Digital public service integration platform and equipment based on intelligent recommendation and medium

Abstract

The invention relates to the technical field of digital public services, in particular to a digital public service integration platform and device based on intelligent recommendation and a medium. The method comprises the following steps: firstly, collecting and analyzing basic information and behavior data of a user in real time through a user portrait construction module, and constructing a multi-dimensional user feature vector; secondly, performing structured processing on the government affair service data by using a service feature extraction module, and extracting a service association relationship and scene features; then realizing accurate matching of the user feature vector and the service feature matrix based on an intelligent recommendation engine; and finally, continuously collecting user feedback and dynamically adjusting a recommendation strategy through a real-time feedback optimization module. According to the method, scene-based precise pushing of the government affair service is realized, a continuous optimization mechanism of the recommendation effect is established, the recommendation strategy can be dynamically adjusted according to the user characteristics and behaviors, and the intelligent level and user experience of the government affair service are remarkably improved.

Description

Technical Field

[0001] This application relates to the technical field of digital public services, in particular to a digital public service integration platform, device, and medium based on intelligent recommendation. Background Art

[0002] With the in-depth promotion of the digital transformation of government services, the public service supply model is changing from passive response to active service. The diversity of government service scenarios, the differences in user needs, and the complexity of service content have put forward higher requirements for accurate recommendation. How to achieve intelligent and accurate push of government services and improve service efficiency has become an important issue in the construction of digital government.

[0003] The existing technology usually adopts a recommendation method based on service type classification. The system classifies and displays government services according to preset service classification rules in dimensions such as functions, departments, and fields. When a user accesses a certain type of service, the system automatically pushes relevant services under this category to achieve automated recommendation of services.

[0004] However, this recommendation method based on fixed classification ignores the personalized needs of users and cannot dynamically adjust the recommendation strategy according to user characteristics and behaviors. This situation needs to be further improved. Summary of the Invention

[0005] In order to solve the problem that the existing recommendation method based on fixed classification cannot dynamically adjust the recommendation strategy according to user characteristics and behaviors, this application provides a digital public service integration platform, device, and medium based on intelligent recommendation, and adopts the following technical solutions: In a first aspect, this application provides a digital public service integration platform based on intelligent recommendation, including: A user portrait construction module, which is used to collect user basic information and behavior data to obtain a user feature vector; A service feature extraction module, which is used to perform structured processing on government service data to obtain a service feature matrix; An intelligent recommendation engine, which is used to perform matching calculations based on the user feature vector and the service feature matrix to obtain an initial recommendation result; A real-time feedback optimization module, which is used to collect user feedback based on the initial recommendation result to obtain an optimized final recommendation result.

[0006] By adopting the above technical solution, the present application first constructs a multi-dimensional user feature vector by the user portrait construction module through real-time collection and analysis of the user's basic information and behavior data; secondly, uses the service feature extraction module to perform structured processing on the government service data to extract service association relationships and scenario features; then, based on the intelligent recommendation engine, realizes the accurate matching of the user feature vector and the service feature matrix; finally, continuously collects user feedback and dynamically adjusts the recommendation strategy through the real-time feedback optimization module; not only realizes the scenario-based accurate push of government services, but also establishes a continuous optimization mechanism for the recommendation effect, can dynamically adjust the recommendation strategy according to user characteristics and behaviors, and significantly improves the intelligent level and user experience of government services.

[0007] Optionally, the user portrait construction module includes: A static feature extraction unit for obtaining the user's basic attribute information to obtain a static feature set; A dynamic feature extraction unit for collecting the user's historical behavior data to obtain a dynamic feature set; A scenario feature extraction unit for obtaining a scenario feature set according to the dynamic feature set and the user's current state; A feature fusion unit for calculating and obtaining a fusion feature vector according to the static feature set, the dynamic feature set and the scenario feature set.

[0008] By adopting the above technical solution, the present application first obtains basic attribute information such as enterprise type and registered capital through the static feature extraction unit to construct a basic user portrait; secondly, uses the dynamic feature extraction unit to collect time-series data such as the user's historical access tracks and operation behaviors to capture changes in user needs; then, based on the scenario feature extraction unit, combines the dynamic features with the user's current state to identify the specific business scenario where the user is located; finally, through the feature fusion unit, using a fusion algorithm with adaptive weights, organically integrates the static feature set, the dynamic feature set and the scenario feature set to generate an all-round user feature vector; not only realizes the multi-dimensional accurate characterization of user features, but also establishes a dynamic association mechanism between features.

[0009] Optionally, the dynamic feature extraction unit includes: A browsing record analyzer for extracting the user's access data to obtain browsing features; An interaction behavior analyzer for analyzing the operation mode according to the browsing features to obtain behavior features; A time-series feature extractor for calculating time dimension features based on the behavior features; A feature fuser for obtaining comprehensive dynamic features according to the browsing features, the behavior features and the time dimension features.

