A method and system for pre-retirement qualification based on large language models

Abstract

The application relates to a retirement qualification pre-examination method and system based on a large language model, wherein the method comprises obtaining declaration data of an application object, obtaining structured declaration information and unstructured declaration information based on the declaration data, calling preset rigid conditions in a preset rule library, checking the structured declaration information to obtain a checking result, when the checking result represents that the preset rigid conditions are met, performing semantic reasoning on the structured declaration information and the unstructured declaration information by using a pre-trained large language model to obtain an analysis result representing whether the preset flexible rules are met. Based on the checking result and the analysis result, retirement qualification pre-examination conclusion data corresponding to the application object and explanatory report data associated with the retirement qualification pre-examination conclusion data are generated. The problem of how to improve the accuracy of pre-examination results in the related art is solved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method and system for pre-screening retirement eligibility based on a large language model. Background Technology

[0002] With the continuous expansion of pension insurance coverage and the massive growth of insured data, retirement eligibility verification, as a core pre-processing step in pension disbursement, is gradually transforming towards intelligent and automated processes. Intelligent pre-review of retirement eligibility, relying on insured contribution data, personal account interest calculation rules, refund calculation standards, and policy rules, pre-verifies and determines the retirement conditions of insured individuals. This has become a key technological direction for social security e-government. Its core lies in accurately executing business logic such as personal account principal and interest calculation, refund amount calculation, and eligibility verification, ensuring that the review results are compliant, efficient, and traceable.

[0003] Current retirement eligibility verification technologies generally rely on fixed code rules and structured data comparison. They use preset formulas to perform rigid calculations such as annual principal and interest payments and personal account refunds, and make eligibility judgments based on fixed thresholds such as age and years of contribution. Some systems output verification results in report form, which to some extent replaces manual basic data calculations and standardized verifications. These technologies aim to automate processes, focusing on improving data processing speed and reducing manual operations. However, they struggle to handle complex scenarios such as rule interpretation and edge case judgment. Especially for flexible policy provisions (such as semantic matching for special occupations, combined calculation of multiple insurance periods, and determination of deemed contribution years), existing technologies lack effective semantic understanding and logical reasoning methods, resulting in insufficient accuracy of verification results.

[0004] Therefore, no effective solution has yet been proposed for improving the accuracy of retirement eligibility pre-screening methods based on large language models. Summary of the Invention

[0005] This application provides a retirement eligibility pre-screening method and system based on a large language model, to at least address the problem of how to improve the accuracy of pre-screening results in related technologies.

[0006] In this regard, embodiments of this application provide a retirement eligibility pre-screening method based on a large language model, the method comprising: Obtain the application data of the applicant, and based on the application data, obtain structured application information and unstructured application information; The structured declaration information is verified by invoking preset rigid conditions from the preset rule base, and the verification result is obtained. When the verification result characterization meets the preset rigid conditions, the structured declaration information and the unstructured declaration information are semantically reasoned through a pre-trained large language model to obtain the analysis result of whether the characterization conforms to the preset flexible rules.

[0007] Based on the verification results and the analysis results, retirement eligibility pre-review conclusion data corresponding to the applicant is generated, as well as explanatory report data associated with the retirement eligibility pre-review conclusion data.

[0008] In one embodiment, obtaining the application data of the applicant, and based on the application data, obtaining structured application information and unstructured application information, includes: Multimodal application data of applicants is obtained through various data collection methods, including voice interaction, text input, image scanning, and interface connection with government databases. The multimodal declaration data is parsed to obtain key fields and text descriptions. Specifically, the declaration data obtained by voice interaction is subjected to voice recognition to extract key fields and text descriptions, and the declaration data obtained by image scanning is subjected to optical character recognition and layout analysis to extract key fields and text descriptions. The key fields and the text descriptions are fused and cleaned to generate structured and unstructured declaration information. The structured declaration information includes formatted and stored numerical fields, and the unstructured declaration information includes descriptive text in natural language form.

[0009] In one embodiment, before invoking preset rigid conditions from a preset rule base, the method further includes: Dynamically acquire policy and regulatory text data; Natural language parsing is performed on the policy and regulation text data to extract policy clauses, applicable objects, preset rigid conditions, and preset flexible rules. The preset rigid conditions include hard numerical thresholds and a list of legal entities. A dynamically updated policy knowledge graph is constructed based on the policy provisions, the applicable objects, the preset rigid conditions, and the preset flexible rules, and the policy knowledge graph is stored as a preset rule base; wherein, the policy knowledge graph uses the policy provisions, the applicable objects, the preset rigid conditions, and the preset flexible rules as nodes, and the relationships between the policy knowledge graph, the policy provisions, the applicable objects, the preset rigid conditions, and the preset flexible rules as edges.

[0010] In one embodiment, the step of calling preset rigid conditions from a preset rule base to verify the structured declaration information and obtain a verification result includes: Based on the field identifiers of each key field in the structured declaration information, the nodes that are associated with each key field are traversed in the policy knowledge graph to obtain the preset rigid conditions corresponding to the key fields. Based on the preset rigid conditions found in the query, determine the condition type of the preset rigid conditions, which includes numerical threshold type and directory list type; When the condition type is a numerical threshold type, the numerical value of the corresponding field is extracted from the structured declaration information, and the numerical value is compared with the hard numerical threshold stored in the preset rigid conditions. Based on the comparison result, a verification pass flag or a verification fail flag is generated. When the condition type is a directory list type, extract the text value of the corresponding field from the structured declaration information, query whether the text value exists in the statutory directory list stored in the preset rigid conditions, and generate a verification pass or verification fail flag based on the query result. The verification pass flag and verification fail flag for each field are used as the verification result.

