A personnel file management system based on artificial intelligence

Abstract

The application discloses a personnel file management system based on artificial intelligence, and relates to the technical field of digital management of personnel files.The system comprises an image sequence acquisition module, an information extraction module, a semantic verification module and a file construction module.The image sequence acquisition module generates a standardized continuous scanning image sequence.The information extraction module performs cross-page semantic flow alignment and synchronously extracts structured information, and constructs a personnel file semantic network with time sequence and logical relationship labels.The semantic verification module adopts a graph neural network with joint constraints of time sequence feature maps and logical rules to perform end-to-end verification, and generates verification confidence of each cross-page semantic unit.The file construction module filters and intelligently completes the semantic unit according to the confidence, and finally constructs a structured digital personnel file embedded with an anti-tampering archiving watermark.The application completely retains cross-page semantic association, and significantly improves the efficiency, information integrity and data quality of digital personnel files.

Description

Technical Field

[0001] This invention relates to the field of digital personnel file management technology, and in particular to a personnel file management system based on artificial intelligence. Background Technology

[0002] The digitization of personnel files is a core component of human resource management informatization. Current technologies generally employ a single-page scanning and page-by-page information processing approach. First, each page of the paper file is scanned independently to generate discrete images. Then, optical character recognition (OCR) technology is used to extract the text content from each page. Finally, based on pre-defined field templates, the text on each page is structured, parsed, and categorized for storage, thus completing the conversion from paper to digital files. In practical applications, personnel files are often multi-page bound documents, and their content naturally exhibits cross-page continuity. For example, the timeline of an individual's work history may span multiple pages, the main text of appointment and dismissal documents and approval opinions may be on different pages, and the content of academic certificates and related attachments may have cross-page connections.

[0003] The existing technology of processing each page independently naturally severs the semantic connections and logical continuity between content across pages. During information extraction, it fails to fully identify the semantic units distributed across pages, resulting in fragmented structured information. This fragmented information cannot directly form a complete archival logical system, requiring manual analysis of the contextual relationships between pages and secondary integration. This not only significantly increases the manual cost and time required for digitization but also makes it highly susceptible to information omissions and logical misalignments due to oversights during manual verification, ultimately hindering the overall efficiency and data quality of personnel file digitization. Summary of the Invention

[0004] This invention provides an artificial intelligence-based personnel file management system to solve the problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides an artificial intelligence-based personnel file management system, characterized in that it includes: The image sequence acquisition module is used to acquire a continuous scan image sequence of the entire personnel file; The information extraction module is used to perform cross-page semantic flow alignment on the continuous scanned image sequence and simultaneously extract structured information during the alignment process to generate a personnel file semantic network with time sequence labels and logical relationship labels. The semantic verification module is used to perform end-to-end verification of the personnel file semantic network using a graph neural network constrained by temporal feature graphs and logical rules, and to generate the verification confidence of each cross-page semantic unit. The file construction module is used to filter cross-page semantic units based on the verification confidence level, and to construct digital personnel files using the filtered cross-page semantic units.

[0006] Preferably, the image sequence acquisition module is specifically used for: Extract unique identifier features and boundary semantic features for each image. The unique identifier features are used to distinguish images from different files, and the boundary semantic features are used to characterize the content continuity of image edges. Based on the aforementioned boundary semantic features, a temporal association between adjacent images is established to generate an initial image sequence; The integrity and uniqueness of the initial image sequence are verified based on the unique identifier feature, and duplicate images and invalid images that do not belong to this file are removed. Generate a sequence of consecutive scanned images with physical archive mapping markers.

[0007] Preferably, the information extraction module is specifically used for: Global semantic features are extracted from a sequence of continuously scanned images to obtain a global semantic feature set; Based on the global semantic feature set and boundary semantic features, establish global semantic associations for cross-page images; Semantic units are bound to the structured information extracted synchronously to obtain a set of structured information corresponding to cross-page semantic associations. During the binding process, semantic node types are assigned to archive elements and subordinate relationship edges are established to generate semantic sub-networks with element types. The structured information set is composed of the semantic sub-network.