[0010] By adopting the above technical solution, the present application first conducts a fine-grained analysis of the user's page access data through a browsing record analyzer, and extracts browsing features including access path, stay duration, click depth, etc.; secondly, uses an interaction behavior analyzer to construct a user operation sequence based on the browsing features, and applies a sequence pattern mining algorithm to identify typical operation patterns; then introduces a time decay function through a time series feature extractor to calculate the weight distribution of behavior features in different time windows; finally, the feature fusion unit adopts an attention mechanism to dynamically integrate the browsing features, behavior features, and time dimension features to generate comprehensive dynamic features expressing the evolution law of user behavior; not only realizes the multi-level analysis of user dynamic behavior, but also establishes a time series correlation mechanism for behavior features.

[0011] Optionally, the intelligent recommendation engine includes: A content matching unit, configured to calculate the similarity between the user feature vector and the service feature matrix to obtain a content matching score; A collaborative filtering unit, configured to analyze the similarity of the user group based on the content matching score to obtain a collaborative recommendation result; A rule filtering unit, configured to apply business rules based on the collaborative recommendation result to obtain a filtered result; A multi-strategy fusion unit, configured to perform a comprehensive ranking based on the filtered result to determine an initial recommendation result.

[0012] By adopting the above technical solution, the present application first uses a deep semantic matching model through the content matching unit to calculate the similarity between the user feature vector and the service feature matrix to obtain a preliminary content matching score; secondly, uses the collaborative filtering unit to construct a user-service interaction network based on the content matching score to mine user groups with similar features and behavior patterns; then introduces a business rule knowledge base through the rule filtering unit to perform compliance filtering and constraint verification on the recommendation result; finally, the multi-strategy fusion unit adopts an online learning algorithm to dynamically adjust the weights of each strategy, and organically integrates the recommendation results of different dimensions to form a final recommendation list that not only meets the personalized needs but also satisfies the business specifications; not only realizes the organic unity of personalized recommendation and rule constraint, but also establishes a dynamic tuning mechanism for recommendation strategies, significantly improving the practicality and reliability of the recommendation service.

[0013] Optionally, the content matching unit includes: A feature extraction sub-unit, configured to extract a static feature set, a dynamic feature set, and a scenario feature set from the user feature vector to obtain a feature representation; A matching calculation sub-unit, configured to calculate the dimensional similarity, time series correlation, and scenario matching degree between the feature representation and the service feature matrix to obtain a multi-dimensional matching value; A score generation subunit, configured to perform weighted fusion according to the multi-dimensional matching value and calculate a content matching score.

[0014] By adopting the above technical solution, the present application first hierarchically deconstructs the user feature vector through a feature extraction subunit, and respectively extracts a static feature set reflecting the user's basic attributes, a dynamic feature set representing the user's behavior evolution, and a scenario feature set depicting the current demand; secondly, a special similarity calculation model is designed by the matching calculation subunit to calculate the dimension similarity, time series correlation, and scenario matching degree between different feature sets and the service feature matrix respectively; finally, the score generation subunit adopts an adaptive weight fusion algorithm to dynamically adjust the weights of the matching values of each dimension according to the feature importance and timeliness, and generate a comprehensive content matching score; not only realizes the multi-dimensional accurate calculation of feature matching, but also establishes a dynamic fusion mechanism for the matching results.

[0015] Optionally, the real-time feedback optimization module includes: A feedback collection unit, configured to collect user interaction data based on the initial recommendation result to obtain original feedback data; A weight calculation unit, configured to calculate a feature weight coefficient according to the original feedback data; A parameter update unit, configured to determine user-service similarity calculation parameters, collaborative filtering threshold parameters, and rule filtering condition parameters based on the feature weight coefficient; A sorting optimization unit, configured to obtain an optimized final recommendation result according to the similarity calculation parameters, collaborative filtering threshold parameters, and rule filtering condition parameters.

[0016] By adopting the above technical solution, the present application first captures multi-dimensional interaction data such as clicks, stays, and conversions of users on the recommendation result in real time through the feedback collection unit to construct a complete user feedback sample; secondly, the weight calculation unit trains a feature importance evaluation model based on the feedback data to dynamically calculate the weight coefficients of different features in the current scenario; then, the parameter update unit adaptively maps the change of the feature weight to the core parameters of the recommendation model, including similarity calculation parameters, collaborative filtering threshold, and rule filtering conditions; finally, the sorting optimization unit adopts an online learning algorithm to re-adjust the sorting strategy of the recommendation result according to the updated parameters, realizing the real-time optimization of the recommendation effect; not only realizes the dynamic adaptive adjustment of the recommendation model, but also establishes a real-time evaluation mechanism for feature importance, significantly improving the timeliness and accuracy of the recommendation system.

[0017] Optionally, the parameter update unit includes: A pre- and post-update data acquirer, configured to acquire pre-parameter-update recommendation effect data within a first preset time period and post-parameter-update recommendation effect data within a second preset time period; A recommendation effect change calculator is used to obtain recommendation effect change data based on the recommendation effect data before parameter update and the recommendation effect data after parameter update.

