and "Pension Insurance".

[0078] User tag generation: 1. User-defined tags: Users manually select preference tags in the National Government Affairs Platform application, and the scores are set by the operators through the management interface.

[0079] 2. User Behavior-Derived Tags: Based on the user's historical click behavior, combined with the service tag score and the number of clicks, a user tag score is calculated. The calculation formula is as follows: User tag score = User-defined score + Coefficient × Click service tag score × Cumulative click count; where the coefficient is used to balance the recommendation weight of national services and local services. Considering that services in the national government affairs platform application are divided into national services and local services (services unique to each province), and that the click volume of local services is usually much lower than that of national services, which may lead to a lower matching degree, the system adjusts the tag scores of national services and local services by setting a coefficient. The coefficient is entered by the operations staff through the management interface.

[0080] Tag matching calculation: 1. For each service, calculate its tag score: Service tag score = coefficient × service tag value.

[0081] 2. For each user, calculate their tag score and extract the top N tags.

[0082] 3. Filter service sets where user tags and service tags overlap.

[0083] 4. For each service in the service set, calculate its match degree with the user. The match degree calculation formula is as follows: User-service matching degree = Σ(user tag score × service tag score).

[0084] Where Σ represents the sum of the scores of the common tags of the user and the service. For example, if the tag score of user 1 is {[tag 1] 60, [tag 4] 50, [tag 6] 70}, and the tag score of service 1 is {[tag 1] 10, [tag 5] 80, [tag 6] 100}, then the matching degree between user 1 and service 1 is: matching degree = 60 × 10 + 70 × 100 = 7600.

[0085] 5. For all services in the service set, sort them from high to low matching degree, and select the top-ranked services as recommended candidates.

[0086] Furthermore, S34 specifically includes: S341: Based on the Hive data warehouse, extract click behavior data about the service over the past N days.

[0087] S342: Based on the click behavior data, calculate the click volume for national services and local services respectively.

[0088] S343: Sort the click volume of national services and local services from high to low, record the serial numbers, and determine the sorting sequence; the serial numbers are 1, 2, 3, ..., m, where m is the total number of serial numbers.

[0089] S345: According to the sorting sequence, output the services with clicks corresponding to the sequence number less than the set sequence number in the sorting sequence to the popular service recommendation strategy table for recommendation; the services are national services or local services.

[0090] Popular Recommendations: Function: Recommend services based on their popularity (clicks) to meet users' needs for popular services.

[0091] Implementation method: 1. The system extracts service click behavior data from the Hive offline data warehouse for the most recent N days and calculates the click volume for services nationwide and services in specific regions.

[0092] 2. For services nationwide, sort them from highest to lowest click count and record the sequence number.

[0093] 3. For local services, group them according to the place of insurance, and within each province, sort them from highest to lowest according to the number of clicks, and record the serial number.

[0094] 4. Based on the ranking results, select the top-ranked services as recommended candidates.

[0095] Furthermore, in an exemplary embodiment, S4 can be replaced by the following steps.

[0096] S41: Add the selected n top services to the recently used service recommendation strategy table in order.

[0097] S42: Remove services that are duplicates of the top service recommendation strategy table from the recently used service recommendation strategy table to determine the first service table.

[0098] S43: Sort the first service table according to the click time from most recent to oldest, and select n services to add to the tag matching service recommendation strategy table.

[0099] S44: Remove services from the tag matching service recommendation strategy table that are duplicates of the top service recommendation strategy table and the recently used service recommendation strategy table to determine the second service table.

[0100] S45: Sort the second service table from high to low according to the matching degree, and select n services to add to the popular service recommendation strategy table.

[0101] S46: Remove services from the popular service recommendation strategy table that are duplicates of the top service recommendation strategy table, the recently used service recommendation strategy table, and the tag matching service recommendation strategy table to determine the third service table.

[0102] S47: Sort the third service table from high to low popularity and add it to the recommendation list to determine the fused service recommendation strategy table.

[0103] Fusion of multiple recommendation strategies: Function: It integrates the results of the above four recommendation strategies to generate a final personalized recommendation service list.

[0104] Implementation method: The recommendation weighting order is: top recommendations > recently used recommendations > tag matching recommendations > popular recommendations. The specific fusion logic is as follows: 1. Top Recommendation: Directly add the N top services set by the operations staff to the recommendation list in sequence.