[0008] Preferably, when the information extraction module generates a semantic network of personnel files with time sequence labels and logical relationship labels, it is specifically used for: Feature clustering is performed on entity nodes in the semantic network of personnel files to obtain entity clustering results; Based on the entity clustering results, multiple nodes describing the same entity are merged into a unified entity node; By integrating the associated edges of unified entity nodes, a semantic network of personnel files is obtained.

[0009] Preferably, when the semantic verification module performs end-to-end verification using a graph neural network that employs joint constraints of temporal feature graphs and logical rules, it is specifically used for: Extract the time-series labels and associated events of each entity node in the semantic network of personnel files to obtain the entity time-series event set; Based on the entity time series event set, construct the entity-level time series feature map corresponding to each entity; The entity-level temporal feature map is input into the graph neural network to obtain the entity-level temporal verification results.

[0010] Preferably, the semantic verification module is further configured to: Extract the file type tags from the semantic network of personnel files to obtain the file type identifier; Based on the file type identifier, load the corresponding set of exclusive logical rules; The logical relationships between elements defined by the exclusive set of logical rules are used as constraints for message passing or node updates in the graph neural network to obtain a graph neural network constrained by the logical rules.

[0011] Preferably, when the semantic verification module generates the verification confidence score for each cross-page semantic unit, it is specifically used for: A threshold judgment is made on the verification confidence of each cross-page semantic unit to obtain a set of low-confidence cross-page semantic units; Traverse the nodes and edges of the semantic subnetwork corresponding to the low-confidence cross-page semantic unit to obtain the set of contradictory nodes and the set of contradictory edges; The contradictory node set and contradictory edge set are marked to the corresponding cross-page semantic unit to obtain the verification result with contradictory marking.

[0012] Preferably, when the archive building module performs the filtering of cross-page semantic units based on the verification confidence level, it is specifically used for: Semantic context extraction is performed on low-confidence cross-page semantic units to obtain a semantic context feature set; Based on the semantic context feature set and the personnel file semantic network, a semantic completion candidate set is generated; The semantic completion candidate set is associated with the filtered high-confidence cross-page semantic units and stored together to obtain the associated cross-page semantic unit set.

[0013] Preferably, when the file construction module constructs digital personnel files using the filtered cross-page semantic units, it is specifically used for: Extract the logical relationship tags of the filtered cross-page semantic units to obtain a logical relationship graph; Based on the logical relationship graph, the hierarchical structure of digital personnel files is divided; Based on the hierarchical structure and logical relationships, the content order of cross-page semantic units is arranged to obtain a structured digital personnel file.

[0014] Preferably, the archive construction module is further configured to: When generating structured digital personnel files, the unique identification information of the personnel in the file and the current timestamp are extracted to generate an anti-tampering archiving watermark, and the archiving watermark is embedded in the metadata or file image layer of the digital personnel file. Attached Figure Description

[0015] Figure 1 This invention provides a system architecture diagram for an artificial intelligence-based personnel file management system. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0017] refer to Figure 1 As shown, an artificial intelligence-based personnel file management system is characterized by comprising: Image sequence acquisition module 100 is used to acquire a continuous scan image sequence of the entire personnel file; The information extraction module 200 is used to perform cross-page semantic flow alignment on the continuous scanned image sequence and simultaneously extract structured information during the alignment process to generate a personnel file semantic network with time sequence labels and logical relationship labels. The semantic verification module 300 is used to perform end-to-end verification of the personnel file semantic network using a graph neural network constrained by temporal feature graphs and logical rules, and to generate the verification confidence of each cross-page semantic unit. The file construction module 400 is used to filter cross-page semantic units according to the verification confidence level, and to construct digital personnel files using the filtered cross-page semantic units.

[0018] In practice, the system first completes the pre-standardization processing of the original archive scan images through the image sequence acquisition module 100.