[0018] By adopting the above technical solution, the present application first collects the recommendation effect data in the first preset time period before parameter update and the second preset time period after parameter update through the pre- and post-update data acquirers, including multi-dimensional indicators such as click-through rate, conversion rate, and user satisfaction; secondly, uses the recommendation effect change calculator to design a comparative analysis model considering time periodicity, and precisely calculates the effect change value before and after parameter update by eliminating interference factors such as basic traffic fluctuations and seasonal changes; and establishes a standardized measurement system for the recommendation effect.

[0019] Optionally, it further includes a recommendation quality monitoring module, and the recommendation quality monitoring module includes: A recommendation stability coefficient acquisition unit is used to acquire the fluctuation data of the recommendation results in the third preset time period in real time, and obtain the recommendation stability coefficient according to the fluctuation data; A user feedback abnormal number acquisition unit is used to acquire the number of times of abnormal triggers in user feedback in real time according to the recommendation effect change data; A system warning information sending unit is used to send system warning information to the management terminal when the number of times of abnormal triggers in user feedback is greater than a preset abnormal threshold or the recommendation stability coefficient is greater than a preset fluctuation threshold.

[0020] By adopting the above technical solution, the present application first monitors the change trend of the recommendation results in the third preset time period in real time through the recommendation stability coefficient acquisition unit, and calculates the fluctuation degree and stability coefficient of the recommendation results based on the time series analysis model; secondly, designs a multi-dimensional abnormal detection model by using the user feedback abnormal number acquisition unit to statistically analyze the abnormal patterns and trigger frequencies in user feedback in real time; finally, establishes a dual-threshold trigger mechanism through the system warning information sending unit, and automatically pushes warning information to the management terminal and starts an emergency response process when the number of times of user feedback abnormalities or the recommendation stability coefficient exceeds the preset threshold; realizes the real-time monitoring of the recommendation quality, and establishes an active intervention mechanism for abnormal warnings, effectively improving the operation reliability and service guarantee ability of the recommendation system.

[0021] In a second aspect, the present application provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the steps performed by the above-mentioned digital public service integration platform based on intelligent recommendation are implemented.

[0022] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps performed by the above-mentioned digital public service integration platform based on intelligent recommendation.

[0023] In summary, the present application includes at least one of the following beneficial technical effects: 1. This application first collects and analyzes the user's basic information and behavior data in real time through the user portrait construction module to construct a multi-dimensional user feature vector; secondly, the service feature extraction module is used to perform structured processing on the government service data to extract service associations and scenario features; then, the user feature vector and the service feature matrix are accurately matched based on the intelligent recommendation engine; finally, the real-time feedback optimization module is used to continuously collect user feedback and dynamically adjust the recommendation strategy; not only the scenario-based accurate push of government services is realized, but also a continuous optimization mechanism for the recommendation effect is established, which can dynamically adjust the recommendation strategy according to user characteristics and behaviors, significantly improving the intelligence level and user experience of government services; 2. This application first obtains basic attribute information such as enterprise type and registered capital through a static feature extraction unit to build a basic user portrait; secondly, the dynamic feature extraction unit is used to collect time series data such as the user's historical access trajectory and operation behavior to capture changes in user needs; then based on the scene feature extraction unit, the dynamic features are combined with the user's current state to identify the specific business scenario in which the user is located; finally, through the feature fusion unit, an adaptive weight fusion algorithm is used to organically integrate the static feature set, dynamic feature set and scene feature set to generate a comprehensive user feature vector; not only does it achieve multi-dimensional and accurate characterization of user features, but it also establishes a dynamic association mechanism between features; 3. This application first uses a deep semantic matching model through a content matching unit to calculate the similarity between the user feature vector and the service feature matrix to obtain a preliminary content matching score; secondly, a collaborative filtering unit is used to build a user-service interaction network based on the content matching score to mine user groups with similar features and behavior patterns; then, a business rule knowledge base is introduced through a rule filtering unit to perform compliance filtering and constraint verification on the recommendation results; finally, a multi-strategy fusion unit uses an online learning algorithm to dynamically adjust the weights of each strategy, organically integrate the recommendation results of different dimensions, and form a final recommendation list that meets both personalized needs and business specifications; it not only realizes the organic unity of personalized recommendations and rule constraints, but also establishes a dynamic tuning mechanism for recommendation strategies, significantly improving the practicality and reliability of recommendation services. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a structural diagram of a digital public service integration platform based on intelligent recommendation according to an embodiment of the present application; Figure 2It is a schematic structural diagram of the user portrait construction module in the digital public service integration platform according to an embodiment of the present application; Figure 3 It is a schematic structural diagram of the dynamic feature extraction unit in the digital public service integration platform according to an embodiment of the present application; Figure 4 It is a schematic structural diagram of the intelligent recommendation engine in the digital public service integration platform according to an embodiment of the present application; Figure 5 It is a schematic structural diagram of the content matching unit in the digital public service integration platform according to an embodiment of the present application; Figure 6 It is a schematic structural diagram of the real-time feedback optimization module in the digital public service integration platform according to an embodiment of the present application; Figure 7 It is a schematic structural diagram of the parameter update unit in the digital public service integration platform according to an embodiment of the present application; Figure 8 It is a schematic structural diagram of the recommendation quality monitoring module in the digital public service integration platform according to an embodiment of the present application; Figure 9 It is an internal structural diagram of an electronic device according to an embodiment of the present application. Detailed implementation manners

[0025] The terms used in the following embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. As used in the specification and appended claims of the present application, the singular forms "a", "an", "the", "above", "said", "this" are also intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term " / and / " used in the present application refers to any or all possible combinations including one or more of the listed items.

