Human resources file intelligent structured display method based on large language model

Abstract

The present application relates to the technical field of data processing, and more particularly to a personnel file intelligent structured display method based on a large language model, the display method comprising: obtaining paragraph data units through target requirements, and judging whether to adjust the weight proportion of the paragraph data units in a display template based on paragraph confidence; if adjustment is not required, judging whether the paragraph data units meet the target requirements through semantic matching degree; if the target requirements are not met, judging whether the paragraph data units can be content-completed through the large language model based on the difference in semantic matching degree; thereby splitting the paragraph data units meeting the target requirements into semantic blocks through the large language model and dynamically redistributing the semantic blocks to generate a display template; and judging whether the display template is qualified based on the model confidence of the display template, thereby generating a final structured display template. Through the multi-level quality evaluation and dynamic adjustment mechanism, the present application realizes intelligent understanding, quality control and adaptive display of personnel file content, and effectively improves the availability and display effect of the file data.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for intelligent structured display of personnel files based on a large language model. Background Technology

[0002] Traditional paper archives generally suffer from problems such as large storage space requirements, low retrieval efficiency, and easy information damage. There is an urgent need to use intelligent methods to achieve automatic identification, structured extraction, and dynamic management of archive content.

[0003] Chinese Patent Publication No. CN119670723A discloses an intelligent archive management method based on a large language model. Existing technical solutions mainly focus on tamper prevention and compliance verification of archives. This method involves randomly selecting original archives to form a content verification group, using an archive verification plugin to perform compliance verification, detecting whether the archives have been tampered with, locating the tampered content and suspicious operations when tampering is detected, and finally displaying the results through graphical comparison layers. This method effectively solves the problems of authenticity assurance, compliance verification, and operation traceability in the process of digital archive transfer, providing technical support for archive security.

[0004] However, the above solutions still have the following problems: the relevant technical solutions mainly focus on verifying the authenticity and compliance of the archives, that is, ensuring that the content of the archives has not been tampered with, but they are significantly lacking in terms of deep semantic understanding, intelligent parsing and structured display of the content of the archives; they cannot handle the problem of content that does not meet the requirements or low confidence information, and they cannot dynamically adjust the display strategy according to the content quality, resulting in low utilization of low-quality archive data.

[0005] Therefore, there is an urgent need for a method that can intelligently analyze and semantically understand the content of personnel files, and dynamically and adaptively display the content based on content quality and user needs, in order to solve the problems of insufficient utilization of file content, rigid display methods, and inability to adapt to intelligent applications in multiple scenarios in existing technologies. Summary of the Invention

[0006] To address this, the present invention provides an intelligent structured display method for personnel files based on a large language model, which overcomes the problems in existing technologies such as lack of deep semantic understanding of file content, inability to process information that does not meet requirements or has low confidence, resulting in static and rigid display methods.

[0007] To achieve the above objectives, this invention provides a method for intelligent structured display of personnel files based on a large language model, comprising: Acquire several paragraph data units based on the target requirements; Calculate the paragraph confidence level corresponding to several paragraph data units, and determine whether to adjust the weight ratio of the paragraph data units in the display template based on the comparison result between the corresponding paragraph confidence level and the confidence level threshold. If it is determined that no adjustment is needed, the semantic matching degree corresponding to several paragraph data units is calculated, and based on the comparison result between the corresponding semantic matching degree and the preset semantic matching degree, it is determined whether the current paragraph data unit meets the requirements. If it is determined that it does not meet the requirements, the completion strategy is determined based on the comparison result between the corresponding semantic matching degree difference and the preset semantic matching degree difference. The semantic matching degree difference is calculated based on the difference between the semantic matching degree and the preset semantic matching degree. If it is determined that the completion strategy will be executed, the large language model will be used to complete the data to make the current paragraph data unit meet the requirements based on the comparison relationship between the corresponding semantic matching degree difference and the hierarchical semantic matching degree difference. If it is determined that the completion strategy will not be executed, the weight ratio of the paragraph data units that do not meet the requirements in the display template will be reduced. Based on the paragraph data unit that meets the requirements, the current paragraph data unit is split into several semantic blocks, and the several semantic blocks are dynamically redistributed to generate the display template; Calculate the model confidence score of the display template, and determine whether the display template is qualified based on the comparison result between the model confidence score and the model confidence score threshold. If it is determined to be unqualified, optimize the display template based on the large language model. If it is determined to be qualified, generate a structured display template by combining the model confidence score and the display template.