[0105] 2. Recently Used Recommendations: From the recently used recommended results, after removing services that are duplicates of the top recommended services, sort them by click time from most recent to oldest, and select N services to add to the recommended list.

[0106] 3. Tag Matching Recommendation: From the tag matching recommendation results, after removing services that are duplicates of the top recommendation and recently used recommendations, sort them from high to low matching degree and select N services to add to the recommendation list.

[0107] 4. Popular Recommendations: From the popular recommendations results, services that are duplicates of the top recommendations, recently used recommendations, and tag-matched recommendations are removed. These services are then sorted by popularity from highest to lowest and added to the recommendation list to a total of N services. The ratio of nationwide services to local services is 1:1. If the number of popular services to be added is even, one more nationwide service will be included.

[0108] 5. Recommendation result storage: The merged recommendation service list is stored in a MySQL database. The fields include user ID and recommendation service list. The recommendation list format is: [service 1, service 2, ..., service 15].

[0109] Applications of personalized service recommendations: Function: Apply the recommendation results to national government affairs platform applications to improve user experience.

[0110] Implementation method: 1. After a user enters the National Government Affairs Platform application, the system calls the recommendation system API interface based on parameters such as user ID and insurance registration location code to obtain a personalized recommendation service list, and displays N recommended services on the application homepage.

[0111] 2. Users can select their insurance location and preference tags through the application interface, and the system will dynamically adjust the recommendation results based on the user's settings.

[0112] 3. If the user chooses to turn off the personalized recommendation function, the system will only apply the top recommendation and popular recommendation strategies, still displaying N service lists, but without including recently used recommendations and tag matching recommendations.

[0113] like Figure 2 As shown, this is applicable to the National Government Affairs Platform. It interacts with the National Government Affairs Platform application through an API interface. Input parameters include user ID, social security registration location code, etc., and the output parameter is a personalized list of recommended services for the current user (N services are recommended by default). Core functions include user behavior data collection, recommendation strategy calculation, strategy fusion, and recommendation result output.

[0114] like Figure 4 As shown, the purpose of this application is to provide a multi-strategy fusion recommendation method to support the implementation process of personalized service recommendation operation method on national government affairs platform applications. The implementation process involves the following modules and their functions.

[0115] Front-end event tracking module: Function: Embed tracking code in the front end of national government affairs platform applications (APP, H5, mini-program) to record user service click behavior.

[0116] Implementation method: The event tracking code sends user behavior logs to the log server in real time, and the log server stores them to the disk using a rolling write method.

[0117] Cloud log collection module: Function: A log service that unifies the collection of behavior logs scattered across multiple log servers and brings them to a private cloud service.

[0118] Implementation method: Deploy probes on each log server to synchronize logs to the cloud service's log service, and create a log library to store the raw logs.

[0119] Data acquisition module: Function: The collected log data is split into two streams for real-time calculation and offline analysis.

[0120] Implementation: Deploy a distributed log collection module, develop a custom Source to consume the log library of the cloud log service, start the log collection module task process, where two processes form a consumer group to write the consumed data to the message queue, and another two processes form a consumer group to write the consumed data to HDFS.

[0121] Message middleware module: Function: Provides a high-throughput message queue for real-time data transmission.

[0122] Implementation: Create a Topic in the message queue to store logs, set the number of partitions to 5 and the number of replicas to 3, to ensure high availability and concurrent processing capabilities of the data.

[0123] Real-time computing module: Functionality: Processes user behavior data in real time, supporting the calculation of recent usage recommendations and tag matching recommendations.

[0124] Implementation method: Task 1: Consume Topic data from the message queue to record user click behavior on services. Fields include user ID, service ID, number of clicks, and update time, used for recent usage recommendations and tag matching recommendations.

[0125] Task 2: Consume Topic data from the message queue and execute the tag matching recommendation calculation logic. Specific steps include: 1. Based on the user ID, retrieve the services and number of clicks the user has historically clicked.

[0126] 2. Associate the service tag table and calculate the user tag score.

[0127] 3. Write the user tag scores into the MySQL database of the strategy fusion module. The record fields include user ID and recommendation list. The recommendation list format is: [{labelId:service1, labelScore:60}, {labelId:service2, labelScore:80}, ...].

[0128] Offline computing module: Function: Performs offline analysis of historical behavior data and supports the calculation of popular recommendations.

[0129] Implementation method: 1. Parse the behavior logs in HDFS and write them to Hive tables, then use the Tez execution engine to improve query performance.