[0026] Hereinafter, the terms "first" and "second" are only used for descriptive purposes, and cannot be construed as implying or indicating relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the embodiments of the present application, unless otherwise specified, the meaning of "a plurality" is two or more.

[0027] The following further describes the embodiments of the present application in detail with reference to the accompanying drawings of the specification.

[0028] In a first aspect, the present application provides a digital public service integration platform based on intelligent recommendation. Referring to Figure 1 , including: A user portrait construction module, configured to collect user basic information and behavior data to obtain user feature vectors.

[0029] In this embodiment, the user portrait construction module adopts a two-layer acquisition architecture of "static data + dynamic behavior". It regularly obtains user basic information from government affairs databases such as industry and commerce, taxation, and social security through preset data interfaces, and uses behavior acquisition scripts deployed on the government service platform to record behavior data such as users' service consultations and application for services in real time. The system has pre-established a feature mapping library, which contains feature extraction rules in dimensions such as enterprise entity features (such as registered capital, business scope), credit features (such as tax payment grade, illegal and irregular records), and business features (such as application frequency, service preferences).

[0030] Specifically, for enterprise users, the system first associates each government affairs database through the unified social credit code, extracts the basic information and converts it into standardized feature values according to the feature mapping rules. For example, numerical features such as registered capital and asset scale are normalized, and text features such as business scope and main business are converted into categorical features through a preset industry classification mapping table. For user behavior data, the system sets up a behavior-feature conversion matrix to convert the original behavior data such as users' search keywords, click paths, and residence duration into quantifiable feature indicators. Finally, the system integrates the static features and dynamic features into a unified feature vector for subsequent recommendation calculations.

[0031] The service feature extraction module is used to perform structured processing on government service data to obtain a service feature matrix.

[0032] In this embodiment, the service feature extraction module adopts a semi-automatic feature extraction scheme, and has pre-established a government service feature template library, which contains structured feature definitions in dimensions such as service objects, application conditions, material lists, and handling processes. The system extracts features in batches from the standardized government service data through template matching. For non-standardized service content, a feature annotation tool is used to assist business personnel for manual processing to ensure the accuracy and integrity of feature extraction.

[0033] Specifically, the system first preprocesses the government service data, including basic operations such as text segmentation and key information extraction. For policy documents with standardized formats, preset parsing templates are directly used to extract service features; for unstructured service descriptions, core elements are identified through keyword matching and rule mapping. For example, when extracting subsidy policy features, the system identifies application conditions through a preset conditional sentence pattern, identifies subsidy standards through a numerical pattern, and identifies handling periods through a time expression. All extracted features are converted into standardized numerical representations and stored in the service feature matrix, and each service corresponds to a feature vector in the matrix.

[0034] The intelligent recommendation engine is used to perform matching calculations based on the user feature vector and the service feature matrix to obtain an initial recommendation result.

[0035] In this embodiment, the intelligent recommendation engine adopts a lightweight recommendation strategy, and realizes the rapid matching of the user feature vector and the service feature matrix through a preset similarity calculation rule and service matching threshold. The system establishes a basic cosine similarity calculation model and combines a rule filtering mechanism to reduce the calculation complexity while ensuring the recommendation accuracy.

[0036] Specifically, the recommendation engine first calculates the similarity scores between the user feature vector and each service vector in the service feature matrix. The system has pre-set a feature weight configuration table to assign different weight coefficients to different types of features. For example, in the enterprise subsidy recommendation scenario, the weight of rigid conditions such as enterprise scale and industry type is relatively high, while the weight of soft features such as historical behavior is relatively low. The similarity scores obtained through weighted calculation are filtered by a preset threshold to form an initial recommendation list. At the same time, the system maintains a rule library for post-processing the recommendation results, such as removing expired services and adjusting the display order.

[0037] The real-time feedback optimization module is used to collect user feedback based on the initial recommendation results to obtain the optimized final recommendation results.

[0038] In this embodiment, the real-time feedback optimization module adopts a feedback collection and optimization strategy, and collects user feedback data on the recommendation results through a preset evaluation index system, including quantitative indexes such as click-through rate, application rate, and completion rate. The system establishes a basic evaluation rule library for judging the recommendation effect and triggering the optimization strategy.