[0008] Furthermore, the process of determining whether to adjust the weight ratio of the paragraph data unit in the display template includes: The confidence level of the paragraph is determined by the large language model based on logical analysis and a pre-set external authoritative database. The current paragraph data unit whose paragraph confidence is less than the confidence threshold is recorded as a low-confidence paragraph data unit; The current paragraph data unit whose paragraph confidence is greater than or equal to the confidence threshold is denoted as a high-confidence paragraph data unit.

[0009] Furthermore, if the current paragraph data unit is the low-confidence paragraph data unit, a first adjustment strategy is determined to reduce the weight ratio of the low-confidence paragraph data unit in the display template. If the current paragraph data unit is the high-confidence paragraph data unit, determine to maintain the weight ratio of the current paragraph data unit in the display template.

[0010] Furthermore, the process of executing the first adjustment strategy based on the large language model includes: Reduce the correlation strength between several of the high-confidence paragraph data units and the current low-confidence paragraph data units in the display template.

[0011] Furthermore, the process of determining whether the current paragraph data unit meets the requirements includes: The semantic matching degree is determined by the large language model based on a preset element dictionary; If the semantic matching degree is less than the preset semantic matching degree, it is determined that the current paragraph data unit does not meet the requirements and it is determined whether to execute the completion strategy. If the semantic matching degree is greater than or equal to the preset semantic matching degree, it is determined that the current paragraph data unit meets the requirements; The preset element dictionary includes a set of necessary intrinsic elements corresponding to different types of paragraph data units.

[0012] Furthermore, the process of determining whether to execute the completion strategy includes: If the semantic matching degree difference is greater than the preset semantic matching degree difference, it is determined that the current paragraph data unit will not execute the completion strategy, and a second adjustment strategy is generated to reduce the weight ratio of the paragraph data unit that does not meet the requirements in the structured display template. If the semantic matching degree difference is less than or equal to the preset semantic matching degree difference, the missing content corresponding to the current paragraph data unit is determined and the completion strategy is executed.

[0013] Furthermore, the process of executing the completion strategy includes: Based on a large language model, prompt words are designed to generate complete content corresponding to the missing content. The completion strength is determined by adjusting the generation parameters of the large language model. The generation parameters include temperature parameter values ​​or kernel sampling values. The temperature parameter values ​​are negatively correlated with the semantic matching degree difference, and the kernel sampling values ​​are positively correlated with the semantic matching degree difference.

[0014] Furthermore, the second adjustment strategy includes reducing the proportion of information displayed in the structured display template for paragraph data units that do not meet the requirements.

[0015] Furthermore, the process of determining whether the display template is qualified includes: The model confidence level is determined by the large language model based on target requirements and a pre-set external authoritative database. Based on the comparison results where the model confidence score is less than the model confidence threshold, the display template is determined to be unqualified, and a third adjustment strategy is generated to optimize the display template based on the large language model. The display template is deemed qualified based on the comparison result where the model confidence level is greater than or equal to the model confidence level threshold, and the structured display template is generated based on the model confidence level and the display template.

[0016] Furthermore, the third adjustment strategy includes adjusting the arrangement order and visual hierarchy of the displayed information in the display template; Alternatively, reallocate semantic blocks.

[0017] Compared with existing technologies, the beneficial effects of the intelligent structured display method for personnel files based on a large language model in this invention are as follows: It calculates the paragraph confidence level in the file data by using the target requirements as the starting point for data processing and determines whether to execute the first adjustment strategy based on the threshold comparison results, thereby achieving a quantitative evaluation of the data source quality and providing a reliable quality basis for subsequent processing. By defining and calculating the semantic matching degree of paragraph data units, judging whether it meets the requirements based on threshold comparison, and judging whether the completion conditions are met based on the semantic matching degree difference, and executing the completion strategy by using the semantic matching degree difference as a quantitative indicator of the degree of information loss, it significantly improves the compliance and usability of file information. Furthermore, by utilizing the deep semantic understanding capabilities of the large language model, it achieves intelligent parsing and reconstruction of file content, and performs confidence evaluation on the final display template to generate a structured display template, avoiding new errors that may be introduced during information reconstruction, ensuring quality monitoring of the intelligent structured display of personnel files, and maintaining consistency in logic and semantics.