[0130] 2. Using a T+1 cycle scheduling task, calculate the click volume of national services and local services in the past N days, generate popular recommendation results, and write them to the MySQL database of the strategy fusion module.

[0131] Strategy Management Module: Function: Provides an interface-based entry point for operations staff to adjust recommendation strategies.

[0132] Implementation method: Through the management interface, operators can set information such as top services, tag scores, and recommendation coefficients, and store the settings results in the MySQL database of the strategy fusion module.

[0133] Strategy fusion module: Function: To perform fusion calculations of multiple recommendation strategies and generate the final recommendation result.

[0134] Implementation: The Mybatis module of the Spring Boot framework is used to read the strategy data in MySQL, calculate the recommendation results according to the fusion logic (deduplication and sorting), and write the results into the MySQL result table. The fields include user ID and the list of recommended services. The recommendation list format is: [service 1, service 2, ..., service 15].

[0135] System interface service module: Function: An API interface that provides recommendation services for national government affairs platform applications.

[0136] Implementation: The API interface input parameters include user ID, insurance location code, and whether the recommendation status is turned off, etc. The output parameter is a list of recommended services.

[0137] Key points of this application include: 1. Design of a multi-strategy fusion recommendation method: In the context of government services where user behavior data is sparse, a personalized recommendation method that integrates four recommendation strategies—"top recommendation, recently used recommendation, tag matching recommendation, and popular recommendation"—is proposed for the first time. Through strategy priority ranking, deduplication mechanism, and weight fusion algorithm, the accuracy, coverage, and fairness of the recommendation results are ensured.

[0138] 2. Dynamic modeling mechanism for user preference tags: Combining user-defined preferences and behavioral data, a dynamic and updatable user tag system is constructed. Based on service tag scores and click behavior, a weighted formula is used to calculate user preference scores, enabling real-time updates and accurate modeling of user tags.

[0139] 3. National and Local Service Recommendation Weight Balancing Mechanism: To address the issue of low exposure for local services, a "recommendation weight coefficient" mechanism is proposed. Operations personnel can configure the weight coefficients for national and local services through the operations management interface to achieve service balance adjustment in the recommendation algorithm, thereby improving the fairness of local service recommendations and user experience.

[0140] 4. Intelligent degradation and fallback mechanism for recommendation strategy: In cases where users do not select service preference tags or there is no user behavior data, the system automatically degrades to the "top recommendation + popular recommendation" strategy to ensure the basic availability of recommendation results and improve system robustness.

[0141] 5. Design for interpretability and operational controllability of recommendation results: The recommendation logic is based on rule and tag matching algorithms, and the results are traceable and interpretable; at the same time, an operation management interface is provided to support operators to flexibly configure top services, tag scores, and weight coefficients, thereby enhancing the maintainability and controllability of the system.

[0142] The points of protection in this application are: The protection points of this application cover the core computational logic of the recommendation algorithm, the policy fusion mechanism, and the overall implementation method and module collaboration process of the system architecture, specifically including the following: 1. Implementation logic and process of multi-strategy fusion recommendation method: Protected Content: A multi-strategy fusion recommendation method suitable for data-sparse government service scenarios, comprising the following steps: (1) Prioritize the output of the top recommendation.

[0143] (2) Recommendation of recently used services based on user click behavior.

[0144] (3) Tag matching recommendation based on the matching degree between users and service tags.

[0145] (4) Recommendation of popular services based on service click popularity.

[0146] (5) Strategy fusion and deduplication mechanism to form the final recommendation list.

[0147] Key areas of protection: Priority ranking rules for recommending strategy fusion.

[0148] Deduplication strategy logic.

[0149] National and local service ratio adjustment methods.

[0150] Recommendation list generation method and other core algorithm processes.

[0151] 2. Dynamic construction and matching methods for user preference tags: Protected content: A method for generating user preference tags by combining user-initiated settings and behavioral data inference, which calculates tag scores based on a weighted average of behavioral click counts and service tag scores, and then matches the tags with the service tags to calculate the recommendation matching degree.

[0152] Key areas of protection: (1) Formula for calculating user tag score.

[0153] (2) Calculation method of service matching degree.

[0154] (2) The weighting adjustment coefficient setting mechanism for national and local services in the tag score.

[0155] (4) Sorting and filtering methods for tag matching recommendation results.