[0039] Specifically, the system records data such as the display times, click times, and application conversion of the recommended service through a user behavior tracking script, and sets a simple counter to count each index. When it is detected that the feedback index of a certain type of user for a specific service is abnormal (such as no click after continuous multiple displays), the system automatically adjusts the recommendation weight of this type of service according to the preset optimization rules. For example, reduce the recommendation frequency of services with poor feedback and increase the recommendation priority of similar services with good user feedback.

[0040] In one embodiment, referring to Figure 2 , the user portrait construction module includes: The static feature extraction unit is used to obtain the user's basic attribute information to obtain a static feature set; The dynamic feature extraction unit is used to collect the user's historical behavior data to obtain a dynamic feature set; The scenario feature extraction unit is used to obtain a scenario feature set according to the dynamic feature set and the user's current state; The feature fusion unit is used to calculate and obtain a fusion feature vector according to the static feature set, the dynamic feature set, and the scenario feature set.

[0041] In this embodiment, the user portrait construction module constructs a multi-dimensional user portrait through the collaborative work of the static feature extraction unit, the dynamic feature extraction unit, the scenario feature extraction unit, and the feature fusion unit. The system has pre-established a user feature classification library, which classifies features into three categories: basic attribute features, historical behavior features, and scenario association features, and corresponding feature extraction rules and quantization indicators are designed for each category of features. Among them, the basic attribute features mainly include fixed features such as user registration information and authentication information; the historical behavior features include dynamic features such as service usage records and interaction behaviors; the scenario association features reflect the temporary features of users in specific service scenarios.

[0042] Specifically, the static feature extraction unit obtains basic attribute information from the user information library through a preset data interface. For enterprise users, the system has established an enterprise basic information mapping table to convert raw data such as registered capital, number of employees, and business scope into standardized feature values. The dynamic feature extraction unit maintains a behavior-feature conversion rule library, and real-time collects and processes user behavior data through a behavior log analysis script. For example, it converts service browsing records into service preference features and converts the frequency of handling affairs into activity features. The scenario feature extraction unit has preset scenario feature templates, and dynamically calculates scenario correlation features based on the user's current state (such as the type of service being applied for and the handling stage) and historical behavior patterns. The feature fusion unit adopts a weighted fusion method. The system has preset a weight configuration table for various features, assigns different weights to features with different timeliness and importance, and generates the final user feature vector through weighted calculation.

[0043] In one embodiment, referring to Figure 3 , the dynamic feature extraction unit includes: A browsing record analyzer for extracting user access data to obtain browsing features; An interaction behavior analyzer for analyzing the operation mode based on the browsing features to obtain behavior features; A time series feature extractor for calculating time dimension features based on the behavior features; A feature fuser for obtaining comprehensive dynamic features based on the browsing features, behavior features, and time dimension features.

[0044] In this embodiment, the dynamic feature extraction unit adopts a multi-level feature analysis architecture. Through the step-by-step processing of the browsing record analyzer, interaction behavior analyzer, time-series feature extractor, and feature fusion unit, the refined extraction of user dynamic features is realized. The system has pre-established a user behavior feature library, including a page access feature mapping table, an operation behavior coding table, and time-series feature calculation rules, which are used for standardizing various dynamic data. The browsing record analyzer is responsible for recording the user's access trajectory on the platform. The interaction behavior analyzer identifies the user's operation patterns. The time-series feature extractor calculates the time distribution features of the behaviors. Finally, the feature fusion unit integrates the features of each dimension into a unified dynamic feature representation.

[0045] Specifically, the browsing record analyzer collects the user's page access data through data collection scripts deployed on each page of the platform, including raw information such as the type of accessed page, stay duration, click position, etc. The system has pre-set a page type coding table to map the pages of different service modules to standardized category codes. At the same time, a stay duration segmentation rule is set to convert the original duration data into discrete attention level features. The interaction behavior analyzer performs pattern matching on the user's click sequence based on the pre-set operation pattern recognition rules, and identifies typical operation behaviors such as "information query", "material upload", "form filling", etc. The system maintains a behavior-purpose mapping table to convert the identified operation behaviors into user intention features. The time-series feature extractor adopts a simple time window statistical method, and pre-sets the feature calculation rules for three time granularities of day, week, and month to statistically analyze the behavior frequency and patterns of users on different time scales. The feature fusion unit integrates the features of each dimension according to the pre-set feature combination template. For example, for the policy query behavior of enterprise users, the system records basic features such as query frequency, keywords, and subsequent operations, identifies the types of policies and query intentions that are of key concern, combines the time rules to judge the policy tracking heat, and finally generates a dynamic feature vector reflecting the policy demand characteristics of enterprises.

[0046] In one embodiment, referring to Figure 4 , the intelligent recommendation engine includes: A content matching unit for calculating the similarity between the user feature vector and the service feature matrix to obtain a content matching score; A collaborative filtering unit for analyzing the similarity of the user group based on the content matching score to obtain a collaborative recommendation result; A rule filtering unit for applying business rules based on the collaborative recommendation result to obtain a filtered result; A multi-strategy fusion unit for comprehensively sorting according to the filtered result to determine the initial recommendation result.