[0018] Furthermore, this invention also determines paragraph confidence levels through a large language model based on logical analysis and a pre-set external authoritative database, elevating credibility assessment from subjective experience judgment to an objective quantitative indicator that combines internal logical self-consistency verification with cross-checking of external objective facts. This makes the first adjustment strategy determined based on paragraph confidence levels more credible and authoritative. Thus, by reducing the non-destructive data processing method of correlation strength, both data integrity and credibility management of output data are achieved.

[0019] Furthermore, this invention also evaluates the suitability of the target requirement through standardized semantic matching degree, and determines whether semantic paragraphs that do not match the target requirement meet the execution conditions of the completion strategy by the difference in semantic matching degree, thereby achieving accurate completion of missing content and avoiding the reduction of content fidelity due to over-processing; semantic paragraphs that do not meet the execution conditions of the completion strategy are downgraded to ensure the completeness of the logic.

[0020] Furthermore, this invention ensures the final quality assessment of the display template after paragraph splitting and semantic redistribution by evaluating the output quality confidence of the display template; and employs a third adjustment strategy to adjust information arrangement, visual hierarchy, or redistribute semantic blocks when the display template is unqualified. This ensures that the final deliverables to the user are always at a high level. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the intelligent structured display method for personnel files based on a large language model, as described in an embodiment of the present invention. Figure 2 This is a logic diagram for determining whether to execute the first adjustment strategy in an embodiment of the present invention; Figure 3 This is a logic diagram for determining whether to execute the second adjustment strategy in an embodiment of the present invention; Figure 4 This is a logic diagram for determining whether to execute the third adjustment strategy in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0023] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0024] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0025] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0026] Please see Figure 1The diagram shows a flowchart of the intelligent structured display method for personnel files based on a large language model, according to an embodiment of the present invention. The flowchart of this embodiment includes at least the following steps: S1: Obtain several paragraph data units based on the target requirements; S2: Calculate the paragraph confidence level corresponding to several paragraph data units, and determine whether to adjust the weight ratio of paragraph data units in the display template based on the comparison results between the corresponding paragraph confidence level and the confidence level threshold. S3: If it is determined that no adjustment is needed, calculate the semantic matching degree corresponding to several paragraph data units, and determine whether the current paragraph data unit meets the requirements based on the comparison result between the corresponding semantic matching degree and the preset semantic matching degree. S4: If it is determined that the requirements are not met, determine whether to execute the completion strategy based on the comparison result between the corresponding semantic matching degree difference and the preset semantic matching degree difference. The semantic matching degree difference is calculated based on the difference between the semantic matching degree and the preset semantic matching degree. S51: If it is determined that the completion strategy should be executed, the large language model should be used to complete the data unit of the current paragraph so that it meets the requirements based on the comparison relationship between the corresponding semantic matching degree difference and the hierarchical semantic matching degree difference. S52: If it is determined that the completion strategy will not be executed, reduce the weight of paragraph data units that do not meet the requirements in the display template; S6: Based on paragraph data units that meet the requirements, the current paragraph data unit is split into several semantic blocks, and these semantic blocks are dynamically redistributed to generate a display template; S7: Calculate the model confidence score of the display template, and determine whether the display template is qualified based on the comparison result between the model confidence score and the model confidence score threshold; S81: If the result is deemed unqualified, the display template will be optimized based on the large language model. S82: If the result is deemed satisfactory, determine whether to generate a structured display template by combining the model confidence level and the display template.

[0027] In this embodiment, the system receives user input or target requirements parsed from a preset template, such as "generating a comprehensive resume view for cadre promotion assessment" or "extracting a resume summary highlighting project management capabilities." Based on these target requirements, the system retrieves, locates, and extracts relevant text paragraphs from the original personnel file database, extracting several paragraph data units to form a set of paragraph data units. Each paragraph data unit is at least a semantic block with independent semantic meaning in the file, such as a complete description of work experience, an educational background introduction, or an award record.