[0011] In one embodiment, when the verification result characterization meets a preset rigid condition, a pre-trained large language model is used to perform semantic reasoning on the structured declaration information and the unstructured declaration information to obtain an analysis result on whether the characterization conforms to a preset flexible rule, including: Extract the preset flexible rules associated with the descriptive text from the unstructured declaration information from the preset rule base; The structured declaration information, the unstructured declaration information, and the preset flexible rules are used as context text and input into the pre-trained large language model. The pre-trained large language model uses the preset flexible rules as the basis for reasoning to perform conformity analysis on the structured declaration information and the unstructured declaration information, and generates analysis results.

[0012] In one embodiment, the pre-trained large language model, using the preset flexible rules as the basis for reasoning, performs a conformity analysis on the structured declaration information and the unstructured declaration information, generating analysis results, including: Extract natural language descriptive text from the unstructured application information, and extract the job title field corresponding to the descriptive text from the structured application information; Query the preset flexible rules to find the special job category list matching rules corresponding to the job title field; The job title field is semantically matched with the legally recognized special job titles recorded in the special job title directory matching rules to generate job matching results; And / or, extract multi-regional insurance records from the structured declaration information, and perform time overlap detection and payment month merging calculation on the multi-regional insurance records according to the payment merging processing rules related to multi-regional insurance in the preset flexible rule clauses, and generate payment merging calculation results; The analysis results are generated based on the job matching results and / or the combined payment calculation results.

[0013] In one embodiment, the analysis result is generated based on the job matching result and / or the combined payment calculation result. When the job name field of the job matching result successfully matches the name of the legal special occupation, a job analysis conclusion that meets the retirement conditions for special occupations is generated; And / or, when the combined payment calculation result indicates that the total number of payment months after the combined calculation of the multi-regional insurance records meets the preset payment month threshold, a payment analysis conclusion that meets the payment years condition is generated; The analysis results are generated based on the job analysis conclusions and / or the payment analysis conclusions.

[0014] In one embodiment, generating retirement eligibility pre-assessment conclusion data corresponding to the applicant and explanatory report data associated with the retirement eligibility pre-assessment conclusion data based on the verification results and the analysis results includes: Based on the verification results, the verification status of each of the preset rigid conditions in the structured declaration information is determined, and based on the analysis results, the conformity status of the structured declaration information and the unstructured declaration information for each of the preset flexible rules is determined. When all the verification statuses are verified as passed and the compliance status indicates compliance, the pre-approval conclusion data indicating that the retirement qualification pre-approval has been passed is generated, and the explanatory report data is generated based on the verification status and the compliance status. When the verification status fails, extract the clause identifier and the original text of the preset rigid condition where the verification status fails, generate the pre-examination conclusion data representing the failure of the retirement qualification pre-examination, and write the clause identifier and the original text of the clause into the explanatory report data; When all the verification statuses are verified as passed but the compliance status characterization is not met, the clause identifier and reasoning description text corresponding to the non-compliance status characterization are extracted, the pre-examination conclusion data representing doubts about the retirement qualification pre-examination is generated, and the clause identifier and the reasoning description text are written into the explanatory report data.

[0015] In one embodiment, after generating retirement eligibility pre-assessment conclusion data corresponding to the applicant and explanatory report data associated with the retirement eligibility pre-assessment conclusion data, the method further includes: Obtain the review and correction results of the preliminary review conclusion data; compare the review and correction results with the preliminary review conclusion data to extract case samples that deviate from the preset flexible rules; Based on the case samples and the corresponding structured declaration information, unstructured declaration information, and preset flexible rules, an incremental training dataset is constructed. The large language model is fine-tuned or trained using the incremental training dataset to update the large language model.

[0016] Secondly, embodiments of this application provide a retirement eligibility pre-screening system based on a large language model. The system is used to execute the above-described method and includes an application information module, a verification module, an analysis result module, and a pre-screening conclusion module; wherein: The application information module is used to obtain the application data of the applicant, and based on the application data, obtain structured application information and unstructured application information; The verification module is used to call preset rigid conditions in the preset rule base to verify the structured declaration information and obtain the verification result. The analysis result module is used to perform semantic reasoning on the structured declaration information and the unstructured declaration information through a pre-trained large language model when the verification result representation meets the preset rigid conditions, so as to obtain the analysis result of whether the representation conforms to the preset flexible rules.

[0017] The preliminary review conclusion module is used to generate retirement eligibility preliminary review conclusion data corresponding to the applicant, and explanatory report data associated with the retirement eligibility preliminary review conclusion data, based on the verification results and the analysis results.

[0018] Thirdly, embodiments of this application provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a retirement eligibility pre-screening method based on a large language model as described above.

[0019] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a retirement eligibility pre-screening method based on a large language model as described above.

[0020] The retirement eligibility pre-screening method and system based on a large language model provided in this application embodiment has at least the following technical effects.

[0021] The system acquires application data from applicants and, based on this data, extracts both structured and unstructured application information. It can uniformly transform disorganized application materials from different sources and in different formats into standardized structured fields and natural language descriptions. It calls upon preset rigid conditions in a pre-defined rule base to validate the structured application information and obtain validation results. It automatically compares numerical fields using hard thresholds and legal directories in the rule base, achieving accurate determination of explicit conditions such as contribution years and age. When the validation result indicates that the system meets the preset rigid conditions, a pre-trained large language model performs semantic reasoning on the structured and unstructured application information to obtain an analysis result on whether the representation conforms to preset flexible rules. Deep reasoning by the large language model is initiated only when the rigid conditions are met, performing contextual understanding and logical deduction on unstructured text such as work experience descriptions. This solves the ambiguity and complex correlation problems that existing technologies struggle to handle, such as matching special occupations and conflicts between multiple insurance locations, significantly improving the accuracy of the preliminary review. Based on the validation and analysis results, it generates retirement qualification preliminary review conclusion data corresponding to the applicant, as well as explanatory report data associated with the retirement qualification preliminary review conclusion data. This enables applicants to clearly understand the review results and their policy basis, and addresses the issue of how to improve the accuracy of pre-review results in related technologies.