[0019] Generally, the image sequence acquisition module 100 imports the original scanned images of the entire personnel file in batches, and synchronously collects two-dimensional core features for each image frame by frame: unique identification features and boundary semantic features. Specifically, the unique identification features are generated by fusing the file's unique number, page number QR code, key personnel feature strings, and image hash value, and are used to accurately distinguish different file subjects.

[0020] Boundary semantic features specifically capture the edge areas around the image, collecting edge fragments, cross-page table lines, and paragraph ending marks, providing core evidence for subsequent judgment of page content continuity.

[0021] After completing the two-dimensional feature acquisition of all images, the image sequence acquisition module 100 performs similarity calculation on the boundary semantic features of all adjacent images and outputs the edge content correlation value in the range of 0 to 1.

[0022] In this embodiment, the judgment threshold can be obtained through batch archive sample traversal experiments, or it can be assigned a value based on the experience of archive digitization engineering. In actual implementation, the engineering usage requirements for automatic matching of image sequences can be preset, multiple sets of parameter values ​​with different gradients can be traversed, and the value that can meet the preset usage requirements can be selected as the final judgment threshold.

[0023] For example, comparative tests were conducted using gradient thresholds of 0.7, 0.75, and 0.8 in sequence. When the value was 0.75, it met the engineering requirements for cross-page image matching in this system, so 0.75 was taken as the critical judgment threshold.

[0024] After determining the judgment threshold, the image sequence acquisition module 100, based on the calculated edge content correlation degree, sequentially connects all adjacent images with a correlation degree higher than the set threshold according to the original paper archive browsing logic to generate a preliminary continuous image sequence.

[0025] As a preferred implementation, if the work resume form or text paragraphs continue from the end of one page of the file to the beginning of the next page, the image sequence acquisition module 100 can identify the cross-page connection relationship through accurate edge feature matching, thus avoiding the discontinuity of continuous file content.

[0026] After the initial image sequence arrangement is completed, the image sequence acquisition module 100 starts the image sequence denoising and calibration operation. It compares all images one by one through global hashing, screens and deletes blank images, duplicate images, and invalid images mixed across files in batches. At the same time, it matches the system's preset file page number range and standard archive page number, detects and marks missing pages in the sequence, and completes the denoising and purification of all images.

[0027] Specifically, each compliant image after purification is individually bound by the image sequence acquisition module 100 with physical mapping tags such as file number, personnel name, archive batch, and physical page number, ultimately generating a standardized continuous image sequence without omissions, errors, or invalid images, providing a regular and reliable data source for subsequent semantic extraction and semantic parsing operations.

[0028] After the image sequence acquisition module 100 outputs a standardized continuous image sequence, the system then performs global semantic processing and structured modeling through the information extraction module 200.

[0029] Generally, the information extraction module 200 can call a dedicated visual semantic model that has been trained in advance using a large number of personnel file samples. It reads all file elements such as text, tables, seals, resumes, awards and punishments, and professional titles contained in continuous images frame by frame, and converts the file content of all images into multi-dimensional semantic feature vectors. These vectors are then integrated and aggregated into a global semantic feature set covering the entire file, thus completely preserving all the semantic information of the entire file.

[0030] Based on this, the information extraction module 200 integrates global semantic features with previously extracted image boundary semantic features, compares the semantic relationships of adjacent pages and non-adjacent pages with content associations in groups, accurately identifies cross-page connecting text, split tables, and continuous resume entries, breaks through the semantic isolation problem brought about by traditional single-frame image recognition, and completes the semantic connection of the same business event, the same archive field, and the same resume entry scattered on different pages, forming a complete and continuous cross-page semantic flow.

[0031] After achieving full semantic flow connectivity, the information extraction module 200 strictly follows the personnel file archiving standards to extract standardized archiving fields such as basic personnel information, academic qualifications, work experience, reward and punishment records, and professional titles and ranks from the complete semantic flow in batches. Each set of extracted structured fields is precisely bound to the corresponding original semantic unit and original image location to form a structured archive dataset that can be traced throughout the entire process.