[0028] Based on the business logic corresponding to preset templates or target requirements, the system assigns initial basic weights to paragraph data units using a weight configuration table through a large language model. This weight configuration table, set based on historical data analysis, defines the relative importance of various information dimensions under different target requirements. For example, for the target requirement "Cadre Promotion Assessment," the configuration can be selected as: {"Job Change Experience": 0.35, "Major Project Performance": 0.30, "Education and Training": 0.20, "Awards and Punishments Records": 0.15}. For the target requirement "Project Management Capability Summary," the configuration is adjusted to: {"Project Participation Experience": 0.50, "Relevant Skill Descriptions": 0.30, "Collaboration and Communication Examples": 0.20}. If it is determined that the weight ratio of paragraph data units in the display template needs adjustment, the system further adjusts it using the weight decay function within the weight configuration table.

[0029] Please see Figure 2 As shown, this is a logic decision diagram for determining whether to execute the first adjustment strategy in an embodiment of the present invention. In this embodiment, a large language model is used in conjunction with design prompts to guide the large language model in analyzing the logical consistency, temporal order, and contradictions within a paragraph. For example, the design prompts could be: "Please analyze the internal logical consistency of the following text paragraphs. Are there any temporal contradictions, factual conflicts, or illogical descriptions?" Key entities identified in the paragraphs, such as the graduating institution, employer, and certificate number, are compared with preset external authoritative databases such as the China Higher Education Student Information System (CHESICC) academic credentials database, the Industrial and Commercial Enterprise Information Database, and the National Vocational Qualification Certificate Inquiry System. Finally, based on the large language model combined with logical consistency and preset external authoritative databases, a paragraph confidence score C is generated. The paragraph confidence score C indicates the reliability quality of the data unit; a perfect match is 1, a complete mismatch or unverifiable result is 0, and the closer the C value is to 1, the more reliable the data unit in the paragraph.

[0030] In one specific embodiment, data units with high reliability should occupy a more prominent position in the final display, while data with low reliability should be downplayed to prevent misleading decisions. The weighting of data units in the display template is determined by comparing the results with a confidence threshold Cth. Cth is determined based on the analysis of historical high-quality personnel files. Specifically, data units deemed "accurate and reliable" are selected from the historical archive, and their C-value set is calculated and statistically analyzed. The 25th percentile of this set can be taken as Cth. For example, Cth can be set to 0.72. It should be noted that Cth can be adjusted according to the user's target needs.

[0031] If C ≥ Cth, it indicates that the information reliability meets the display standard. The confidence level of the current paragraph data unit is deemed acceptable, and its weight ratio in the display template is not adjusted. The current paragraph data unit is then recorded as a high-confidence paragraph data unit.

[0032] If C < Cth, it indicates that the data unit is insufficient to be presented as core or strongly related information in the display template. The confidence level of the current paragraph data unit is determined to be too low, and a first adjustment strategy is generated. The current paragraph data unit is then recorded as a low-confidence paragraph data unit.

[0033] Specifically, the weight of the current paragraph data unit in the display template is reduced by decreasing the association strength between the high-confidence paragraph data unit and the current low-confidence paragraph data unit in the display template. This is achieved by using a large language model for dynamic reallocation, constructing a semantic association network within the display template, and reducing the association edge weight between all semantic blocks of the low-confidence paragraph data unit and other unit semantic blocks based on design prompt words.

[0034] In this embodiment, for high-confidence paragraph data units with acceptable confidence levels, a semantic matching degree S is determined by a large language model based on a preset element dictionary. The preset element dictionary is a structured rule base that defines the necessary set of intrinsic elements required for different types of paragraph data units to meet specific target requirements. For example, for the target requirement "cadre promotion assessment," for paragraphs of the "work experience" category, the necessary set of intrinsic elements includes at least: start and end dates, work unit, position held, and main responsibilities or achievements. The large language model is guided by designed prompts to identify the type of the current paragraph data unit and, based on the corresponding necessary set of intrinsic elements, checks whether each element is explicitly mentioned or fully described in the paragraph. Finally, the large language model generates a semantic matching degree S, where the semantic matching degree S indicates the degree to which the content framework of the paragraph data unit fits the target requirement; a larger S value indicates that the content structure better meets the target requirement.

[0035] In one specific embodiment, a paragraph data unit that meets the target requirements should have complete information elements to support the analysis of the requirements. If the paragraph data unit is missing in structure, it may not be able to effectively serve the target analysis. Therefore, it is necessary to determine whether the severity of the missing information has reached a level that requires and is worth intervening in.