[0156] 3. Recommend a fairness guarantee mechanism: Protection content: In response to the problem of low click-through rates and low recommendation probability for local services, a dynamic balance mechanism is proposed to be achieved by setting a service tag score weighting coefficient to realize the recommendation opportunities of local services and national services.

[0157] Key areas of protection: (1) Method for setting the service recommendation weight coefficient.

[0158] (2) Weighting adjustment method of coefficients in behavior analysis and label matching process.

[0159] (3) Recommend a ratio control mechanism for local services and national services during the integration process (such as a 1:1 allocation strategy).

[0160] 4. Recommendation system architecture design and module collaboration mechanism: Protected content: A personalized service recommendation system for a national government affairs platform built on a big data technology stack. The system includes a front-end data tracking module, a log collection module, a data distribution module, a message middleware module, a real-time computing module, an offline analysis module, a strategy fusion module, and an interface service module.

[0161] Key areas of protection: (1) Data flow logic between modules.

[0162] (2) The division of labor and cooperation mechanism between real-time computing and offline analysis tasks.

[0163] (3) The generation, storage and retrieval of recommendation results in the system.

[0164] (4) API interface structure and calling logic of recommendation service.

[0165] 5. Downgrade and fallback mechanisms for recommendation strategies: Content protection: In scenarios where users have not selected service preference tags or there is no user behavior data, the system automatically adopts a top recommendation and popular recommendation strategy to ensure the integrity and relevance of the recommendation list.

[0166] Key areas of protection: The strategy degradation logic of the recommendation system.

[0167] The mechanism for enabling the fallback recommendation strategy.

[0168] How to handle whether a user chooses to turn off personalized recommendations.

[0169] The protection points of this application not only cover the core computational logic and strategy fusion mechanism of the recommendation algorithm, but also the overall implementation method of the system architecture and the module collaboration process. At the same time, in response to the pain points of data sparsity and regional differences in government service scenarios, it innovatively proposes a multi-strategy fusion recommendation and recommendation fairness adjustment mechanism, which has broad applicability and promotion value, and can simultaneously ensure the practicality, controllability and stability of the recommendation system at the technical and operational levels.

[0170] While the "Multi-Strategy Fusion Recommendation Method and System Based on Big Data" proposed in this application has demonstrated superior practicality and adaptability in government service scenarios such as national government service platforms, theoretically, several other technical approaches can still be used to achieve similar recommendation goals. The following describes several typical alternative solutions and compares them with this application to further highlight the advantages and innovation of the proposed solution: Alternative Option 1: Deep Learning-Based Recommendation Model Solution Description: Utilize deep neural networks (such as Wide&Deep, DeepFM, etc.) to learn high-dimensional feature representations of users and services, train the model to predict the probability of user preference for services, and generate recommendation results accordingly.

[0171] advantage: It can automatically learn the deep feature combinations between users and services and uncover potential relationships.

[0172] In scenarios with sufficient data, the recommendation accuracy is relatively high.

[0173] shortcoming: 1. A large amount of training data is required, but user behavior data in the application scenarios of national government affairs platforms is relatively sparse, which may lead to poor model performance.

[0174] 2. The model has high complexity and high training and inference costs, making it difficult to meet the needs of real-time recommendation.

[0175] The interpretability of the recommendation results is poor, making it difficult to meet the needs of operations personnel to adjust the recommendation strategy.

[0176] Comparison with this application: This application adopts a multi-strategy fusion recommendation method. Through rule-based recommendation logic and dynamic label matching algorithm, it can achieve fast, accurate and real-time recommendation in data-sparse environment, and has strong interpretability and operational controllability, which is significantly better than the applicability of deep learning models in this scenario.

[0177] Alternative Solution 2: Knowledge Graph-Based Recommendation Solution Description: Construct a knowledge graph for the service domain of the national government affairs platform, model entities such as users, services, policies, and service guides and their relationships, and use the reasoning capabilities of the knowledge graph to discover potential connections between users and services and generate recommendation results.

[0178] advantage: 1. It can incorporate external knowledge (such as policies, regulations, and service guides) to improve the accuracy and interpretability of recommendation results.

[0179] It is suitable for uncovering deep semantic relationships between users and services.

[0180] shortcoming: 1. The construction and maintenance of knowledge graphs are costly, especially in the context of national government affairs platforms, where there is a lack of external data such as policies and service guides, making it difficult to form a complete knowledge system.