[0032] It should be noted that the information extraction module 200 then classifies all semantic units into types, generating multiple types of entity nodes such as basic information nodes, time nodes, duty performance event nodes, qualification nodes, and reward and punishment nodes. At the same time, based on the inherent business logic of personnel files, it builds subordinate association edges between different nodes, and initially constructs the semantic sub-network of the files.

[0033] For fragmented information of the same type scattered across multiple pages of files, the information extraction module 200 integrates text features, semantic attribute features, and temporal correlation features to carry out multi-dimensional clustering and merging. It integrates scattered nodes corresponding to the same person, the same business event, and the same qualification information into a unique standardized entity node. At the same time, it performs deduplication and logical correction on the correlation edges associated with entity nodes, and uniformly configures temporal labels and logical relationship labels for all semantic nodes and correlation edges. Finally, it generates a complete, computable, and logically self-consistent global semantic network of personnel files.

[0034] In this embodiment, information on professional title evaluation and qualification acquisition recorded on multiple pages can be clustered into the same qualification entity, thereby achieving the standardization and deduplication of fragmented information.

[0035] After the information extraction module 200 completes the semantic network construction, the system switches to the semantic verification module 300 to start the full-domain intelligent verification operation.

[0036] Specifically, the semantic verification module 300 first traverses all entity nodes within the semantic network, collects all business events and bound time tags corresponding to each entity, sorts out the arrangement logic and interval relationship of each event according to the chronological order, constructs a time sequence feature map exclusive to each file entity, and fully records the timeline of events throughout the entire life cycle of personnel education, job performance, rewards and punishments, assessments, etc.

[0037] For example, for personnel resume entities, the semantic verification module 300 can uniformly incorporate time-series events such as onboarding, job assignment, promotion, and job transfer into the same time-series graph, providing a basis for subsequent time-series compliance verification.

[0038] Subsequently, the semantic verification module 300 automatically identifies the category of the currently processed file, accurately matches the system's pre-stored general mandatory constraint rules with category-specific differentiated rules, and embeds hard business logic such as the time of obtaining academic qualifications, the time of employment, the time of reward and punishment, and the business relationship of the position into the node update and reasoning process of the graph neural network. This ensures that every semantic judgment is subject to the rigid constraints of manually preset rules, avoiding unfounded autonomous reasoning.

[0039] After the rules are loaded, the semantic verification module 300 starts the graph neural network parameter configuration and verification operation.

[0040] In a preferred embodiment, the parameters of the two-layer graph convolutional network architecture can be determined through sample traversal experiments or engineering experience assignment. In actual implementation, engineering usage requirements such as feature carrying capacity, computing power overhead, and iterative convergence stability can be preset, multiple sets of gradient parameter configurations can be traversed, and the parameter combination that can meet the preset usage requirements can be selected as the final configuration.

[0041] For example, by testing multiple combinations of feature input dimensions, hidden layer dimensions, iteration rounds, initial learning rate, and regularization dropout rate, if the input dimension is 256, the hidden layer dimension is 128, the total number of iteration rounds is 200, the initial learning rate is 0.001, and the dropout rate is 0.2, then this parameter configuration can meet the operational requirements of the system's document semantic verification. In this case, the parameters are configured, and the Adam optimizer is used to perform iterative updates to ensure that the steps for each document verification are consistent and the results are reproducible.

[0042] After the parameters are configured, the semantic verification module 300 combines the temporal feature map with business rule constraints to carry out full-domain verification. It simultaneously integrates the temporal matching score and the logical compliance score to complete the weighted calculation and quantifies the confidence value of each cross-page semantic unit. All confidence values ​​are uniformly classified into the value range of 0 to 1.

[0043] It should be noted that the confidence threshold can be obtained through batch sample experimentation or assigned based on archival quality inspection experience. In actual implementation, the requirements for archival quality inspection accuracy, anomaly judgment standards, and other engineering needs can be preset, multiple sets of gradient parameters can be iterated, and the value that can meet the preset requirements can be selected as the final judgment threshold.