[0022] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0023] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of a retirement eligibility pre-screening method based on a large language model; Figure 2 This is a flowchart illustrating step S102 according to an exemplary embodiment; Figure 3 This is a flowchart illustrating step S103 according to an exemplary embodiment; Figure 4 This is a system architecture block diagram of a retirement eligibility pre-screening system based on a large language model, according to an exemplary embodiment. Figure 5 This is a structural block diagram of an electronic device according to an exemplary embodiment. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the application is described and illustrated below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application.

[0025] Obviously, the accompanying drawings described below are merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar scenarios based on these drawings without any inventive effort. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, any changes to design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as insufficient disclosure of the content of this application.

[0026] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0027] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms “a,” “an,” “an,” “the,” and similar words used in this application do not indicate quantity limitation and may indicate singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms “connected,” “linked,” “coupled,” and similar words used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following objects are in an "or" relationship. The terms "", "second", "third", etc., used in this application are merely to distinguish similar objects and do not represent a specific ordering of objects.

[0028] In this document, it should be understood that the terms used may be technical means used to implement part of the present invention or other summary technical terms. For example, the terms may include: Structured declaration information: refers to the collection of numeric and enumerated fields extracted from various declaration data submitted by applicants and formatted. For example: date of birth (date type), cumulative number of months of payment (numeric type), job title (enumerated type), and multi-regional insurance records (list structure).

[0029] Unstructured application information: refers to descriptive text in natural language form extracted from various application data submitted by applicants that cannot be directly stored in a fixed field format. Examples include: "Worked in underground tunneling at a certain mine from 1985 to 1990" or "2 years of service as an educated youth, 3 years of military service" and other descriptions of work experience.

[0030] Pre-defined rigid conditions: These refer to rigid review standards identified and structured from policy and regulatory texts, which can be directly determined through numerical comparison calculations or set member queries. For example, the threshold of 180 corresponds to "cumulative payment months ≥ 180 months"; the statutory list corresponds to "special types of work".

[0031] Pre-defined flexible rules: These refer to soft review standards identified and structured from policy and regulatory texts, which require semantic understanding and logical deduction based on context to make a judgment. Examples include: rules for determining deemed contribution years, rules for merging overlapping time records from multiple locations, and regional supplementary regulations for special occupation lists.

[0032] Directory list type: refers to one of the condition types in the preset rigid conditions, indicating that the judgment logic of the rigid condition is "to determine whether the value of a certain field in the structured declaration information exists in a given set of legal directories".

[0033] Special job category matching rules: These refer to the semantic matching rules in the preset flexible rules used to determine whether the job performed by the applicant belongs to a legally defined special job category.

[0034] Consolidated payment calculation result: This refers to the intermediate data product generated by the large language model after conflict detection and consolidation simulation of insurance records from multiple regions. It includes the total number of payment months after time overlap detection and deduplication consolidation, and the determination status of whether it meets the preset payment month threshold.

[0035] Firstly, embodiments of this application provide a retirement eligibility pre-screening method based on a large language model. Figure 1 This is a flowchart of a retirement eligibility pre-screening method based on a large language model, such as... Figure 1 As shown, the method includes: Step S101: Obtain the application data of the applicant, and based on the application data, obtain structured application information and unstructured application information.

[0036] Step S102: Call the preset rigid conditions in the preset rule base to verify the structured declaration information and obtain the verification result.

[0037] Step S103: When the verification result representation meets the preset rigid conditions, semantic reasoning is performed on the structured and unstructured declaration information through a pre-trained large language model to obtain the analysis result of whether the representation conforms to the preset flexible rules.

[0038] Step S104: Based on the verification results and analysis results, generate retirement eligibility pre-examination conclusion data corresponding to the applicant, as well as explanatory report data associated with the retirement eligibility pre-examination conclusion data.

[0039] In summary, this application provides a retirement eligibility pre-screening method based on a large language model. By acquiring the applicant's declaration data, structured and unstructured declaration information are obtained. This method can uniformly transform disorganized declaration materials from different sources and formats into standardized structured fields and natural language descriptions. Preset rigid conditions from a pre-defined rule base are invoked to verify the structured declaration information, yielding verification results. Numerical fields are automatically compared using hard thresholds and legal directories in the rule base, enabling accurate determination of explicit conditions such as contribution years and age. When the verification result indicates that the pre-defined rigid conditions are met, a pre-trained large language model performs semantic reasoning on the structured and unstructured declaration information to obtain an analysis result determining whether the indication conforms to pre-defined flexible rules. Deep reasoning by the large language model is initiated only when the rigid conditions are met, performing contextual understanding and logical deduction on unstructured text such as work experience descriptions. This solves the problems of ambiguity and complex correlations, such as matching special occupations and conflicts between multiple insurance locations, which are difficult to handle with existing technologies, significantly improving the accuracy of pre-screening. Based on the verification and analysis results, preliminary retirement eligibility assessment conclusion data corresponding to the applicant is generated, along with explanatory report data associated with the preliminary retirement eligibility assessment conclusion data. This enables applicants to clearly understand the assessment results and their policy basis, addressing the issue of how to improve the accuracy of preliminary assessment results in related technologies.

[0040] In one embodiment, step S101 involves obtaining the application data of the applicant, and based on the application data, obtaining structured application information and unstructured application information. Specifically, this includes the following steps: Step S1011: Obtain multimodal application data of the applicant through various data collection methods, including voice interaction, text input, image scanning and government database interface connection; Step S1012: Analyze the multimodal declaration data to obtain key fields and text descriptions. Specifically, for declaration data obtained through voice interaction, key fields and text descriptions are extracted through voice recognition. For declaration data obtained through image scanning, key fields and text descriptions are extracted through optical character recognition and layout analysis. Step S1013: Perform data fusion and cleaning on key fields and text descriptions to generate structured and unstructured declaration information. Structured declaration information includes formatted and stored numerical fields, while unstructured declaration information includes descriptive text in natural language form.