[0047] In this embodiment, the intelligent recommendation engine adopts a multi-level recommendation architecture. It conducts a preliminary screening through the content matching unit, expands the recommendation scope through the collaborative filtering unit, ensures business compliance through the rule filtering unit, and finally generates the final sorting result by the multi-strategy fusion unit. The system has pre-established a basic similarity calculation model, a user group clustering rule library, a business rule configuration table, and a policy weight configuration table, achieving accurate and efficient recommendation calculations. For collaborative filtering with high computational costs, the system adopts a simplified solution based on user grouping, improving the computational efficiency while ensuring the recommendation effect.

[0048] Specifically, the content matching unit adopts an improved cosine similarity algorithm. The system has preset a feature importance weight table, assigning different weight coefficients to different types of features. For example, higher weights (0.6) are assigned to hard condition features such as enterprise qualifications and scale, and lower weights (0.4) are assigned to soft features such as historical behavior. The similarity scores between the user feature vector and each service feature vector are obtained through weighted calculation. The collaborative filtering unit maintains a simplified user grouping mechanism, pre-establishing a user grouping mapping table based on the core features of users (such as enterprise scale, industry affiliation, etc.), and conducting collaborative recommendations based on the historical behavior data of users within the group, avoiding the high cost of calculating the similarity of all users. The rule filtering unit sets up a business rule library, including service timeliness rules (such as project application deadline), user qualification rules (such as access conditions), policy conflict rules (such as mutually exclusive policies), etc., and filters the compliance of the recommendation results. The multi-strategy fusion unit adopts a weighted sorting method. The system has preset a policy weight configuration table, comprehensively considering factors such as content similarity (weight 0.4), group recommendation degree (weight 0.3), and rule adaptation degree (weight 0.3) to generate the final recommendation sorting.

[0049] In one embodiment, referring to Figure 5 , the content matching unit includes: A feature extraction sub-unit, used to extract a static feature set, a dynamic feature set, and a scenario feature set from the user feature vector to obtain a feature representation; A matching calculation sub-unit, used to calculate the dimension similarity, time series correlation, and scenario matching degree between the feature representation and the service feature matrix to obtain a multi-dimensional matching value; A score generation sub-unit, used to perform weighted fusion based on the multi-dimensional matching value and calculate the content matching score.

[0050] In this embodiment, the content matching unit performs feature decomposition through the feature extraction subunit, multi-dimensional similarity calculation through the matching calculation subunit, and weighted fusion through the score generation subunit. The system has pre-established a feature classification mapping table, a dimension calculation rule library, and a weight configuration table to achieve precise matching between user features and service features. The feature classification mapping table defines the standardized representation formats of static features (such as enterprise qualifications), dynamic features (such as service records), and scenario features (such as current requirements); the dimension calculation rule library contains similarity calculation methods for different feature dimensions; and the weight configuration table sets the fusion weights for each dimension according to the service type.

[0051] Specifically, the feature extraction subunit first parses the user feature vector according to the preset feature classification mapping table. The system maintains a feature type recognition rule to map each component in the user feature vector to the corresponding feature set. For example, for enterprise user features, the basic information such as registered capital and industry type is mapped to static features, the recent service records and areas of concern are mapped to dynamic features, and the currently browsed service type and the processing stage are mapped to scenario features. The matching calculation subunit adopts a differentiated calculation strategy: for dimension similarity, it uses an improved cosine similarity algorithm to calculate the matching degree of static features; for temporal correlation, it calculates the correlation degree of dynamic features by setting a time decay factor; for scenario matching degree, it calculates the matching degree of scenario features based on a preset scenario correlation matrix. The score generation subunit fuses the multi-dimensional matching values through a dynamic weight allocation mechanism, and the system has preset weight templates according to different service types.

[0052] In one embodiment, referring to Figure 6 , the real-time feedback optimization module includes: A feedback collection unit for collecting user interaction data based on the initial recommendation result to obtain the original feedback data; A weight calculation unit for calculating the feature weight coefficient according to the original feedback data; A parameter update unit for determining the user-service similarity calculation parameter, the collaborative filtering threshold parameter, and the rule filtering condition parameter based on the feature weight coefficient; A sorting optimization unit for obtaining the optimized final recommendation result according to the similarity calculation parameter, the collaborative filtering threshold parameter, and the rule filtering condition parameter.

[0053] In this embodiment, the real-time feedback optimization module obtains user response data through the feedback collection unit, analyzes the feature importance by the weight calculation unit, adjusts the recommendation strategy parameters by the parameter update unit, and regenerates the recommendation results by the sorting optimization unit. The system has pre-established a feedback index system, a feature weight calculation rule library, and a parameter adjustment template to achieve dynamic optimization of the recommendation effect. The feedback index system includes two types of indicators: explicit feedback (such as clicks, favorites) and implicit feedback (such as browsing duration, bounce rate); the feature weight calculation rule defines the influence mode of different feedback behaviors on feature importance; the parameter adjustment template stipulates the adjustment range and step size of various parameters.

