A personalized recommendation engine system for government service one-network handling

Abstract

The present application relates to the technical field of personalized recommendation, in particular to a personalized recommendation engine system for government service one-network handling, comprising an identity feature extraction module, a feature array decomposition module, a qualification feature completion module, a rule boundary comparison module and a terminal rendering driving module.In the present application, an initial qualification sequence is generated by collecting and identifying signals and splicing credit data, singular value decomposition is performed on a two-dimensional array, missing qualifications are calculated by combining the core diagonal line and the specific row and column, the original sequence is merged to complete the subject's complete government qualification, the access logic is established by performing size comparison on the complete qualification, the target sequence is packaged and converted into terminal display instructions, the static comparison misalignment defect caused by limited input is overcome, the solid equivalent verification drawbacks are abandoned, the subject's implied handling conditions are excavated, the distribution of invalid business links is effectively blocked, and the personalized and accurate recommendation effect of the government terminal is ensured.

Description

Technical Field

[0001] This invention relates to the field of personalized recommendation technology, and in particular to a personalized recommendation engine system for one-stop online government services. Background Technology

[0002] Personalized recommendation technology involves methods for providing customized information services to users based on their historical behavior, attributes, and context. This involves collecting user browsing, click, and form submission records, and constructing feature vectors based on attributes such as age, gender, and place of residence. Simultaneously, it involves structured parsing of candidate item descriptions and attribute tags to filter content lists for specific users. The personalized recommendation engine system for government services refers to an architecture for filtering and distributing administrative approvals and public services aggregated from various levels of government departments. This is typically accomplished using a pre-set business rule dictionary. Typically, the system involves manually sifting through the service guides of various commissions and bureaus, extracting the age range, place of residence, enterprise registration type, and industry qualification certificate prerequisites required for processing the service. These conditions are compiled into a static mapping database table containing multi-level key-value pairs. When the system receives the basic identity field of an ID card number or unified social credit code input from the terminal, it performs a row-by-row equal-value comparison in the aforementioned static mapping database table through conditional branching statements. If the input identity field is completely equal to the prerequisites defined for a certain government service item in the database table, the processing link for that government service item is extracted and output to the terminal display page for layout and rendering.

[0003] In the current process of personalized government service recommendations, the main approach relies on building a static mapping database table by sorting out rules. The received identity field is only compared row by row. This operating mode has certain limitations in real-world scenarios. Due to the limited information input by the terminal, the fixed branch comparison mechanism is difficult to uncover the implicit qualification attributes of the applicant. When the basic parameters and preset conditions are not completely equal, the recommendation engine cannot accurately know the complete application qualifications. This leads to the system frequently pushing invalid government service links that do not meet the actual access conditions, which seriously restricts the improvement of the accuracy of self-service terminals and increases the time spent on invalid trial and error operations by the applicant. Summary of the Invention

[0004] To address the technical problems existing in the prior art, this invention provides a personalized recommendation engine system for one-stop online government services. The technical solution is as follows:

[0005] On the one hand, a personalized recommendation engine system for one-stop online government services is provided, including:

[0006] The identity feature extraction module collects government service identity recognition signals and converts them into a digital number sequence. It extracts the corresponding historical number of services and credit score and performs character concatenation operation with the digital number sequence to obtain the initial sparse qualification sequence.

[0007] The feature array decomposition module combines the initial sparse quality sequence into a two-dimensional feature array by row, and performs singular value decomposition on the two-dimensional feature array to split it into a left singular array, a core array and a right singular array, thus constructing a singular decomposition feature array.

[0008] The qualification feature completion module extracts the target service entity line sequence, constructs a predicted missing qualification sequence by combining it with the singular decomposition feature array, and merges and splices the predicted missing qualification sequence with the initial sparse qualification sequence to generate a complete government qualification sequence.

[0009] The rule boundary comparison module extracts the lower limit boundary of the admission age and the upper limit boundary of the registered capital, compares the qualification parameters corresponding to the complete government qualification sequence with the upper and lower limit boundaries, and establishes an admission permit logical identifier.

[0010] The terminal rendering driver module performs an extraction operation on the access permission logical identifier, associated with the government name sequence and the jump address sequence, encapsulates the two sequences and converts them into terminal coordinate display instructions, and establishes personalized recommendation results for one-stop government services.

[0011] As a further aspect of the present invention, in the process of constructing the predicted missing qualification sequence, the diagonal sequence of the core array is extracted by combining the singular decomposition feature array, the target service subject row sequence is aligned and merged with the diagonal sequence of the core array to construct an intermediate transition feature sequence, the corresponding column sequence of the right singular array is extracted simultaneously, and the intermediate transition feature sequence is mapped and fused with the corresponding column sequence.

[0012] As a further aspect of the present invention, during the size comparison process, if the value is greater than the lower limit and less than the upper limit, a high-level indicator is output; if the value is less than the lower limit or greater than the upper limit, a low-level indicator is output.

[0013] As a further aspect of the present invention, the initial sparse qualification sequence includes an identity authentication code string, performance behavior rating, and government interaction frequency; the singular decomposition feature array includes a user implicit preference vector, an attribute dimensionality reduction weight matrix, and a transaction potential association table; the complete government qualification sequence includes full-dimensional verification indicators, estimated asset level, and potential application categories; the access permit logical identifier includes a compliance pass, a conditional rejection code, and a manual review requirement marker; and the personalized recommendation results of the one-stop government service platform include customized service guides, dedicated business direct links, and intelligent form filling guidance pages.