[0041] Optionally, multimodal application data for retirement qualification processing can be obtained through four data collection methods: voice interaction, text input, image scanning, and government database interface. For voice-based application data collected through voice interaction, automatic speech recognition technology is used to transcribe the speech, accurately extracting key fields and corresponding text descriptions required for business processing. For image-based documents such as certificates, archives, and insurance materials collected through image scanning, optical character recognition combined with layout structure analysis is used to recognize text content and parse related information such as table layout and material format, simultaneously extracting effective key fields and text descriptions. For authorized data synchronously retrieved from text input and government database interfaces, standardized field information is directly captured. Data fusion, deduplication, error correction, and format cleaning are uniformly performed on key fields and text descriptions obtained from all channels to integrate multi-source heterogeneous information and finally generate structured application information containing formatted numerical content such as date of birth, cumulative months of payment, and years of insurance, as well as unstructured application information containing natural language content such as work experience and special job experience, forming a complete and standardized comprehensive information file for the applicant.

[0042] Step S101 is compatible with multiple forms of data collection for declarations, broadens information acquisition channels, and simplifies user declaration operations; multimodal data collaborative analysis and fusion cleaning can fill information gaps and ensure the completeness and standardization of declaration information.

[0043] In one embodiment, before invoking preset rigid conditions in a preset rule base, the method further includes: Dynamically acquire policy and regulatory text data; Natural language parsing is performed on policy and regulatory text data to extract policy clauses, applicable objects, pre-set rigid conditions and pre-set flexible rules. Pre-set rigid conditions include hard numerical thresholds and legal list lists. A dynamically updated policy knowledge graph is constructed based on policy provisions, applicable objects, pre-set rigid conditions, and pre-set flexible rules. The policy knowledge graph is stored as a pre-set rule base. The policy knowledge graph uses policy provisions, applicable objects, pre-set rigid conditions, and pre-set flexible rules as nodes, and the relationships between the policy knowledge graph, policy provisions, applicable objects, pre-set rigid conditions, and pre-set flexible rules as edges.

[0044] Optionally, the system dynamically captures retirement-related policies, administrative regulations, departmental notices, and policy interpretations issued by local human resources and social security departments in real time, continuously updating and synchronously collecting policy source data. Natural language processing technology is used to perform deep semantic analysis and content segmentation on the collected policy and regulatory texts, extracting standardized policy clauses, clearly defined applicable populations and scopes, hard numerical thresholds, and statutory lists as pre-defined rigid conditions. Simultaneously, pre-defined flexible rules, including exceptions, supplementary provisions, and scenario-based requirements, are identified. Using the parsed policy clauses, applicable populations, pre-defined rigid conditions, and pre-defined flexible rules as nodes, and the subordinate, limiting, and relational constraints between nodes as connecting edges, a real-time iteratively updated policy knowledge graph is constructed. This completed policy knowledge graph is then uniformly stored and used as a pre-defined rule base for long-term business review and judgment.

[0045] It can keep up with policy changes in real time and realize the dynamic iteration and updating of policy rules; through structured parsing and knowledge graph modeling, it transforms scattered text policies into a related and visualized rule system, unifies the management of rigid constraints and flexible clauses, and greatly improves the comprehensiveness and accuracy of subsequent qualification review and judgment.

[0046] In one embodiment, Figure 2 This is a flowchart illustrating step S102 according to an exemplary embodiment, as follows: Figure 2 As shown, step S102 involves calling the preset rigid conditions in the preset rule base to verify the structured declaration information and obtain the verification result. Specifically, this includes the following steps: Step S1021: Based on the field identifiers of each key field in the structured declaration information, traverse the nodes that are related to each key field in the policy knowledge graph and query the preset rigid conditions corresponding to the key fields. Step S1022: Based on the queried preset rigid conditions, determine the condition type of the preset rigid conditions. The condition types include numerical threshold type and directory list type. Step S1023: When the condition type is a numerical threshold type, extract the value of the corresponding field from the structured declaration information, compare the value with the hard numerical threshold stored in the preset rigid conditions, and generate a verification pass flag or verification fail flag based on the comparison result. Step S1024: When the condition type is the directory list type, extract the text value of the corresponding field from the structured declaration information, query whether the text value exists in the statutory directory list stored in the preset rigid conditions, and generate a verification pass or verification fail flag based on the query result. Step S1025: Use the validation pass and validation fail flags of each field as the validation result.

[0047] Optionally, by using the unique field identifier corresponding to each key field in the structured declaration information, the system traverses the pre-built and dynamically updated policy knowledge graph one by one to accurately retrieve graph nodes that have a relationship of constraint with each key field, quickly match and retrieve the preset rigid conditions of the corresponding content; further, all preset rigid conditions obtained from the query are classified and identified, distinguishing between two types of conditions: numerical threshold type and directory list type; for rigid conditions of numerical threshold type, the actual filled value of the corresponding business field is extracted from the structured declaration information, and the actual value is compared with the pre-stored rigid numerical threshold in the policy knowledge graph for comparison, range determination, etc. Based on the matching results, the corresponding verification pass or verification fail identifier is automatically generated; for rigid conditions of directory list type, the text content of the corresponding project is extracted from the structured declaration information, and it is searched whether the text content is included in the statutory directory list. Based on the query results of whether it is a match or not, the corresponding verification identifier is generated simultaneously; all the identifiers obtained after verifying all key fields one by one are summarized and integrated into a complete rigid condition verification result, completing the automated compliance verification of all rigid indicators.

[0048] Step S102 uses a policy knowledge graph to achieve precise association and matching between fields and rigid conditions, and classifies and differentiates numerical comparisons and directory retrieval verifications to effectively improve the efficiency and accuracy of rigid condition verification of application materials.