[0036] In this embodiment, the current paragraph data unit is determined to meet the target requirements by comparing the semantic matching degree S with the preset semantic matching degree Sth. The preset semantic matching degree Sth is the minimum qualification requirement for the completeness and matching degree of the content of a paragraph data unit. The Sth setting is determined based on business logic and historical statistical analysis. Specifically, for the scenario of "cadre promotion assessment", key information such as work experience must be highly complete. Therefore, a higher Sth can be set. For example, Sth=0.8 can be set.

[0037] If S < Sth, it indicates that there are structural deficiencies in the paragraph content. The current paragraph data unit is determined to be unsuitable for the requirements. The decision on whether to execute the completion strategy is based on the semantic matching degree difference ΔS. If S≥Sth, it indicates that the complete information elements are met. The current paragraph data unit is determined to meet the requirements. The completion decision is not executed, and a display template is generated based on the large language model.

[0038] In one specific embodiment, the completion strategy is based on inferential completion of contextual semantics and a preset external authoritative database. It is not the creation of content out of thin air, but a reasonable completion that strictly follows the principles of structured extraction and intelligent summarization.

[0039] Specifically, once the system determines to perform completion, it uses existing information within the current paragraph data unit and information from other related paragraphs in the same file to construct a timeline and logical chain for inferential completion. It also provides objective factual constraints for completion based on a pre-set external authoritative database. The system guides the large language model to infer based on the clearly defined reasoning scope of the prompts. For example, a work experience record states: "From July 2015 to 2018, worked at [XX Province, YY City, ZZ Bureau], mainly responsible for information technology construction and network security management." However, it lacks a clear "position held." The system identifies this experience as being in the middle of the individual's career, with the previous experience being "staff member" and the subsequent experience being "deputy section chief." By querying the external database, the system finds the regular internal departments and job sequences of the "ZZ Bureau" from 2015 to 2018 (such as the office, science and technology information department, etc.) and matches the job description "information technology construction and network security." The corresponding position is usually "deputy section chief," and the system uses this inferred "deputy section chief" as the completion content.

[0040] If it is determined that a completion decision should be made, the semantic matching degree difference ΔS = Sth - S is calculated. The severity of the missing information is quantified by ΔS. The completion strategy is determined based on the comparison between the corresponding semantic matching degree difference ΔS and the preset semantic matching degree difference ΔSth. The completion strategy is determined by the comparison between the semantic matching degree difference ΔS and the hierarchical semantic matching degree difference θ. The preset semantic matching degree difference ΔSth is determined based on the maximum value of the completion capability of the large language model. For example, ΔSth = 0.40 can be set. The hierarchical semantic matching difference θ includes a first hierarchical semantic matching difference θ1 and a second hierarchical semantic matching difference θ2. The hierarchical semantic matching difference θ is set based on historical completion case data and historical usage experience. Specifically, from the database of the historical record management system, the final verified completion operation records are selected, and the semantic matching degree S before completion and the semantic matching degree difference ΔS are calculated. The strategy type used for completion is extracted to form a historical case dataset. Based on this dataset, statistical analysis is performed, and the ΔS value is divided into multiple continuous intervals. The proportion of completion results determined to be completely acceptable within each interval is counted as the basic completion success rate for that interval. The ΔS value corresponding to the starting point where the completion success rate curve begins to decline significantly and steadily is set as the first threshold θ1. When the success rate of the conventional completion strategy has dropped to an extremely low level, and the highest level of completion mode must be used to achieve the desired effect, the ΔS value is set as the second threshold θ2. For example, θ1 = 0.15, θ2 = ΔSth = 0.40.

[0041] Specifically, if ΔS≤ΔSth, the completion strategy is executed. Based on the semantic understanding and element recognition capabilities of the large language model and the design prompt words, the large language model is guided to check the necessary intrinsic element set corresponding to the preset element dictionary to verify elements that are not explicitly mentioned or are not adequately described. Different completion strengths are adopted for different levels of missing elements.

[0042] The completion strategy determined based on the comparison results of the semantic matching degree difference ΔS and the preset quality grading threshold θ can be found in Table 1.