[0014] As a further aspect of the present invention, the identity feature extraction module includes:

[0015] The signal parsing submodule collects government service identity recognition signals and converts them into digital number sequences. It receives simulated business test data, extracts basic comparison identifiers from the simulated business test data, performs a bit-by-bit comparison operation between the basic comparison identifiers and the digital number sequences, determines heterogeneous deviation intervals, removes sequence segments that fall within the heterogeneous deviation intervals, sorts the removed digital number sequences according to their positional order, and generates a feature relationship mapping quantity.

[0016] The qualification concatenation submodule extracts the corresponding historical service counts and credit scores, performs format conversion processing on the historical service counts and credit scores to obtain a fixed-length bit string, compares the fixed-length bit string with the first and last nodes of the feature relationship mapping to determine overlapping nodes, removes redundant characters in the overlapping nodes, and performs character concatenation operation on the removed fixed-length bit string and the feature relationship mapping to obtain the initial sparse qualification sequence.

[0017] As a further aspect of the present invention, the feature array decomposition module includes:

[0018] The feature combination submodule extracts the internal separators of the initial sparse quality sequence, performs segmentation based on the separators to obtain multiple groups of segments, and combines the multiple groups of segments row by row according to a fixed step size to generate a two-dimensional feature array.

[0019] The singular decomposition submodule performs singular value decomposition on the two-dimensional feature array to split it into a left singular array, a core array, and a right singular array. It compares the elements in the core array with the retention benchmark threshold to determine the difference range, removes the values ​​falling outside the difference range, and obtains the dimensionality-reduced core feature quantity.

[0020] The feature construction submodule, based on the dimensionality reduction core feature quantity, performs transpose on the left singular array and the right singular array to obtain a transposed singular array, rearranges and maps the dimensionality reduction core feature quantity with the transposed singular array to obtain a corrected array, and performs dot product fusion on the corrected array and the left singular array to generate a singular decomposition feature array.

[0021] As a further aspect of the present invention, the qualification feature completion module includes:

[0022] The sequence merging submodule collects the target service entity line sequence, extracts the core array diagonal sequence by combining the singular decomposition feature array, performs length padding operation on the core array diagonal sequence and the target service entity line sequence to obtain equal-length numerical segments, adds the elements in the equal-length numerical segments to generate the merging transition feature quantity;

[0023] The mapping and fusion submodule extracts the corresponding column sequence of the right singular array based on the merged transition feature quantity, performs a dot product mapping between the corresponding column sequence and the merged transition feature quantity to obtain the fusion prediction value, and arranges the fusion prediction value in position order to generate the prediction supplementary feature quantity.

[0024] The qualification splicing submodule compares the predicted supplementary feature quantity with the initial sparse qualification sequence to determine the placeholder node, and embeds the predicted supplementary feature quantity into the placeholder node in the initial sparse qualification sequence to generate a complete government qualification sequence.

[0025] As a further aspect of the present invention, the rule boundary comparison module includes:

[0026] The boundary extraction submodule acquires a sample dataset, extracts the lower limit boundary of the admission age and the upper limit boundary of the registered capital of the sample dataset, combines the lower limit boundary of the admission age and the upper limit boundary of the registered capital to obtain the dual-end judgment benchmark parameters, extracts the corresponding qualification parameters in the complete government qualification sequence, performs a sorting and alignment operation on the corresponding qualification parameters and the dual-end judgment benchmark parameters to construct a comparison tuple array, and generates a boundary comparison reference set.

[0027] The logic level submodule, based on the boundary comparison reference set, performs a size comparison logic operation between the internally arranged elements and the dual-end judgment benchmark parameter. If it is greater than the lower limit and less than the upper limit, it outputs a high-level flag to the data transmission port. If it is less than the lower limit or greater than the upper limit, it outputs a low-level flag to the data transmission port. It performs timing conversion on the collected high-level flags and low-level flags to establish an access permission logic flag.

[0028] As a further aspect of the present invention, the terminal rendering driver module includes:

[0029] The association extraction submodule extracts the association comparison parameters within the preset sample data, compares the access permission logical identifier with the association comparison parameters to determine the logical node, extracts the government name sequence and the jump address sequence, and generates an association mapping matrix.

[0030] The sequence encapsulation submodule, based on the association mapping matrix, performs alignment and merging on the government name sequence and the jump address sequence to construct the network payload, performs protocol packaging operation on the network payload, collects device screen parameters, combines the positioning value with the device screen parameters to determine the rendering landing point, and generates terminal coordinate display instructions;

[0031] The rendering driver submodule extracts internal coordinates and service carrying data based on the terminal coordinate display instructions. It performs raster mapping processing on the internal coordinates and the service carrying data to construct a bottom-level pixel matrix. It then performs timing allocation on the bottom-level pixel matrix with the refresh frequency to establish personalized recommendation results for government services through the online platform.