[0181] 2. The computational complexity of inference is high, making it difficult to meet the needs of real-time recommendations.

[0182] A weak reliance on user behavior data may lead to recommendations that are out of touch with users' actual needs.

[0183] Comparison with this application: This application does not rely on external knowledge systems and can achieve efficient recommendations based entirely on user behavior and service tags. It has low deployment and maintenance costs, strong real-time performance, and is suitable for rapid iteration and deployment in government service platform scenarios.

[0184] Alternative Solution 3: Recommendation Based on Graph Neural Networks Solution Description: Users and services are constructed as a graph structure. Graph Neural Networks (GNNs) are used to learn the feature representations of user and service nodes. Graph embedding techniques are used to predict user preferences for services and generate recommendation results.

[0185] advantage: 1. It can capture complex relationships between users and services, and between services.

[0186] 2. Applicable to mining potential associations in graph structures.

[0187] shortcoming: 1. In the application scenarios of the national government affairs platform, user behavior data is sparse, the number of services is small, and the scale of the graph structure is limited, which makes it difficult to train the graph neural network and the model converges slowly.

[0188] 2. The computational complexity is high, making it difficult to meet the needs of real-time recommendations.

[0189] High data quality is required; data noise can significantly affect model performance.

[0190] Comparison with this application: Graph neural network recommendation models are more suitable for highly active user groups and large-scale graph structure data environments, while this application, through label matching and policy fusion, can better adapt to the "low-frequency, high-value" usage characteristics of government service platforms, and has stronger recommendation efficiency and application adaptability.

[0191] in conclusion: While the aforementioned alternatives may have advantages in certain high-frequency internet applications, in the field of government services such as national government platforms, which are characterized by sparse user behavior, high demand diversity, and large regional differences, this application achieves high-quality recommendation output under limited data conditions through "multi-strategy integration + dynamic tag modeling + recommendation fairness adjustment mechanism," and has better practicality, scalability, and promotional value.

[0192] To ensure the core technological advantages of this application in the national government affairs platform and other government service platforms, the applicant proposes the following patent application requirements: Scope of patent protection: The applicant wishes to seek comprehensive and systematic patent protection for the following technical content: 1. Personalized service recommendation methods: 1.1 Involves a multi-strategy fusion recommendation method (top recommendation, recently used recommendation, tag matching recommendation, popular recommendation).

[0193] 1.2 Recommendation strategy integration logic, priority ranking, and service deduplication mechanism.

[0194] 1.3 Recommended proportion control method (national service and local service balance mechanism).

[0195] 2. User tag modeling and matching mechanism: 2.1 Includes user behavior analysis, tag score calculation, and matching degree formula.

[0196] 2.2 Adjustment mechanism for weight coefficients of national and local service labels.

[0197] 3. Recommendation System Architecture Design: 3.1 Includes all functional modules of the recommendation system (tracking, data collection, traffic splitting, calculation, fusion, and API interface).

[0198] 3.2 Data flow and collaboration mechanism between modules.

[0199] 3.3 The linkage between real-time and offline computing.

[0200] 4. Downgrade and fallback mechanisms for recommendation strategies: 4.1 Logic for adjusting recommendation strategies for users in cold start scenarios or scenarios with missing behaviors.

[0201] 4.2 Default display scheme after disabling personalized recommendations.

[0202] 5. Operational controllability and explainability of recommendation design: 5.1 Includes features such as visual configuration of recommended content, tag management, and recommendation weight adjustment interface.

[0203] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 5 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database is used for multi-policy fusion. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. The computer program is executed by the processor to implement multi-policy fusion.

[0204] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0205] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0206] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0207] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0208] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0209] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0210] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0211] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0212] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A multi-strategy fusion recommendation method, characterized in that, The multi-strategy fusion recommendation method includes: Collect user click behavior data related to the service; The click behavior data is processed, and the processed click behavior data is stored in the Hive data warehouse; Based on the Hive data warehouse, different service recommendation strategies are determined; the service recommendation strategies include a top service recommendation strategy, a recently used service recommendation strategy, a tag matching service recommendation strategy, and a popular service recommendation strategy. By integrating different service recommendation strategies, a list of integrated service recommendation strategies is determined.

2. The multi-strategy fusion recommendation method based on big data according to claim 1, characterized in that, Collect user click behavior data related to the service, specifically including: Embed tracking code in the front end of web applications; Based on the tracking code, user click behavior data related to the service is collected and recorded in real time to generate log records; the log records include user ID, service name, click event and insurance location code.