[0049] In one embodiment, Figure 3 This is a flowchart illustrating step S103 according to an exemplary embodiment, as follows: Figure 3 As shown, in step S103, when the verification result representation meets the preset rigid conditions, a pre-trained large language model is used to perform semantic reasoning on the structured and unstructured declaration information to obtain the analysis result of whether the representation conforms to the preset flexible rules. Specifically, this includes the following steps: Step S1031: Extract the preset flexible rules associated with the description text from the preset rule base in the unstructured declaration information.

[0050] Optionally, preset flexible policy clauses (preset flexible rules) related to the applicant's work experience description can be extracted from the preset rule base, such as the list of special occupations attached to the "Administrative Measures for Early Retirement of Special Occupations" and the supplementary provisions for the corresponding year in the region, and the implementation details for the combined processing of insurance contributions in multiple regions.

[0051] Step S1032: Input the structured declaration information, unstructured declaration information, and preset flexible rules as context text into the pre-trained large language model.

[0052] Optionally, the structured application information, the work experience description text in the unstructured application information, and the preset flexible rules are concatenated into contextual text and input as a whole into a pre-trained large language model that has been specially fine-tuned.

[0053] Step S1033: The pre-trained large language model, using preset flexible rules as the basis for reasoning, performs a conformity analysis on structured and unstructured declaration information, generating analysis results. Specifically, this includes: Extract descriptive text in natural language form from unstructured application information, and extract the job title field corresponding to the descriptive text from structured application information; Query the preset flexible rules to find the matching rules for the special job category list corresponding to the job title field; The job title field is semantically matched with the legally defined special job titles recorded in the special job title directory matching rules to generate job title matching results; And / or, extract multi-regional insurance records from structured declaration information, and perform time overlap detection and payment month merging calculation on the multi-regional insurance records according to the payment merging processing rules related to multi-regional insurance in the preset flexible rule clauses, and generate payment merging calculation results.

[0054] Analysis results are generated based on job matching results and / or combined payment calculation results. Specifically, these include: When the job title field of the job matching result successfully matches the name of a legally defined special occupation, a job analysis conclusion that meets the retirement conditions for special occupations is generated. And / or, when the combined calculation result of the payment indicates that the total number of payment months after the combined calculation of insurance records from multiple regions meets the preset threshold for the number of payment months, a payment analysis conclusion that meets the payment years condition is generated; Analysis results are generated based on the conclusions of job analysis and / or payment analysis.

[0055] Optionally, after receiving the context text, the big language model performs three aspects of deep reasoning: First, in terms of context understanding and information association, the big language model extracts the descriptive text "worked in underground tunneling at a certain mine from 1985 to 1990" from the unstructured declaration information, extracts the corresponding job title field "underground tunneling worker" from the structured declaration information, and then queries the special job category matching rules from the preset flexible rules to semantically match "underground tunneling worker" with the legal title "underground coal miner" recorded in the category—based on the understanding that both "tunneling" and "coal mining" belong to underground mining operations, the big language model judges that the two are substantially equivalent and generates a job matching result of "job matching successful"; Second, according to the policy of converting deemed contribution years, the big language model applies the "2 years for educated youth" and "military service years" in the unstructured declaration information. The system identifies "3 years" and calculates, based on the conversion factor for the length of service of educated youth and the rules for determining the deemed contribution of military service during the corresponding period, 24 months of educated youth service and 36 months of military service can be converted, for a total of 60 months of deemed contribution. Third, the big data model extracts multi-regional insurance data recorded in the structured declaration information. For example, the insurance period in City A is from January 2006 to December 2010, and the insurance period in City B is from June 2009 to August 2015. It detects that there is an overlap in the insurance period between the two places from June 2009 to December 2010. Then, according to the principle of "no double accumulation within the overlapping time period" in the rules for merging insurance contributions in multiple regions, the overlapping period is deducted and the effective contribution months of the two places are merged to generate the contribution merger calculation result. At the same time, the overlapping period is marked as a potential problem point for subsequent manual review. The large language model integrates the above job matching results, deemed payment conversion results, and payment consolidation calculation results into an analysis result, which includes conformity identifiers representing compliance with preset flexible rules, as well as natural language conclusion description text for each inference dimension.

[0056] Step S103 uses a large language model to achieve semantic understanding and complex logical reasoning of flexible clauses, breaking through the limitations of existing technologies that can only perform rigid field comparisons. It can accurately adapt to non-standard and complex business scenarios such as special job identification, multi-location payment consolidation and accounting, and year conversion, thereby improving the accuracy of pre-review results and solving the problem of how to improve the accuracy of pre-review results in related technologies.

[0057] In one embodiment, step S104 involves generating retirement eligibility pre-assessment conclusion data corresponding to the applicant, and explanatory report data associated with the retirement eligibility pre-assessment conclusion data, based on the verification and analysis results. Specifically, this includes the following steps: Step S1041: Based on the verification results, determine the verification status of each preset rigid condition in the structured declaration information, and based on the analysis results, determine the compliance status of the structured declaration information and the unstructured declaration information with respect to each preset flexible rule. Step S1042: When all verification statuses are verified as passed and the compliance status indicates compliance, generate preliminary review conclusion data indicating that the retirement qualification pre-review has been passed, and generate explanatory report data based on the verification status and compliance status. Step S1043: When the verification status fails, extract the clause identifier and the original text of the clause with the preset rigid condition of verification status failing, generate the pre-examination conclusion data representing the failure of the retirement qualification pre-examination, and write the clause identifier and the original text of the clause into the explanatory report data. Step S1044: When all verification statuses are verified as passed but the compliance status characterization is not met, extract the clause identifier and reasoning description text corresponding to the non-compliance status characterization, generate the pre-examination conclusion data indicating doubt in the pre-examination of retirement eligibility, and write the clause identifier and reasoning description text into the explanatory report data.