[0032] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following:

[0033] By collecting and identifying signals and splicing credit data to generate an initial qualification sequence, singular value decomposition is performed on the two-dimensional array, and missing qualifications are inferred by combining the core diagonal with specific rows and columns. The original sequence is merged to complete the subject's complete government qualifications. The access boundary is extracted and a size comparison is performed on the complete qualifications to establish access logic identifiers. The target sequence is encapsulated and converted into terminal display instructions. This overcomes the defects of static comparison inaccuracy caused by input limitations, abandons the drawbacks of fixed equivalent verification, and explores the subject's implicit processing conditions. It effectively blocks the distribution of invalid business links and ensures the personalized and accurate recommendation effect of government terminals. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a schematic diagram of a personalized recommendation engine system for one-stop online government services provided in an embodiment of the present invention;

[0036] Figure 2 This is a schematic diagram of the system framework of the present invention;

[0037] Figure 3 This is a flowchart of the identity feature extraction module of the present invention;

[0038] Figure 4 This is a flowchart of the feature array decomposition module of the present invention;

[0039] Figure 5 Flowchart of the qualification feature completion module for this invention;

[0040] Figure 6 This is a flowchart of the rule boundary comparison module of the present invention;

[0041] Figure 7 This is a flowchart of the terminal rendering driver module of the present invention. Detailed Implementation

[0042] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0043] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0044] This invention provides a personalized recommendation engine system for one-stop online government services, such as... Figure 1-2 The schematic diagram shown includes:

[0045] The identity feature extraction module collects government service identity recognition signals and converts them into a digital number sequence. It extracts the corresponding historical number of services and credit score and performs character concatenation operation with the digital number sequence to obtain the initial sparse qualification sequence.

[0046] The feature array decomposition module combines the initial sparse quality sequence into a two-dimensional feature array by row, and performs singular value decomposition on the two-dimensional feature array to split it into a left singular array, a core array and a right singular array, thus constructing a singular decomposed feature array.

[0047] The qualification feature completion module extracts the target service entity row sequence, extracts the core array diagonal sequence by combining the singular decomposition feature array, performs alignment and merging operation on the target service entity row sequence and the core array diagonal sequence to construct the intermediate transition feature sequence, simultaneously extracts the corresponding column sequence of the right singular array, performs mapping and fusion on the intermediate transition feature sequence and the corresponding column sequence to construct the predicted missing qualification sequence, and merges and splices the predicted missing qualification sequence with the initial sparse qualification sequence to generate the complete government qualification sequence;

[0048] The rule boundary comparison module extracts the lower limit boundary of the admission age and the upper limit boundary of the registered capital. It compares the qualification parameters corresponding to the complete government qualification sequence with the upper and lower limit boundaries. If the value is greater than the lower limit and less than the upper limit, it outputs a high-level indicator. If the value is less than the lower limit or greater than the upper limit, it outputs a low-level indicator, thus establishing the access permit logic identifier.

[0049] The terminal rendering driver module performs extraction operations on the access permission logical identifier, associated government name sequence and jump address sequence, encapsulates the two sequences and converts them into terminal coordinate display instructions, and establishes personalized recommendation results for one-stop government services.

[0050] The initial sparse qualification sequence includes identity authentication code string, performance behavior rating, and frequency of government interaction; the singular decomposition feature array includes user implicit preference vector, attribute dimensionality reduction weight matrix, and transaction potential association table; the complete government qualification sequence includes full-dimensional verification indicators, estimated asset level, and potential application categories; the access permit logical identifier includes compliance pass certificate, conditional rejection code, and manual review required marker; and the personalized recommendation results of the one-stop government service platform include customized service guides, dedicated business direct links, and intelligent form filling guidance pages.

[0051] Specifically, such as Figure 2 , 3 As shown, the identity feature extraction module includes:

[0052] The signal parsing submodule collects government affairs processing identity recognition signals and converts them into digital number sequences. It receives simulated business test data, extracts basic comparison identifiers from the simulated business test data, performs bit-by-bit comparison operations between the basic comparison identifiers and the digital number sequences, determines heterogeneous deviation intervals, removes sequence segments that fall within the heterogeneous deviation intervals, sorts the removed digital number sequences according to their position order, and generates feature relationship mapping quantities.

[0053] The signal parsing submodule collects government service identity recognition signals through photoelectric sensing components, directly quantizing analog voltage fluctuations and converting them into a numerical sequence composed of Arabic numerals. This step is a prerequisite for achieving unified identity authentication for one-stop online government services and supporting accurate distribution in subsequent systems. This module receives simulated business test data from a pre-built relational database, reads field content, and extracts the basic comparison identifier from the simulated business test data. The signal parsing submodule performs a bit-by-bit alignment operation between the basic comparison identifier and the numerical sequence and calculates the absolute difference between corresponding values. When the basic comparison identifier... The number of bits is And the number sequence number is the The number of bits is At that time, the absolute deviation value is obtained by calculating the difference. This module compares the absolute deviation value with the heterogeneity determination benchmark threshold. When the absolute deviation value is greater than the set threshold, the module will detect the heterogeneity. When the heterogeneity determination benchmark threshold is reached, the position is determined to be within the heterogeneity deviation range. For sequence segments falling within the heterogeneity deviation range, the signal analysis submodule performs a physical memory address overwrite and clearing operation to remove them. The removed numerical sequence is then sorted according to its original physical address position in the memory module. Scattered values ​​from adjacent memory blocks are integrated to reconstruct a continuous data stream, generating a feature relationship mapping. Through standard data cleaning and noise filtering steps in the machine learning process, the interference of underlying hardware jitter on the feature extraction of the upper-layer model is effectively eliminated, laying a solid foundation for underlying relationship calculation for the subsequent personalized recommendation business of the government service one-stop online platform.

[0054] The qualification concatenation submodule extracts the corresponding historical service counts and credit scores, performs format conversion on the historical service counts and credit scores to obtain a fixed-length bit string, compares the fixed-length bit string with the first and last nodes of the feature relationship mapping to determine the overlapping area nodes, removes redundant characters in the overlapping area nodes, and performs character concatenation operation on the removed fixed-length bit string and the feature relationship mapping to obtain the initial sparse qualification sequence.