3. The multi-strategy fusion recommendation method according to claim 1, characterized in that, The click behavior data is processed, and the processed click behavior data is stored in a Hive data warehouse, specifically including: Click behavior data is collected from log records via cloud log service; The click behavior data is stored in the created log database; The click behavior data in the log library is split into two parts: the first part of the click behavior data is written to a message queue, and the second part of the click behavior data is written to the Hadoop Distributed File System. The click behavior data in the Hadoop Distributed File System is parsed, and the parsed click behavior data is written to the Hive data warehouse.

4. The multi-strategy fusion recommendation method according to claim 1, characterized in that, Based on the Hive data warehouse, different service recommendation strategies are determined, specifically including: Based on the Hive data warehouse, the operations staff selects n services for top recommendation and determines the top service recommendation strategy. Recommendations are made based on services that users have recently used frequently, and a recommendation strategy for recently used services is determined. Recommendations are made based on the tag matching degree between users and services, and a tag matching service recommendation strategy is determined. Recommendations are made based on the number of clicks for each service, and a strategy for recommending popular services is determined.

5. The multi-strategy fusion recommendation method according to claim 4, characterized in that, Based on the Hive data warehouse, operations personnel select n services for top recommendation and determine the top service recommendation strategy, which specifically includes: The operations staff selects n services as top-recommended services and sets the order in which they are topped. The recommendation results of the pinning order are output to the pinning service recommendation strategy table for pinning recommendation.

6. The multi-strategy fusion recommendation method according to claim 4, characterized in that, Recommendations are made based on services that users have recently used frequently. The recommendation strategy for recently used services includes: Extract user click behavior data related to the service from the Hive data warehouse over the past N days; Based on the click behavior data, retain the last record of each service category that the user recently clicked; The last record in each service category is sorted from most recent to oldest according to the user's click time to generate a click sequence; Recommendation results are determined based on the generated click sequence; The recommendation results are then output to the recently used service recommendation strategy table for further recommendation.

7. The multi-strategy fusion recommendation method according to claim 4, characterized in that, Recommendations are made based on the tag matching degree between users and services, and a tag matching service recommendation strategy is determined, specifically including: Set service tags and determine the score for clicking on service tags; Based on the click service tag score, determine the user tag score and the service tag score; the user tag score = user-actively set score + coefficient × click service tag score × cumulative number of clicks; the service tag score = coefficient × service tag score; The degree of tag matching between users and services is determined based on user tag scores and service tag scores; The user and service tag matching degree is output to the tag matching service recommendation strategy table for recommendation.

8. The multi-strategy fusion recommendation method according to claim 4, characterized in that, Recommendations are made based on service click-through rates, and a strategy for recommending popular services is determined, including: Based on the Hive data warehouse, extract click behavior data related to the service over the past N days; Based on the click behavior data, the click volume of national services and local services is calculated respectively; Sort the click volume of national services and local services from highest to lowest, record the serial numbers, and determine the sorting sequence; the serial numbers are 1, 2, 3, ..., m, where m is the total number of serial numbers; According to the sorting sequence, the services with clicks corresponding to those with serial numbers less than a set serial number are output to the popular service recommendation strategy table for recommendation; the services are national services or local services.

9. The multi-strategy fusion recommendation method according to claim 8, characterized in that, By integrating different service recommendation strategies, a list of integrated service recommendation strategies is determined, including: Add the selected n top-rated services to the recently used service recommendation strategy table in order; Remove services that are duplicated in the recently used service recommendation strategy table from the top service recommendation strategy table to determine the first service table; Sort the first service table by click time from most recent to oldest, and select n services to add to the tag matching service recommendation strategy table; Remove services from the tag matching service recommendation strategy table that are duplicated in the top service recommendation strategy table and the recently used service recommendation strategy table to determine the second service table; Sort the second service table from high to low according to the matching degree, and select n services to add to the popular service recommendation strategy table; Remove services from the popular service recommendation strategy table that are duplicated with the top service recommendation strategy table, the recently used service recommendation strategy table, and the tag matching service recommendation strategy table to determine the third service table; The third service table is sorted from high to low popularity and added to the recommendation list to determine the integrated service recommendation strategy table.

10. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the multi-strategy fusion recommendation method according to any one of claims 1-9.

Images