[0044] For example, sample tests were conducted using gradient thresholds of 0.7, 0.75, 0.8, 0.85, and 0.9 in sequence. When the value was 0.8, it met the engineering requirements for the quality inspection of the archives in this system. Therefore, 0.8 was selected as the critical threshold.

[0045] Subsequently, the semantic verification module 300 completes the anomaly determination based on the threshold, uniformly determining semantic units with a confidence level below 0.8 as abnormal units, traversing the semantic sub-networks corresponding to the abnormal units one by one, accurately locating contradictory nodes and related edges with temporal conflicts, logical contradictions, missing information, and content mismatches, accurately marking the abnormal location and abnormal type, and completing the anomaly tracing of the semantics of the entire archive.

[0046] For example, if the file contains contradictory information such as "obtained a bachelor's degree in 2019 and was employed in a professional and technical position in 2018", the semantic verification module 300 can proactively identify the conflict, reduce the confidence of the corresponding unit, and accurately complete the anomaly marking through preset rules.

[0047] Furthermore, after the semantic verification module 300 completes the full-domain anomaly marking and source tracing, the system's archive construction module 400 undertakes the subsequent standardization, correction, and archiving operations.

[0048] Specifically, the archive construction module 400 first performs hierarchical classification and screening based on the confidence scores of semantic units. It retains high-confidence compliant semantic units with a confidence score of not less than 0.8 throughout the process, directly removes invalid semantic units with fundamental errors that cannot be corrected by context, and retains low-confidence semantic units that need to be corrected separately for subsequent optimization processing.

[0049] For all retained low-confidence anomalous semantic units, the file construction module 400 comprehensively extracts the high-confidence contextual semantics, global correlation features, and preset business constraint rules surrounding the unit, constructs a local semantic reference set covering the anomalous unit, and intelligently infers multiple sets of compliance correction content based on the complete semantic network and reference set information, builds a semantic completion candidate set, and binds and merges the corrected compliance content with the original high-confidence semantic units to complete the intelligent correction of anomalous content.

[0050] In this embodiment, if there is a missing job description at the end of a resume page, the file construction module 400 can supplement the compliant content that fits the semantics of the original file by relying on the job description features of similar positions on previous and subsequent pages and industry standards.

[0051] After completing the semantic content correction and integration across the entire domain, the archive construction module 400 extracts the logical relationship tags carried by all compliant semantic units and integrates them to generate a global logical relationship graph covering all business relationships in the entire archive. Based on this graph and combined with the current national personnel archive archiving standards, it automatically divides a multi-level standardized archive hierarchy structure. With basic personnel identity information as the top core, it sequentially categorizes secondary sub-modules such as academic qualifications, work experience, rewards and punishments, and professional titles and ranks. Within each module, the archive items are refined according to the logical relationships of subordination, chronology, parallelism, and causation, thereby achieving standardized hierarchical classification of the entire archive content.

[0052] Generally, the archive construction module 400 then performs unified formatting on all the hierarchical and standardized semantic content, standardizing the layout style, item format and text description, thoroughly correcting the problems of cross-page content fragmentation, disordered item arrangement and mixed and inconsistent formats that exist in traditional digital archives, and generating structured digital personnel archives with clear structure, rigorous logic and complete content.

[0053] After the structured archives are generated, the archive construction module 400 initiates the secure archiving process. It extracts unique identification information such as the ID card number, employee number, and archive number of the personnel corresponding to the archives in real time. Combined with the system's high-precision timestamp, it generates a globally unique anti-tampering archive watermark using the SHA-256 irreversible hash encryption algorithm. The watermark is embedded in both the archive metadata layer and the image display layer. Relying on the dual-layer watermark, it enables back-end traceability auditing and front-end visual anti-counterfeiting verification. At the same time, the irreversible, undeletable, and tamper-proof characteristics of the watermark ensure the security and traceability of the archived archives. Finally, it completes the intelligent archiving process and outputs standardized, accurate, and secure finished digital personnel archives.