[0055] The qualification splicing submodule extracts the historical service frequency and credit score of the corresponding target object from the government service record storage component. This multi-dimensional historical behavioral data is a key business indicator for constructing personalized user profiles for the one-stop government service platform, and can deeply reveal the frequency of public service requests and their creditworthiness. When the historical service frequency value is... And the credit score is At this time, this module performs binary encoding conversion on the historical number of transactions and credit score, converting the decimal values ​​into a fixed-length bit string with a preset placeholder length. The qualification concatenation submodule extracts the beginning and end of this fixed-length bit string. The end of the bit character and feature relationship mapping quantity The module compares the first and last nodes of a fixed-length bit string with the feature relation mapping to determine if there are identical character arrangements. Identical consecutive bits are considered overlapping regions. For redundant characters recurring within these overlapping regions, the module performs a one-sided retention / removal operation to eliminate them, freeing up register space. The qualification concatenation submodule then performs a concatenation operation between the eliminated fixed-length bit string and the feature relation mapping, writing each character data sequentially into consecutive flash memory sectors. When the remaining length of the fixed-length bit string is... And the length of the feature relation mapping is When the two are joined together, the total length is The initial sparse qualification sequence of a certain number of characters is integrated through this serialization, so that the user feature elements initially have the ability to splice and represent discrete and continuous variables required in machine learning feature engineering, and thus smoothly enter the data form of personalized recommendation engine for preference matching.

[0056] Specifically, such as Figure 2 , 4 As shown, the feature array decomposition module includes:

[0057] The feature combination submodule extracts the internal separators of the initial sparse quality sequence, performs segmentation based on the separators to obtain multiple groups of segments, and combines the multiple groups of segments row by row according to a fixed step size to generate a two-dimensional feature array.

[0058] The feature combination submodule reads the internal data of the initial sparse qualification sequence and extracts the consecutively occurring data within the sequence. Each consecutive zero value serves as a separator. Based on these consecutive zero-value separators, the module performs segmentation processing on the initial sparse quality sequence, forcibly blocking the data flow at the hardware index position of each separator to remove redundant separator characters, thus obtaining the total. Groups of independent data segments. The feature combination submodule will... The group of segments is set as follows Extract the internal values ​​with a fixed step size of character width, and arrange them row by row in the cache matrix area according to the left-to-right arrangement. Group fragments are placed in the matrix. Line start position, exceeding Each character step triggers a carry mechanism and is placed at the beginning of the next line. This module continues to read the remaining segments until all data is arranged within a two-dimensional planar grid, generating a two-dimensional feature array with row and column coordinates. When the total number of characters in all segments is... At that time, according to Perform matrix padding on characters of width to precisely construct a string with... OK A two-dimensional feature array with column space structure is used to construct a multi-dimensional input tensor format that meets the requirements of machine learning convolutional neural networks or deep feedforward networks. This thoroughly structures the messy government affairs processing logs and provides a standard input container for further extracting high-dimensional features to optimize personalized recommendation strategies in the one-stop government services platform.

[0059] Table 1. Data Table of Two-Dimensional Feature Array

[0060]

[0061] As shown in Table 1, the feature combination submodule completes the construction and output of the two-dimensional feature array.

[0062] The singular decomposition submodule performs singular value decomposition on the two-dimensional feature array to split it into a left singular array, a core array, and a right singular array. It compares the elements in the core array with the retention benchmark threshold to determine the difference range, removes the values ​​falling outside the difference range, and obtains the dimensionality-reduced core feature quantity.

[0063] The singularity decomposition submodule receives a two-dimensional feature array from the storage medium. Based on the element arrangement rules within the two-dimensional feature array, it performs matrix orthogonalization calculations, decomposing its singular values ​​into a left singular array, a core diagonal array containing eigenvalue information, and a corresponding right singular array. This module extracts the numerical elements at the main diagonal positions of the core diagonal array and compares the extracted core array elements with a system-set retention threshold to determine their magnitude. When the core array's main diagonal elements sequentially present... , , , And the baseline threshold is strictly set to At that time, the singular splitting submodule executes the size comparison instructions item by item through the central processing unit to obtain the value. With numerical values Clearly less than the retention benchmark threshold The coordinates of the lower-value items in this section are defined as the difference interval. The singular splitting submodule removes values ​​falling outside the difference interval, triggering a memory address zeroing mechanism to erase the values. With numerical values All associated numerical elements in the row and column. After thoroughly removing low-value associated items, the high-weight numerical elements retained in the singular splitting submodule integration matrix are reassembled to generate a higher-density dimensionality-reduced core feature quantity. This dimensionality reduction effectively filters out invalid feature interference generated by the daily redundant interactions of the government service one-stop online service system, significantly improving the computational efficiency and accuracy of the personalized recommendation core logic under limited computing resources.

[0064] The feature construction submodule, based on the dimensionality reduction core feature quantity, performs transpose on the left singular array and the right singular array to obtain the transposed singular array, rearranges and maps the dimensionality reduction core feature quantity with the transposed singular array to obtain the corrected array, and performs dot product fusion on the corrected array and the left singular array to generate the singular decomposition feature array.