[0043] Table 1. Mapping relationship between completion strategy and semantic matching degree difference ΔS

[0044] Please see Figure 3As shown, this is a logic diagram for determining whether to execute the second adjustment strategy in an embodiment of the present invention. Specifically, if ΔS > ΔSth, it indicates severe missing information, and the existing context information is insufficient to support reliable completion reasoning. Forced completion will result in extremely low fidelity. Therefore, it is determined that the current paragraph data unit will not execute the completion strategy, and a second adjustment strategy is generated to reduce the information priority of the paragraph data unit that does not meet the requirements in the template.

[0045] Specifically, the second adjustment strategy includes guiding the large language model to conduct a multi-dimensional semantic importance assessment of the current paragraph data unit based on design prompt words. The importance assessment includes the completeness of the paragraph content and the relative value of meeting the target requirements in the overall context, thereby reducing the information proportion of the current paragraph data unit in the semantic association network during the semantic block redistribution process.

[0046] In one specific implementation, based on a large language model, paragraph data units that meet the target requirements are divided into several semantic blocks, such as "time blocks," "main behavior blocks," "achievement indicator blocks," and "skill tag blocks." These semantic blocks are then dynamically redistributed according to the display logic corresponding to the target requirements. For example, for a "cadre resume timeline" requirement, the "time blocks" and "main behavior blocks" extracted from all paragraphs are reorganized in chronological order; for a "capability matrix" requirement, all "skill tag blocks" and "achievement indicator blocks" are categorized and aggregated, ultimately generating a display template that includes the organizational structure of the archival information, the content sources of each module, and the preliminary visual presentation relationships.

[0047] Please see Figure 4 As shown, this is the logical decision diagram for determining whether to execute the third adjustment strategy in an embodiment of the present invention. The system and the display template generate a confidence network in parallel. This confidence network corresponds one-to-one with the information nodes of the display template, aiming to provide a transparent and traceable quality assessment for each information dimension of the final output, corresponding to each specific information node in the display structure. For example, a specific work experience entry (e.g., "2015-2018, Senior Engineer at Company A"), a skill tag (e.g., "Java Programming"), or an honor received. Each information node is formed by the redistribution of semantic blocks, and a comprehensive confidence level corresponding one-to-one with each information node is calculated.

[0048] Specifically, the overall confidence score is obtained by comprehensively evaluating the paragraph confidence score C and completeness assessment of the original paragraph data unit of the current information node through a large language model, as well as the confidence score assessment of the supplementary content. Based on the semantic associations such as information nodes, representing time sequence, reporting relationship, and project affiliation, the certainty and support of each association connection line in the context are evaluated, and an association confidence score is generated for it. For example, the confidence score of the causal relationship "experience A leads to skill B" may be high, while the confidence score of the association "project C is highly related to project D" may be medium.

[0049] In this embodiment, the model confidence Cm is determined based on the target requirements and a preset external authoritative database using a large language model on the basis of a confidence network. Specifically, the large language model evaluates whether the overall narrative of the presentation template is smooth, whether there are logical jumps or contradictions between different parts, checks whether the presentation template fully responds to all dimensions of the target requirements, and uses related information from the preset external authoritative database to corroborate the core conclusive information extracted from the presentation template. Finally, the model confidence Cm of the presentation template is output by weighted averaging the comprehensive confidence of all information nodes in the confidence network. The model confidence Cm indicates the credibility of the final generated presentation template. The closer the Cm value is to 1, the higher the overall quality and reliability of the presentation template.

[0050] In this embodiment, the qualification of the display template is determined based on the comparison between the model confidence score Cm and the model confidence threshold Cmth. The model confidence threshold Cmth is determined by backtesting historical qualified display templates, calculating the distribution of their Cm values, and selecting the 70th percentile as the model confidence threshold Cmth. For example, Cmth can be set to 0.75.

[0051] If Cm < Cmth, the display template is deemed unqualified, and a third adjustment strategy is generated to optimize the overall structure and content organization of the display template.