[0058] Optionally, based on the results of the rigid condition verification, the pass / fail status of each preset rigid condition, such as age, minimum contribution period, and legally mandated inclusion, is determined. Simultaneously, based on the results of the flexible condition analysis, the compliance status of matters such as special work type identification, multi-location contribution consolidation, and deemed years of service calculation is determined. If all rigid condition verifications are passed and all flexible condition compliance statuses are compliant, a preliminary approval conclusion is automatically generated, and all verification details and flexible reasoning processes are summarized to form a complete explanatory report. If any rigid condition verification fails, for example... If the accumulated years of contribution do not reach the statutory threshold or the occupation is not on the statutory list, the corresponding rigid conditions clause identifiers and policy texts will be automatically retrieved, a preliminary review failure conclusion will be generated, and the problematic clause content will be simultaneously written into the explanatory report. If all rigid conditions are verified as qualified, but the semantic matching of special occupations is not up to standard, or there are contradictions in the calculation of contributions in multiple locations, etc., the flexible matters will be judged as non-compliant, the corresponding flexible policy clause identifiers and complete reasoning description texts will be extracted, a preliminary review doubt conclusion will be generated, and the reasons for doubt, reasoning logic and corresponding policy basis will be simultaneously marked and retained in the explanatory report, thus retaining the complete judgment basis of the entire process.

[0059] Step S104 differentiates the results into three categories: pre-approved, unapproved, and questionable. Simultaneously, it automatically generates a traceable explanatory report, fully preserving the policy basis, verification details, and reasoning process. This enhances the transparency and traceability of the audit results, facilitates manual review, and reduces the risk of audit disputes.

[0060] In one embodiment, after generating retirement eligibility pre-assessment conclusion data corresponding to the applicant and explanatory report data associated with the retirement eligibility pre-assessment conclusion data, the method further includes: Obtain the results of the review and correction of the preliminary review conclusion data; compare the differences between the review and correction results and the preliminary review conclusion data, and extract case samples that deviate from the preset flexible rules; An incremental training dataset is constructed based on case samples and the corresponding structured and unstructured declaration information and preset flexible rules. By using incremental training datasets, large language models can be fine-tuned or trained using reinforcement learning to update them.

[0061] Optionally, a second review is conducted on the retirement eligibility pre-examination conclusion data automatically output by the system, and correction opinions are entered to obtain the review and correction results. The review and correction results are compared with the pre-examination conclusion data item by item to accurately screen and select problematic cases that lead to inconsistent judgments due to misunderstandings of flexible conditions or insufficient reasoning in complex scenarios. These cases are used as samples of deviation cases. For each sample of deviation cases, the structured application information, unstructured application description text, the relevant preset flexible rules, and the judgment logic after manual correction are completely collected and retained. These are integrated and organized to form a standardized and reusable incremental training dataset. The completed incremental training dataset is used to continuously iterate and optimize the pre-trained large language model used for flexible condition judgment by using instruction fine-tuning or reinforcement learning training methods. This allows the model to continuously learn subtle differences in policies, marginal special cases, and manual correction logic, thereby completing the dynamic iterative upgrade of the model's capabilities.

[0062] By reviewing discrepancies and establishing a closed-loop feedback mechanism, we continuously identify model shortcomings and policy adaptation blind spots. Based on incremental data iterative training, we achieve autonomous optimization and upgrading of the large language model. It can adaptively adapt to local policy updates and changes in special business scenarios, continuously improve the semantic understanding of flexible clauses and the ability to reason about complex logic, and continuously enhance the accuracy, universality, and adaptive evolution capabilities of the overall pre-review and review process.

[0063] In summary, this application provides a retirement eligibility pre-screening method based on a large language model. By acquiring the applicant's declaration data, structured and unstructured declaration information are obtained. This method can uniformly transform disorganized declaration materials from different sources and formats into standardized structured fields and natural language descriptions. Preset rigid conditions from a pre-defined rule base are invoked to verify the structured declaration information, yielding verification results. Numerical fields are automatically compared using hard thresholds and legal directories in the rule base, enabling accurate determination of explicit conditions such as contribution years and age. When the verification result indicates that the pre-defined rigid conditions are met, a pre-trained large language model performs semantic reasoning on the structured and unstructured declaration information to obtain an analysis result determining whether the indication conforms to pre-defined flexible rules. Deep reasoning by the large language model is initiated only when the rigid conditions are met, performing contextual understanding and logical deduction on unstructured text such as work experience descriptions. This solves the problems of ambiguity and complex correlations, such as matching special occupations and conflicts between multiple insurance locations, which are difficult to handle with existing technologies, significantly improving the accuracy of pre-screening. Based on the verification and analysis results, preliminary retirement eligibility assessment conclusion data corresponding to the applicant is generated, along with explanatory report data associated with the preliminary retirement eligibility assessment conclusion data. This enables applicants to clearly understand the assessment results and their policy basis, addressing the issue of how to improve the accuracy of preliminary assessment results in related technologies.

[0064] Secondly, embodiments of this application provide a retirement eligibility pre-screening system based on a large language model. Figure 4 This is a system architecture block diagram illustrating a retirement eligibility pre-screening system based on a large language model, according to an exemplary embodiment. For example... Figure 4 As shown, the system includes an application information module 410, a verification module 420, an analysis results module 430, and a preliminary review conclusion module 440; wherein: The application information module 410 is used to obtain the application data of the applicant and, based on the application data, obtain structured application information and unstructured application information. The verification module 420 is used to call the preset rigid conditions in the preset rule base to verify the structured declaration information and obtain the verification result; The analysis results module 430 is used to perform semantic reasoning on structured and unstructured declaration information through a pre-trained large language model when the verification result representation meets the preset rigid conditions, and obtain the analysis results of whether the representation conforms to the preset flexible rules.