[0065] The feature construction submodule, based on the extracted dimensionality-reduced core features, performs a transpose operation on the left and right singular arrays in the temporary storage space, swapping the coordinates of their row and column elements to obtain the reconstructed transposed singular array. This module then performs a rearrangement mapping process on the dimensionality-reduced core features and the transposed singular array based on the product of their corresponding coordinate indices. When the coordinate elements within the dimensionality-reduced core features are... And the elements at the exact corresponding positions of the transposed singular array are At that time, the arithmetic logic unit is invoked to multiply the two to obtain the correction element. The calculated values ​​are then precisely filled into the corresponding memory coordinates to construct a corrected array. The feature construction submodule performs a dot product fusion operation on the corrected array, which is filled with product results, and the untransposed original left singular array, summing the corresponding elements of the row and column vectors. When the product values ​​of the first row vector element of the corrected array and the first column vector element of the left singular array are respectively... and At that time, the accumulator is called to sum the results and obtain the total value. As the first coordinate element of the new fusion matrix, this module sequentially calculates the multiplicative sum of all row and column vector combinations, outputs a fully filled singular decomposition feature array, completes the feature cross-operation of machine learning implicit representation vectors, deeply explores the potential nonlinear correlation between various government historical qualifications, and provides high-dimensional and comprehensive fusion feature matrix support for subsequent personalized recommendations across departments and business scenarios.

[0066] Specifically, such as Figure 2 , 5 As shown, the qualification feature completion module includes:

[0067] The sequence merging submodule collects the target service entity line sequence, extracts the core array diagonal sequence by combining the singular decomposition feature array, performs length padding operation on the core array diagonal sequence and the target service entity line sequence to obtain equal-length numerical segments, adds the elements within the equal-length numerical segments, and generates the merging transition feature quantity.

[0068] The sequence merging submodule collects the target service subject line sequence, which records the attribute information of service personnel, through the system data reading channel. It then extracts the core array diagonal sequence at the main diagonal position by combining this sequence with the singular decomposition feature array in memory. When the collected target service subject line sequence contains a total of... The core array diagonal sequence contains only a limited number of data elements, and is limited by dimensionality. When dealing with a single data element, this module performs a length padding operation on the core array diagonal sequence and the target service entity row sequence, continuously increasing the physical address at the end of the core array diagonal sequence. All values ​​are set to The compensation elements are constructed to all contain The sequence merging submodule performs basic arithmetic addition on the internal elements with the same position index within the two sets of equal-length numerical segments. When the first element of each set of equal-length numerical segments is extracted as... and At that time, the arithmetic unit is invoked to calculate the first and second combined values ​​by adding the numerical values. This module processes all instances using the same logic. Combine all element pairs and sum the results to generate a total containing The merging and transitional feature quantity of each independent data item, the completion and addition process satisfies the input dimension matching rules that are mandatory during the forward propagation calculation of the machine learning model, and strictly ensures the data alignment and dimension consistency of various heterogeneous service entities in the government service one-stop online service system before entering the final personalized recommendation prediction engine.

[0069] The mapping fusion submodule extracts the corresponding column sequence of the right singular array based on the merged transition feature quantity, performs a dot product mapping between the corresponding column sequence and the merged transition feature quantity to obtain the fused prediction value, and arranges the fused prediction value according to the position order to generate the prediction supplementary feature quantity.

[0070] The mapping fusion submodule, based on the merged transition feature quantity generated by the preceding computation, extracts the corresponding column sequence that precisely corresponds to the dimension of the target attribute from the right singular array stored in the cache. This module will contain a total of The corresponding column sequence of each element also contains The merged transition feature of each element undergoes a dot product mapping process based on the product of the values ​​at the corresponding array index positions. When the value extracted from the first position of the corresponding column sequence is... And the value extracted from the first position of the merged transition feature is At that time, the multiplier component is invoked to directly multiply the two values ​​to obtain the first fused prediction value. For all After performing the product operation on each element, we get the result from... The data set consists of individual predicted values. The mapping and fusion submodule arranges and connects these fused predicted values ​​in ascending order of their original element indices. After rigorous sequence assembly, the predicted values ​​form a continuous, uninterrupted numerical stream in a contiguous memory space. This module packages and outputs the entire stream as a continuous bit string, generating supplementary predicted features. This allows the personalized recommendation model to maintain the stability, consistency, and timeliness of government service recommendations even when there is a temporary lack of up-to-date user qualification data, based on high-quality predicted values.

[0071] The qualification splicing submodule compares the predicted supplementary features with the initial sparse qualification sequence to determine the missing nodes, and embeds the predicted supplementary features into the missing nodes in the initial sparse qualification sequence to generate a complete government qualification sequence.

[0072] The qualification splicing submodule starts a bit-by-bit scanner to traverse all memory storage nodes of the initial sparse qualification sequence, checking for characters that represent missing data and are set to zero values. It compares the total number of elements within the predicted supplementary feature quantity with the total number of nodes in the initial sparse qualification sequence to determine the precise index position of the zero-value element, which is the placeholder node that needs to be compensated. When the initial sparse qualification sequence reaches the [missing information], ... Bit register address and the When a bit register address contains a zero value, this module explicitly designates these two address segments as placeholder vacancy nodes. The qualification splicing submodule extracts the corresponding first-order supplementary feature quantity within the predicted supplementary feature quantity based on the position mapping relationship. Position and First The predicted supplementary value of the bit sequence is called, and the data writing instruction is called to overwrite the specific value extracted from the predicted supplementary feature quantity into the register where the placeholder node in the initial sparse qualification sequence is located, forcibly erasing the original zero value placeholder, and completing the physical level replacement at the data content level.

[0073] Table 2 Complete Government Qualification Sequence List

[0074]

[0075] Based on the data entry process in Table 2, the qualification splicing submodule integrates the discontinuous data segments into independent data strings that are connected end to end, generating a complete government qualification sequence. This completes the panoramic dynamic qualification splicing reconstruction for specific applicants, providing a seamless underlying core basis for the personalized recommendation engine to make accurate matching decisions.