[0052] Specifically, the logical contradictions in the current display template are analyzed using a large language model, and semantic blocks are reordered to improve the information flow. For example, resumes arranged in chronological order are adjusted to prioritize "experiences most relevant to the research objective". Based on content importance analysis and user attention models, the visual presentation parameters of the display template are adjusted, such as redistributing title levels according to the importance of semantic blocks and their relevance to the target needs. Key information is displayed in bold or larger font size, while secondary information uses a regular font. Different color saturation and brightness values ​​are assigned based on the information's confidence level and importance. The relative position and space occupied by each information module on the page are recalculated. Alternatively, semantic blocks are redistributed based on the importance and confidence level of the display template content, migrating some semantic blocks from the original modules to new modules that better match their semantics, thereby optimizing the display template and regenerating the model confidence network.

[0053] If Cm≥Cmth, the display template is deemed qualified. Based on the large language model and combined with the model confidence network, the output is a structured display template that can be directly delivered to the front-end interface for visualization.

[0054] Specifically, the structured presentation template is a dynamically growing tree-like data structure used to intuitively represent an individual's overall development within the organizational system. The tree structure uses the "personal topic" as the root node. Several main branches derive from the root node, representing core information dimensions such as timelines, organizational levels, and competency dimensions. On each main branch, leaf nodes grow in chronological order or logical relationships, representing specific experiences, events, or competencies (e.g., "Manager of XX Department from 2005-2010," "Leaded the YY project and won an innovation award"). Semantic relationships between nodes are expressed through connecting lines, indicating chronological order, reporting relationships, and project affiliation.

[0055] When new archival materials (such as the latest appointment documents and training certificates) are processed by the system, there is no need to reconstruct the whole. The system will automatically identify its information type and timestamp, and intelligently insert it as a new node into the appropriate position of the corresponding main branch, or establish a new association with existing nodes, so as to intuitively restore the continuous trajectory of personal growth and ability evolution.

[0056] While generating the display template, the system combines a confidence network to ultimately output a structured display template that integrates a complete tree structure, confidence data for all nodes / edges, and dynamic updates. This template can be directly rendered by the front-end visualization engine. During rendering, the size, color, and border thickness of nodes, as well as the visual attributes of related edges such as solidity and transparency, can be dynamically mapped according to their confidence values, allowing users to intuitively perceive the reliability of the information. The final structured display template is output in standard data formats (such as JSON and XML) and supports seamless integration with government information systems such as cadre management systems, talent databases, and digital personnel platforms. It can provide a data foundation for intelligent application scenarios such as cadre selection, talent profiling, policy matching, and risk warning.

[0057] In this embodiment, the structured display template achieves high-precision and highly adaptable display of personnel files. While ensuring the compliance of government affairs, it can effectively handle unstructured files with diverse formats and ambiguous content. It breaks through the traditional flat file storage mode and uses a tree structure to truly restore an individual's growth path and ability evolution in the organizational system. It realizes the leap from "keyword matching" to "semantic reasoning" and can generate high-level tags with policy context and organizational management significance. It meets the stringent requirements of organizational departments for the authenticity, completeness, traceability and calculability of files.

[0058] All technologies not mentioned in the above embodiments are applicable to existing technologies. It is understood that no specific limitation is made to any preset parameter or critical parameter in the embodiments of the present invention, and the above values ​​are not limited thereto. Those skilled in the art can adjust the preset parameters or critical parameters accordingly based on actual needs, analysis of historical data, or equipment usage.

[0059] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for intelligent structured display of personnel files based on a large language model, characterized in that, include: Acquire several paragraph data units based on the target requirements; Calculate the paragraph confidence level corresponding to several paragraph data units, and determine whether to adjust the weight ratio of the paragraph data units in the display template based on the comparison result between the corresponding paragraph confidence level and the confidence level threshold. If it is determined that no adjustment is needed, the semantic matching degree corresponding to several paragraph data units is calculated, and based on the comparison result between the corresponding semantic matching degree and the preset semantic matching degree, it is determined whether the current paragraph data unit meets the requirements. If it is determined that it does not meet the requirements, the completion strategy is determined based on the comparison result between the corresponding semantic matching degree difference and the preset semantic matching degree difference. The semantic matching degree difference is calculated based on the difference between the semantic matching degree and the preset semantic matching degree. If it is determined that the completion strategy will be executed, the large language model will be used to complete the data to make the current paragraph data unit meet the requirements based on the comparison relationship between the corresponding semantic matching degree difference and the hierarchical semantic matching degree difference. If it is determined that the completion strategy will not be executed, the weight ratio of the paragraph data units that do not meet the requirements in the display template will be reduced. Based on the paragraph data unit that meets the requirements, the current paragraph data unit is split into several semantic blocks, and the several semantic blocks are dynamically redistributed to generate the display template; Calculate the model confidence score of the display template, and determine whether the display template is qualified based on the comparison result between the model confidence score and the model confidence score threshold. If it is determined to be unqualified, optimize the display template based on the large language model. If it is determined to be qualified, generate a structured display template by combining the model confidence score and the display template.