[0065] The preliminary review conclusion module 440 is used to generate retirement eligibility preliminary review conclusion data corresponding to the applicant, as well as explanatory report data associated with the retirement eligibility preliminary review conclusion data, based on the verification results and analysis results.

[0066] In summary, the retirement eligibility pre-screening system based on a large language model provided in this application solves the problem of how to improve the accuracy of pre-screening results in related technologies through the synergistic effect of the application information module 410, the verification module 420, the analysis result module 430, and the pre-screening conclusion module 440.

[0067] It should be noted that the retirement eligibility pre-screening system based on a large language model provided in this embodiment is used to implement the above-described implementation methods, and details already described will not be repeated. As used above, terms such as "module," "unit," and "subunit" can refer to combinations of software and / or hardware that perform predetermined functions. Although the apparatus described in the above embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0068] Thirdly, embodiments of this application provide an electronic device, Figure 5 This is a block diagram illustrating an electronic device according to an exemplary embodiment. (e.g.) Figure 5 As shown, the electronic device may include a processor 81 and a memory 82 storing computer program instructions.

[0069] Specifically, the processor 81 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0070] The memory 82 may include a mass storage device for data or instructions. For example, and not limitingly, the memory 82 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 82 may include removable or non-removable (or fixed) media. Where appropriate, the memory 82 may be internal or external to a data processing device. In a particular embodiment, the memory 82 is non-volatile memory. In a particular embodiment, the memory 82 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random-Access Memory (FPMDRAM), Extended Data Out Dynamic Random-Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0071] The memory 82 can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 81.

[0072] The processor 81 reads and executes computer program instructions stored in the memory 82 to implement any of the retirement eligibility pre-screening methods based on a large language model in the above embodiments.

[0073] In one embodiment, a retirement eligibility pre-screening device based on a large language model may further include a communication interface 83 and a bus 80. Wherein, as... Figure 5 As shown, the processor 81, memory 82, and communication interface 83 are connected through bus 80 and complete communication with each other.

[0074] The communication interface 83 is used to enable communication between the various modules, devices, units, and / or equipment in the embodiments of this application. The communication interface 83 can also enable data communication with other components such as external devices, image / data acquisition devices, databases, external storage, and image / data processing workstations.

[0075] Bus 80 includes hardware, software, or both, that couples together components of a retirement eligibility pre-screening device based on a large language model. Bus 80 includes, but is not limited to, at least one of the following: Data Bus, Address Bus, Control Bus, Expansion Bus, and Local Bus. For example, and not as a limitation, bus 80 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. Where appropriate, bus 80 may include one or more buses. Although specific buses are described and illustrated in the embodiments of this application, this application considers any suitable bus or interconnection.

[0076] Fourthly, embodiments of this application provide a computer-readable storage medium storing a program thereon, which, when executed by a processor, implements a retirement eligibility pre-screening method based on a large language model provided in this aspect.

[0077] The readable storage medium may be more specifically adopted, including but not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.

[0078] In a possible implementation, the invention can also be implemented as a program product comprising program code that, when the program product is run on a terminal device, causes the terminal device to perform steps of a retirement eligibility pre-screening method based on a large language model provided in the implementation.

[0079] The program code for executing the present invention can be written in any combination of one or more programming languages. The program code can be executed entirely on the user device, partially on the user device, as a standalone software package, partially on the user device and partially on a remote device, or entirely on a remote device.

[0080] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0081] The above embodiments merely illustrate several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for pre-screening retirement eligibility based on a large language model, characterized in that, The method includes: Obtain the application data of the applicant, and based on the application data, obtain structured application information and unstructured application information; The structured declaration information is verified by invoking preset rigid conditions from the preset rule base, and the verification result is obtained. When the verification result characterization meets the preset rigid conditions, the structured declaration information and the unstructured declaration information are semantically reasoned through a pre-trained large language model to obtain the analysis result of whether the characterization conforms to the preset flexible rules. Based on the verification results and the analysis results, retirement eligibility pre-examination conclusion data corresponding to the applicant is generated, as well as explanatory report data associated with the retirement eligibility pre-examination conclusion data.

2. The retirement eligibility pre-screening method based on a large language model according to claim 1, characterized in that, When the verification result characterization meets the preset rigid conditions, a pre-trained large language model is used to perform semantic reasoning on the structured declaration information and the unstructured declaration information to obtain an analysis result on whether the characterization conforms to the preset flexible rules, including: Extract the preset flexible rules associated with the descriptive text from the unstructured declaration information from the preset rule base; The structured declaration information, the unstructured declaration information, and the preset flexible rules are used as context text and input into the pre-trained large language model. The pre-trained large language model uses the preset flexible rules as the basis for reasoning to perform conformity analysis on the structured declaration information and the unstructured declaration information, and generates analysis results.

3. The retirement eligibility pre-screening method based on a large language model according to claim 2, characterized in that, The pre-trained large language model, using the preset flexible rules as the basis for reasoning, performs conformity analysis on the structured declaration information and the unstructured declaration information, generating analysis results, including: Extract natural language descriptive text from the unstructured application information, and extract the job title field corresponding to the descriptive text from the structured application information; Query the preset flexible rules to find the special job category list matching rules corresponding to the job title field; The job title field is semantically matched with the legally recognized special job titles recorded in the special job title directory matching rules to generate job matching results; And / or, extract multi-regional insurance records from the structured declaration information, and perform time overlap detection and payment month merging calculation on the multi-regional insurance records according to the payment merging processing rules related to multi-regional insurance in the preset flexible rule clauses, and generate payment merging calculation results; The analysis results are generated based on the job matching results and / or the combined payment calculation results.