[0076] Specifically, such as Figure 2 , 6 As shown, the rule boundary comparison module includes:

[0077] The boundary extraction submodule acquires the sample dataset, extracts the lower limit boundary of the admission age and the upper limit boundary of the registered capital in the sample dataset, combines the lower limit boundary of the admission age and the upper limit boundary of the registered capital to obtain the dual-end judgment benchmark parameters, extracts the corresponding qualification parameters in the complete government qualification sequence, performs a sorting and alignment operation on the corresponding qualification parameters and the dual-end judgment benchmark parameters to construct a comparison tuple array, and generates a boundary comparison reference set.

[0078] The boundary extraction submodule acquires a manually annotated historical approval sample dataset via an external high-speed serial bus interface. It then reads fields from the underlying database tables to extract the minimum age limit for legal representatives and the maximum registered capital limit for companies. When the extracted minimum age limit boundary value is... And the result of extracting the upper limit value of registered capital is At this time, the module combines and stores the two extreme values ​​in the same one-dimensional memory array structure to obtain the dual-end judgment benchmark parameter. The boundary extraction submodule extracts the corresponding qualification parameters of the age and capital dimensions in the complete government qualification sequence according to the system's preset fixed field names, and performs a sorting and alignment operation on the corresponding qualification parameters and the dual-end judgment benchmark parameter with the same field names. This module combines and encapsulates the extracted applicant age value with the lower limit boundary of the admission age, and combines and encapsulates the applicant capital value with the upper limit boundary of the registered capital, constructing a comparison tuple array containing multiple sets of data key-value pair structures, integrating and processing all paired groups, and outputting a boundary comparison reference set. This key rule defense line fundamentally ensures that all personalized recommendations generated by the government service one-stop online platform are strictly limited to the user's legally permissible permissions and legal age and capital requirements, completely eliminating invalid and unauthorized service recommendations.

[0079] The logic level submodule, based on the boundary comparison reference set, performs a size comparison logic operation on the internal arrangement elements and the double-ended judgment benchmark parameter. If it is greater than the lower limit and less than the upper limit, it outputs a high level flag to the data transmission port. If it is less than the lower limit or greater than the upper limit, it outputs a low level flag to the data transmission port. It performs timing conversion on the collected high level flags and low level flags to establish an access permission logic flag.

[0080] The logic level submodule, based on the boundary comparison reference set within the buffer area, reads the qualification parameter values ​​corresponding to the addresses of the internally arranged elements. For each internally arranged element and its corresponding double-ended judgment benchmark parameter, it calls a logic comparison gate to perform numerical subtraction calculations and sign recognition comparison logic operations. When the applicant's age parameter value is extracted as... Furthermore, the lower limit of the admission age within the dual-end judgment benchmark parameter is extracted as follows: When the comparator outputs a positive number, indicating it is greater than the lower limit, and the applicant's funding parameter value is... Furthermore, the upper limit of registered capital within the dual-end judgment benchmark parameters is When the comparator outputs a negative number, it indicates the value is less than the upper limit. If all qualification parameters meet the condition of being greater than the lower limit and less than the upper limit, the module triggers the driver to continuously output a stable voltage amplitude to the data transmission port of the hardware device. A high-level indicator for volts. If the applicant's age parameter value is... Less than the lower limit Or the financial parameter value is Greater than the upper limit The logic level submodule cuts off the voltage drive and outputs a voltage amplitude of [value missing] to the data transmission port. The module performs a timing conversion based on the system clock cycle, arranging the collected high and low level indicators in chronological order, to establish a sequence of high and low level indicators. and The access permission logic identifier, composed of binary level states, constitutes the underlying electronic gate for the compliance diversion of government service online services. It directly determines the display priority of each recommended service item in the final personalized recommendation list and its clickable interactive state.

[0081] Specifically, such as Figure 2 , 7 As shown, the terminal rendering driver module includes:

[0082] The association extraction submodule extracts association comparison parameters from the preset sample data, compares the access permission logical identifier with the association comparison parameters to determine the logical node, extracts the government name sequence and jump address sequence, and generates an association mapping matrix.

[0083] The association extraction submodule extracts association comparison parameters used to determine the mapping relationship between approval node connectivity and business type within the pre-set sample data on the system's local hard drive through a text character matcher invoked via memory addressing. This module reads the binary level state sequence arranged within the access permission logic identifier and compares the access permission logic identifier with the association comparison parameter to determine the specific business flow logic node corresponding to the level state transition action. When continuous captures occur in the access permission logic identifier data packet... The number of high-level states and the requirement for the number of high-level states within the correlation comparison parameter must be greater than [a certain number]. At that time, the judgment component determines that the current business processing status has indeed reached the approval node. Based on the triggered approval logic node, this module triggers a database query command to extract the business registration service name contained within the government name sequence and the Hypertext Transfer Protocol address bound within the jump address sequence from the business routing relationship table fields. The association extraction submodule establishes a one-way key-value pair association between the extracted government name sequence and the corresponding jump address sequence in the memory buffer, and stores all mapping key-value pairs in a two-dimensional addressing matrix structure, outputting an association mapping matrix. This establishes a transparent direct communication path between the underlying approval release node and the processing entry of the government service one-stop online service system front-end display page. This is the core data link for personalized recommendations to achieve a superior interactive experience of "click to process, enjoy without application".