2. The intelligent structured display method for personnel files based on a large language model according to claim 1, characterized in that, The process of determining whether to adjust the weight ratio of the paragraph data unit in the display template includes: The confidence level of the paragraph is determined by the large language model based on logical analysis and a pre-set external authoritative database. The current paragraph data unit whose paragraph confidence is less than the confidence threshold is recorded as a low-confidence paragraph data unit; The current paragraph data unit whose paragraph confidence is greater than or equal to the confidence threshold is denoted as a high-confidence paragraph data unit.

3. The intelligent structured display method for personnel files based on a large language model according to claim 2, characterized in that, If the current paragraph data unit is the low-confidence paragraph data unit, determine to generate a first adjustment strategy to reduce the weight ratio of the low-confidence paragraph data unit in the display template; If the current paragraph data unit is the high-confidence paragraph data unit, determine to maintain the weight ratio of the current paragraph data unit in the display template.

4. The intelligent structured display method for personnel files based on a large language model according to claim 3, characterized in that, The process of executing the first adjustment strategy based on the large language model includes: Reduce the correlation strength between several of the high-confidence paragraph data units and the current low-confidence paragraph data units in the display template.

5. The intelligent structured display method for personnel files based on a large language model according to claim 1, characterized in that, The process of determining whether the current paragraph data unit meets the requirements includes: The semantic matching degree is determined by the large language model based on a preset element dictionary; If the semantic matching degree is less than the preset semantic matching degree, it is determined that the current paragraph data unit does not meet the requirements and it is determined whether to execute the completion strategy. If the semantic matching degree is greater than or equal to the preset semantic matching degree, it is determined that the current paragraph data unit meets the requirements; The preset element dictionary includes a set of necessary intrinsic elements corresponding to different types of paragraph data units.

6. The intelligent structured display method for personnel files based on a large language model according to claim 5, characterized in that, The process of determining whether to execute the completion strategy includes: If the semantic matching degree difference is greater than the preset semantic matching degree difference, it is determined that the current paragraph data unit will not execute the completion strategy, and a second adjustment strategy is generated to reduce the weight ratio of the paragraph data unit that does not meet the requirements in the structured display template. If the semantic matching degree difference is less than or equal to the preset semantic matching degree difference, the missing content corresponding to the current paragraph data unit is determined and the completion strategy is executed.

7. The intelligent structured display method for personnel files based on a large language model according to claim 6, characterized in that, The process of executing the completion strategy includes: Based on a large language model, prompt words are designed to generate complete content corresponding to the missing content. The completion strength is determined by adjusting the generation parameters of the large language model. The generation parameters include temperature parameter values ​​or kernel sampling values. The temperature parameter values ​​are negatively correlated with the semantic matching degree difference, and the kernel sampling values ​​are positively correlated with the semantic matching degree difference.

8. The intelligent structured display method for personnel files based on a large language model according to claim 6, characterized in that, The second adjustment strategy includes reducing the proportion of information displayed in the structured display template for paragraph data units that do not meet the requirements.

9. The intelligent structured display method for personnel files based on a large language model according to claim 1, characterized in that, The process of determining whether the display template is qualified includes: The model confidence level is determined by the large language model based on target requirements and a pre-set external authoritative database. Based on the comparison results where the model confidence score is less than the model confidence threshold, the display template is determined to be unqualified, and a third adjustment strategy is generated to optimize the display template based on the large language model. The display template is deemed qualified based on the comparison result where the model confidence level is greater than or equal to the model confidence level threshold, and the structured display template is generated based on the model confidence level and the display template.

10. The intelligent structured display method for personnel files based on a large language model according to claim 9, characterized in that, The third adjustment strategy includes adjusting the arrangement order and visual hierarchy of the displayed information in the display template; Alternatively, reallocate semantic blocks.

Images