4. The retirement eligibility pre-screening method based on a large language model according to claim 3, characterized in that, The analysis results are generated based on the job matching results and / or the combined payment calculation results, including: When the job name field of the job matching result successfully matches the name of the legal special occupation, a job analysis conclusion that meets the retirement conditions for special occupations is generated; And / or, when the combined payment calculation result indicates that the total number of payment months after the combined calculation of the multi-regional insurance records meets the preset payment month threshold, a payment analysis conclusion that meets the payment years condition is generated; The analysis results are generated based on the job analysis conclusions and / or the payment analysis conclusions.

5. The retirement eligibility pre-screening method based on a large language model according to claim 1, characterized in that, Before invoking the preset rigid conditions in the preset rule base, the method further includes: Dynamically acquire policy and regulatory text data; Natural language parsing is performed on the policy and regulation text data to extract policy clauses, applicable objects, preset rigid conditions, and preset flexible rules. The preset rigid conditions include hard numerical thresholds and a list of legal entities. A dynamically updated policy knowledge graph is constructed based on the policy provisions, the applicable objects, the preset rigid conditions, and the preset flexible rules, and the policy knowledge graph is stored as a preset rule base; wherein, the policy knowledge graph uses the policy provisions, the applicable objects, the preset rigid conditions, and the preset flexible rules as nodes, and the relationships between the policy knowledge graph, the policy provisions, the applicable objects, the preset rigid conditions, and the preset flexible rules as edges.

6. The retirement eligibility pre-screening method based on a large language model according to claim 5, characterized in that, The method of calling preset rigid conditions from a preset rule base to verify the structured declaration information and obtaining verification results includes: Based on the field identifiers of each key field in the structured declaration information, the nodes that are associated with each key field are traversed in the policy knowledge graph to obtain the preset rigid conditions corresponding to the key fields. Based on the preset rigid conditions found in the query, determine the condition type of the preset rigid conditions, which includes numerical threshold type and directory list type; When the condition type is a numerical threshold type, the numerical value of the corresponding field is extracted from the structured declaration information, and the numerical value is compared with the hard numerical threshold stored in the preset rigid conditions. Based on the comparison result, a verification pass flag or a verification fail flag is generated. When the condition type is a directory list type, extract the text value of the corresponding field from the structured declaration information, query whether the text value exists in the statutory directory list stored in the preset rigid conditions, and generate a verification pass or verification fail flag based on the query result. The verification pass flag and verification fail flag for each field are used as the verification result.

7. The retirement eligibility pre-screening method based on a large language model according to claim 1, characterized in that, The process of generating retirement eligibility pre-assessment conclusion data corresponding to the applicant and explanatory report data associated with the retirement eligibility pre-assessment conclusion data, based on the verification results and the analysis results, includes: Based on the verification results, the verification status of each of the preset rigid conditions in the structured declaration information is determined, and based on the analysis results, the conformity status of the structured declaration information and the unstructured declaration information for each of the preset flexible rules is determined. When all the verification statuses are verified as passed and the compliance status indicates compliance, the pre-approval conclusion data indicating that the retirement qualification pre-approval has been passed is generated, and the explanatory report data is generated based on the verification status and the compliance status. When the verification status fails, extract the clause identifier and the original text of the preset rigid condition where the verification status fails, generate the pre-examination conclusion data representing the failure of the retirement qualification pre-examination, and write the clause identifier and the original text of the clause into the explanatory report data; When all the verification statuses are verified as passed but the compliance status characterization is not met, the clause identifier and reasoning description text corresponding to the non-compliance status characterization are extracted, the pre-examination conclusion data representing doubts about the retirement qualification pre-examination is generated, and the clause identifier and the reasoning description text are written into the explanatory report data.

8. The retirement eligibility pre-screening method based on a large language model according to claim 1, characterized in that, The process of obtaining the applicant's declaration data, and based on that data, obtaining structured and unstructured declaration information, includes: Multimodal application data of applicants is obtained through various data collection methods, including voice interaction, text input, image scanning, and interface connection with government databases. The multimodal declaration data is parsed to obtain key fields and text descriptions. Specifically, the declaration data obtained by voice interaction is subjected to voice recognition to extract key fields and text descriptions, and the declaration data obtained by image scanning is subjected to optical character recognition and layout analysis to extract key fields and text descriptions. The key fields and the text descriptions are fused and cleaned to generate structured and unstructured declaration information. The structured declaration information includes formatted and stored numerical fields, and the unstructured declaration information includes descriptive text in natural language form.

9. The retirement eligibility pre-screening method based on a large language model according to claim 1, characterized in that, After generating retirement eligibility pre-assessment conclusion data corresponding to the applicant, and explanatory report data associated with the retirement eligibility pre-assessment conclusion data, the method further includes: Obtain the review and correction results of the preliminary review conclusion data; compare the review and correction results with the preliminary review conclusion data to extract case samples that deviate from the preset flexible rules; Based on the case samples and the corresponding structured declaration information, unstructured declaration information, and preset flexible rules, an incremental training dataset is constructed. The large language model is fine-tuned or trained using the incremental training dataset to update the large language model.

10. A retirement eligibility pre-screening system based on a large language model, characterized in that, The system is used to execute the method according to any one of claims 1 to 9, and the system includes an application information module, a verification module, an analysis result module, and a preliminary examination conclusion module; wherein: The application information module is used to obtain the application data of the applicant, and based on the application data, obtain structured application information and unstructured application information; The verification module is used to call preset rigid conditions in the preset rule base to verify the structured declaration information and obtain the verification result. The analysis result module is used to perform semantic reasoning on the structured declaration information and the unstructured declaration information through a pre-trained large language model when the verification result representation meets the preset rigid conditions, so as to obtain the analysis result of whether the representation conforms to the preset flexible rules. The preliminary review conclusion module is used to generate retirement eligibility preliminary review conclusion data corresponding to the applicant, and explanatory report data associated with the retirement eligibility preliminary review conclusion data, based on the verification results and the analysis results.

Images