[0084] The sequence encapsulation submodule, based on the association mapping matrix, performs alignment and merging of government name sequences and jump address sequences to construct network payloads, performs protocol packaging operations on network payloads, collects device screen parameters, combines positioning values ​​with device screen parameters to determine rendering landing points, and generates terminal coordinate display instructions.

[0085] The sequence encapsulation submodule reads the contents of the previously generated association mapping matrix, extracts the associated key-value pair elements distributed within the matrix, and performs alignment and merging processing on the Unicode character encoding of the government name sequence and the Uniform Resource Locator (URI) of the jump address sequence according to a fixed byte spacing to construct the network payload. This module directly pushes the constructed network payload to the Transmission Control Network (TCN) protocol stack to perform protocol packaging operations, including encapsulating the data packet header and cyclic redundancy check (CRC) verification bits. The sequence encapsulation submodule obtains the screen hardware specifications of the user's terminal device through the system's underlying application programming interface (API), extracting the horizontal and vertical physical pixel counts of the display panel. When the device screen parameter return value shows a horizontal pixel count of... And the number of vertical pixels is At that time, the module combines the received touch center point positioning value with the device screen parameters and calls the arithmetic logic unit to perform center point offset calculation, dividing the horizontal pixels by Obtain the x-coordinate of the rendering landing point Divide the vertical pixels by Obtain the ordinate of the rendering landing point .

[0086] Table 3 Terminal Coordinate Command Mapping Table

[0087]

[0088] Referring to the mapping rules in Table 3, the sequence encapsulation submodule packages and generates terminal coordinate display instructions that are sent to the graphics card register based on the coordinate calculation results. This precise positioning instruction ensures that the personalized recommendation card component generated by the government service platform can be adaptively centered and optimized on screens of various physical sizes and resolutions, greatly improving the visual acquisition efficiency of users receiving recommendation information.

[0089] The rendering driver submodule extracts internal coordinates and business data based on terminal coordinate display instructions. It performs raster mapping processing on the internal coordinates and business data to construct the underlying pixel matrix. It then performs timing allocation on the underlying pixel matrix with the refresh frequency to establish personalized recommendation results for government services through the online platform.

[0090] The rendering driver submodule, based on the data packet structure of the terminal coordinate display instructions, extracts the internal coordinate values ​​of the independent light-emitting units controlling the screen, as well as the service-carrying data that needs to be displayed to the operator, from the underlying instruction parsing cache queue. This module performs raster mapping processing on all extracted internal coordinates and service-carrying data containing specific color depth codes, filling the values ​​one-to-one into the video memory register addresses. This forces the abstract service-carrying data stream into a dot matrix composed of multiple rows and columns of primary color codes, thereby constructing the underlying pixel matrix. When the constructed underlying pixel matrix precisely contains... take When the pixel count is 1, the rendering driver submodule will match the underlying pixel matrix with the display hardware's factory-set speed per second. The refresh rate is handled by a timer interrupt-based timing allocation process for the frame output interval, ensuring that the graphics card control core refreshes every [time period]. The module sends a complete pixel matrix data frame to the display panel with millisecond precision. By maintaining a high frequency of pixel matrix timing output, it controls the update of screen content, ultimately creating a personalized recommendation result for government services that is intuitively presented to end users. This transforms the originally highly homogenized and cumbersome government processing procedures into a highly intelligent and compliant underlying computing power driven by the integration of machine learning prediction algorithms, achieving a deeply customized interface that perfectly meets the specific application needs of people of different ages and asset sizes. This fundamentally puts into practice the modern and efficient public service concept of deeply integrating dynamic and personalized recommendation technology for government services.

[0091] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection of the described technical solutions.

Claims

1. A personalized recommendation engine system for one-stop online government services, characterized in that, include: The identity feature extraction module collects government service identity recognition signals and converts them into a digital number sequence. It extracts the corresponding historical number of services and credit score and performs character concatenation operation with the digital number sequence to obtain the initial sparse qualification sequence. The feature array decomposition module combines the initial sparse quality sequence into a two-dimensional feature array by row, and performs singular value decomposition on the two-dimensional feature array to split it into a left singular array, a core array and a right singular array, thus constructing a singular decomposition feature array. The qualification feature completion module extracts the target service entity line sequence, constructs a predicted missing qualification sequence by combining it with the singular decomposition feature array, and merges and splices the predicted missing qualification sequence with the initial sparse qualification sequence to generate a complete government qualification sequence. The rule boundary comparison module extracts the lower limit boundary of the admission age and the upper limit boundary of the registered capital, compares the qualification parameters corresponding to the complete government qualification sequence with the upper and lower limit boundaries, and establishes an admission permit logical identifier. The terminal rendering driver module performs an extraction operation on the access permission logical identifier, associated with the government name sequence and the jump address sequence, encapsulates the two sequences and converts them into terminal coordinate display instructions, and establishes personalized recommendation results for one-stop government services.

2. The personalized recommendation engine system for one-stop online government services as described in claim 1, characterized in that: In the process of constructing the predicted missing qualification sequence, the diagonal sequence of the core array is extracted by combining the singular decomposition feature array, the target service subject row sequence is aligned and merged with the diagonal sequence of the core array to construct the intermediate transition feature sequence, the corresponding column sequence of the right singular array is extracted simultaneously, and the intermediate transition feature sequence is mapped and fused with the corresponding column sequence.

3. The personalized recommendation engine system for one-stop online government services as described in claim 1, characterized in that: During the size comparison process, if the value is greater than the lower limit and less than the upper limit, a high-level indicator is output; if the value is less than the lower limit or greater than the upper limit, a low-level indicator is output.