[0054] Specifically, the feedback collection unit records the user's interaction behaviors in real time through the data collection script deployed on the recommendation result display page. The system sets up a behavior-feedback conversion rule to convert behaviors such as the user's clicks (weight 0.3), favorites (weight 0.2), applications (weight 0.5), etc. into standardized feedback scores. The weight calculation unit uses the statistical method of a sliding time window, presets a basic window period of 7 days, and statistically calculates the correlation intensity between each feature and positive feedback within the window period. For example, when it is found that the correlation between the industry feature to which the enterprise belongs and the highly feedback service reaches 0.8, the weight coefficient of this feature is correspondingly increased. The parameter update unit maintains a parameter update rule table and dynamically adjusts the recommendation parameters according to the change of feature weights: for the similarity calculation parameter, adjusts the weight ratio through feature importance; for the collaborative filtering threshold, adjusts the screening criteria for similar users based on group feedback; for the rule filtering condition, dynamically adjusts the filtering intensity according to the rule effectiveness. The sorting optimization unit adopts a real-time rearrangement strategy, applies the updated parameters to the recommendation calculation process, and obtains an optimized recommendation sequence.

[0055] In one embodiment, referring to Figure 7 , the parameter update unit includes: A pre-update and post-update data acquirer, which is used to obtain the recommendation effect data before parameter update within the first preset time period and the recommendation effect data after parameter update within the second preset time period; A recommendation effect change calculator, which is used to obtain the recommendation effect change data according to the recommendation effect data before parameter update and the recommendation effect data after parameter update.

[0056] In this embodiment, the parameter update unit adopts a comparative analysis mechanism, collects the effect data before and after parameter adjustment through the pre-update and post-update data acquirer, and evaluates the effectiveness of parameter update through the recommendation effect change calculator. The system has pre-established an effect index evaluation system and a change trend analysis rule library to achieve quantitative evaluation of the parameter adjustment effect. The effect index evaluation system includes direct indicators (such as click-through rate, conversion rate) and indirect indicators (such as user satisfaction, complaint rate); the change trend analysis rule defines the judgment criteria for effect changes in different scenarios, considering factors such as seasonal fluctuations and activity impacts.

[0057] Specifically, the data acquirer before and after the update adopts a dual-time-window sampling scheme. The system presets 7 days before the parameter update as the first preset time period and 7 days after the update as the second preset time period. For each time period, multi-dimensional effect indicators are obtained through the data acquisition interface: interaction indicators (including display times, click times, average stay duration, etc.), conversion indicators (including service application numbers, completion rates, counseling consultation volumes, etc.), and quality indicators (including user ratings, number of feedback suggestions, etc.). The recommendation effect change calculator sets an index normalization rule to convert indicators with different dimensions into a unified score range of [-1, 1]. At the same time, the system maintains a reference table of benchmark values and sets the fluctuation thresholds of each indicator according to historical data. When the click-through rate of a certain service increases from 5% to 8% and exceeds the preset fluctuation threshold of 2%, the system determines that the parameter adjustment has produced a significant positive effect. In addition, to eliminate the influence of random fluctuations, the system also adopts a simple mean smoothing process, taking the moving average of 3 consecutive days as the final basis for effect evaluation to avoid misjudgment caused by single-day data fluctuations.

[0058] In one embodiment, referring to Figure 8 , it further includes a recommendation quality monitoring module, and the recommendation quality monitoring module includes: A recommendation stability coefficient acquisition unit, configured to obtain the recommendation result fluctuation data within a third preset time period in real time, and obtain the recommendation stability coefficient according to the fluctuation data; A user feedback abnormal number acquisition unit, configured to obtain the number of times of user feedback abnormal triggers in real time according to the recommendation effect change data; A system warning information sending unit, configured to send a system warning information to the management terminal when the number of times of user feedback abnormal triggers is greater than a preset abnormal threshold or the recommendation stability coefficient is greater than a preset fluctuation threshold.

[0059] In this embodiment, the recommendation quality monitoring module adopts a real-time monitoring mechanism. The recommendation stability coefficient acquisition unit monitors the fluctuation of the recommendation results, the user feedback abnormal number acquisition unit counts the abnormal feedback, and the system warning information sending unit triggers the warning in time. The system has pre-established a fluctuation statistics rule library, an abnormal pattern recognition rule library, and a warning trigger configuration table to realize the automatic monitoring of the recommendation quality. The fluctuation statistics rule defines the calculation method of the recommendation result stability; the abnormal pattern recognition rule contains the determination criteria for various abnormal feedbacks; the warning trigger configuration table stipulates the trigger conditions and processing procedures for different levels of warnings.

[0060] In one embodiment, the present application provides an electronic device, which may be a server, and its internal structure diagram may be as Figure 9As shown in the figure. The electronic device includes a processor, a memory, and a network interface connected through a system bus. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the electronic device is used to store data. The network interface of the electronic device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements the steps performed by a digital public service integration platform based on intelligent recommendation.

[0061] Those skilled in the art can understand that Figure 9 the structure shown in the figure is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the electronic device to which the solution of this application is applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0062] In one embodiment, an electronic device is further provided, including a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, the steps in the above method embodiments are implemented.

[0063] Those of ordinary skill in the art can understand that all or part of the processes of implementing the above method embodiments can be completed by instructing relevant hardware through a computer program. The above computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above method embodiments. Among them, any reference to the memory, storage, database, or other media used in the various embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical memory, etc. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0064] The above are all preferred embodiments of this application. The protection scope of this application is not limited thereby. Therefore, any equivalent changes made according to the structure, shape, and principle of this application should be covered within the protection scope of this application.

Claims

1. A digital public service integration platform based on intelligent recommendation, characterized in that Including: A user profile construction module for collecting user basic information and behavioral data to obtain user feature vectors; A service feature extraction module for performing structured processing on government service data to obtain a service feature matrix; An intelligent recommendation engine for performing matching calculations based on the user feature vectors and the service feature matrix to obtain an initial recommendation result; A real-time feedback optimization module for collecting user feedback based on the initial recommendation result to obtain an optimized final recommendation result.

2. The digital public service integration platform based on intelligent recommendation according to claim 1, characterized in that The user profile construction module includes: A static feature extraction unit for obtaining user basic attribute information to obtain a static feature set; A dynamic feature extraction unit for collecting user historical behavioral data to obtain a dynamic feature set; A scenario feature extraction unit for obtaining a scenario feature set according to the dynamic feature set and the user's current state; A feature fusion unit for calculating a fused feature vector according to the static feature set, the dynamic feature set and the scenario feature set.

3. The digital public service integration platform based on intelligent recommendation according to claim 2, characterized in that, The dynamic feature extraction unit includes: A browsing record analyzer for extracting user access data to obtain browsing features; An interaction behavior analyzer for analyzing the operation mode according to the browsing features to obtain behavior features; A time series feature extractor for calculating time dimension features based on the behavior features; A feature fuser for obtaining comprehensive dynamic features according to the browsing features, the behavior features and the time dimension features.

4. The digital public service integration platform based on intelligent recommendation according to claim 2, wherein, The intelligent recommendation engine includes: A content matching unit for calculating the similarity between the user feature vectors and the service feature matrix to obtain a content matching score; A collaborative filtering unit for analyzing user group similarity according to the content matching score to obtain a collaborative recommendation result; A rule filtering unit for applying business rules based on the collaborative recommendation result to obtain a filtered result; A multi-strategy fusion unit for performing comprehensive sorting according to the filtered result to determine the initial recommendation result.

5. The digital public service integration platform based on intelligent recommendation according to claim 4, characterized in that, The content matching unit includes: A feature extraction subunit for extracting a static feature set, a dynamic feature set and a scenario feature set from the user feature vectors to obtain a feature representation; A matching calculation subunit for calculating the dimensional similarity, the time series correlation and the scenario matching degree between the feature representation and the service feature matrix to obtain a multi-dimensional matching value; A score generation subunit for performing weighted fusion according to the multi-dimensional matching value to calculate the content matching score.

6. The digital public service integration platform based on intelligent recommendation according to claim 4, characterized in that The real-time feedback optimization module includes: A feedback collection unit for collecting user interaction data based on the initial recommendation result to obtain raw feedback data; A weight calculation unit for calculating feature weight coefficients according to the raw feedback data; A parameter update unit for determining user-service similarity calculation parameters, collaborative filtering threshold parameters and rule filtering condition parameters based on the feature weight coefficients; A sorting optimization unit for obtaining an optimized final recommendation result according to the similarity calculation parameters, the collaborative filtering threshold parameters and the rule filtering condition parameters.

7. The digital public service integration platform based on intelligent recommendation according to claim 6, characterized in that, The parameter update unit includes: A data acquirer before and after update, which is used to acquire the recommended effect data before parameter update within a first preset time period and the recommended effect data after parameter update within a second preset time period; A recommended effect change calculator, which is used to acquire recommended effect change data according to the recommended effect data before parameter update and the recommended effect data after parameter update.

8. The digital public service integration platform based on intelligent recommendation according to claim 1, wherein It further includes a recommended quality monitoring module, and the recommended quality monitoring module includes: A recommended stability coefficient acquisition unit, which is used to acquire the fluctuation data of recommended results within a third preset time period in real time, and acquire the recommended stability coefficient according to the fluctuation data; A user feedback abnormal times acquisition unit, which is used to acquire the triggered times of user feedback abnormal in real time according to the recommended effect change data; A system warning information sending unit, which is used to send system warning information to a management terminal when the triggered times of user feedback abnormal are greater than a preset abnormal threshold or the recommended stability coefficient is greater than a preset fluctuation threshold.

9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the steps executed by the digital public service integration platform based on intelligent recommendation according to any one of claims 1-8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps executed by the digital public service integration platform based on intelligent recommendation according to any one of claims 1-8.

Images