4. The personalized recommendation engine system for one-stop online government services as described in claim 1, characterized in that: The initial sparse qualification sequence includes an identity authentication code string, performance behavior rating, and government interaction frequency. The singular decomposition feature array includes a user implicit preference vector, an attribute dimensionality reduction weight matrix, and a transaction potential association table. The complete government qualification sequence includes full-dimensional verification indicators, estimated asset level, and potential application categories. The access permit logical identifier includes a compliance pass, a conditional rejection code, and a manual review requirement marker. The personalized recommendation results of the one-stop government service platform include customized service guides, dedicated business direct links, and intelligent form filling guidance pages.

5. The personalized recommendation engine system for one-stop online government services as described in claim 1, characterized in that, The identity feature extraction module includes: The signal parsing submodule collects government service identity recognition signals and converts them into digital number sequences. It receives simulated business test data, extracts basic comparison identifiers from the simulated business test data, performs a bit-by-bit comparison operation between the basic comparison identifiers and the digital number sequences, determines heterogeneous deviation intervals, removes sequence segments that fall within the heterogeneous deviation intervals, sorts the removed digital number sequences according to their positional order, and generates a feature relationship mapping quantity. The qualification concatenation submodule extracts the corresponding historical service counts and credit scores, performs format conversion processing on the historical service counts and credit scores to obtain a fixed-length bit string, compares the fixed-length bit string with the first and last nodes of the feature relationship mapping to determine overlapping nodes, removes redundant characters in the overlapping nodes, and performs character concatenation operation on the removed fixed-length bit string and the feature relationship mapping to obtain the initial sparse qualification sequence.

6. The personalized recommendation engine system for one-stop online government services according to claim 1, characterized in that, The feature array decomposition module includes: The feature combination submodule extracts the internal separators of the initial sparse quality sequence, performs segmentation based on the separators to obtain multiple groups of segments, and combines the multiple groups of segments row by row according to a fixed step size to generate a two-dimensional feature array. The singular decomposition submodule performs singular value decomposition on the two-dimensional feature array to split it into a left singular array, a core array, and a right singular array. It compares the elements in the core array with the retention benchmark threshold to determine the difference range, removes the values ​​falling outside the difference range, and obtains the dimensionality-reduced core feature quantity. The feature construction submodule, based on the dimensionality reduction core feature quantity, performs transpose on the left singular array and the right singular array to obtain a transposed singular array, rearranges and maps the dimensionality reduction core feature quantity with the transposed singular array to obtain a corrected array, and performs dot product fusion on the corrected array and the left singular array to generate a singular decomposition feature array.

7. The personalized recommendation engine system for one-stop online government services as described in claim 1, characterized in that, The qualification feature completion module includes: The sequence merging submodule collects the target service entity line sequence, extracts the core array diagonal sequence by combining the singular decomposition feature array, performs length padding operation on the core array diagonal sequence and the target service entity line sequence to obtain equal-length numerical segments, adds the elements in the equal-length numerical segments to generate the merging transition feature quantity; The mapping and fusion submodule extracts the corresponding column sequence of the right singular array based on the merged transition feature quantity, performs a dot product mapping between the corresponding column sequence and the merged transition feature quantity to obtain the fusion prediction value, and arranges the fusion prediction value in position order to generate the prediction supplementary feature quantity. The qualification splicing submodule compares the predicted supplementary feature quantity with the initial sparse qualification sequence to determine the placeholder node, and embeds the predicted supplementary feature quantity into the placeholder node in the initial sparse qualification sequence to generate a complete government qualification sequence.

8. The personalized recommendation engine system for one-stop online government services according to claim 1, characterized in that, The rule boundary comparison module includes: The boundary extraction submodule acquires a sample dataset, extracts the lower limit boundary of the admission age and the upper limit boundary of the registered capital of the sample dataset, combines the lower limit boundary of the admission age and the upper limit boundary of the registered capital to obtain the dual-end judgment benchmark parameters, extracts the corresponding qualification parameters in the complete government qualification sequence, performs a sorting and alignment operation on the corresponding qualification parameters and the dual-end judgment benchmark parameters to construct a comparison tuple array, and generates a boundary comparison reference set. The logic level submodule, based on the boundary comparison reference set, performs a size comparison logic operation between the internally arranged elements and the dual-end judgment benchmark parameter. If it is greater than the lower limit and less than the upper limit, it outputs a high-level flag to the data transmission port. If it is less than the lower limit or greater than the upper limit, it outputs a low-level flag to the data transmission port. It performs timing conversion on the collected high-level flags and low-level flags to establish an access permission logic flag.

9. The personalized recommendation engine system for one-stop online government services according to claim 1, characterized in that, The terminal rendering driver module includes: The association extraction submodule extracts the association comparison parameters within the preset sample data, compares the access permission logical identifier with the association comparison parameters to determine the logical node, extracts the government name sequence and the jump address sequence, and generates an association mapping matrix. The sequence encapsulation submodule, based on the association mapping matrix, performs alignment and merging on the government name sequence and the jump address sequence to construct the network payload, performs protocol packaging operation on the network payload, collects device screen parameters, combines the positioning value with the device screen parameters to determine the rendering landing point, and generates terminal coordinate display instructions; The rendering driver submodule extracts internal coordinates and service carrying data based on the terminal coordinate display instructions. It performs raster mapping processing on the internal coordinates and the service carrying data to construct a bottom-level pixel matrix. It then performs timing allocation on the bottom-level pixel matrix with the refresh frequency to establish personalized recommendation results for government services through the online platform.

Images