[0054] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0055] The embodiments described above can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0056] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A personnel file management system based on artificial intelligence, characterized in that, include: The image sequence acquisition module is used to acquire a continuous scan image sequence of the entire personnel file; The information extraction module is used to perform cross-page semantic flow alignment on the continuous scanned image sequence and simultaneously extract structured information during the alignment process to generate a personnel file semantic network with time sequence labels and logical relationship labels. The semantic verification module is used to perform end-to-end verification of the personnel file semantic network using a graph neural network constrained by temporal feature graphs and logical rules, and to generate the verification confidence of each cross-page semantic unit. The file construction module is used to filter cross-page semantic units based on the verification confidence level, and to construct digital personnel files using the filtered cross-page semantic units.

2. The personnel file management system based on artificial intelligence as described in claim 1, characterized in that, The image sequence acquisition module is specifically used for: Extract unique identifier features and boundary semantic features for each image. The unique identifier features are used to distinguish images from different files, and the boundary semantic features are used to characterize the content continuity of image edges. Based on the aforementioned boundary semantic features, a temporal association between adjacent images is established to generate an initial image sequence; The integrity and uniqueness of the initial image sequence are verified based on the unique identifier feature, and duplicate images and invalid images that do not belong to this file are removed. Generate a sequence of consecutive scanned images with physical archive mapping markers.

3. The personnel file management system based on artificial intelligence as described in claim 2, characterized in that, The information extraction module is specifically used for: Global semantic features are extracted from a sequence of continuously scanned images to obtain a global semantic feature set; Based on the global semantic feature set and boundary semantic features, establish global semantic associations for cross-page images; Semantic units are bound to the structured information extracted synchronously to obtain a set of structured information corresponding to cross-page semantic associations. During the binding process, semantic node types are assigned to archive elements and subordinate relationship edges are established to generate semantic sub-networks with element types. The structured information set is composed of the semantic sub-network.

4. The personnel file management system based on artificial intelligence as described in claim 1, characterized in that, When the information extraction module generates a semantic network of personnel files with time-series labels and logical relationship labels, it is specifically used for: Feature clustering is performed on entity nodes in the semantic network of personnel files to obtain entity clustering results; Based on the entity clustering results, multiple nodes describing the same entity are merged into a unified entity node; By integrating the associated edges of unified entity nodes, a semantic network of personnel files is obtained.

5. The personnel file management system based on artificial intelligence as described in claim 1, characterized in that, When the semantic verification module performs end-to-end verification using a graph neural network that employs joint constraints of temporal feature maps and logical rules, it is specifically used for: Extract the time-series labels and associated events of each entity node in the semantic network of personnel files to obtain the entity time-series event set; Based on the entity time series event set, construct the entity-level time series feature map corresponding to each entity; The entity-level temporal feature map is input into the graph neural network to obtain the entity-level temporal verification results.

6. The personnel file management system based on artificial intelligence as described in claim 5, characterized in that, The semantic verification module is also used for: Extract the file type tags from the semantic network of personnel files to obtain the file type identifier; Based on the file type identifier, load the corresponding set of exclusive logical rules; The logical relationships between elements defined by the exclusive set of logical rules are used as constraints for message passing or node updates in the graph neural network to obtain a graph neural network constrained by the logical rules.

7. The personnel file management system based on artificial intelligence as described in claim 4, characterized in that, The semantic verification module, when generating the verification confidence score for each cross-page semantic unit, is specifically used for: A threshold judgment is made on the verification confidence of each cross-page semantic unit to obtain a set of low-confidence cross-page semantic units; Traverse the nodes and edges of the semantic subnetwork corresponding to the low-confidence cross-page semantic unit to obtain the set of contradictory nodes and the set of contradictory edges; The contradictory node set and contradictory edge set are marked to the corresponding cross-page semantic unit to obtain the verification result with contradictory marking.

8. The personnel file management system based on artificial intelligence as described in claim 8, characterized in that, When the archive construction module performs the filtering of cross-page semantic units based on the verification confidence level, it is specifically used for: Semantic context extraction is performed on low-confidence cross-page semantic units to obtain a semantic context feature set; Based on the semantic context feature set and the personnel file semantic network, a semantic completion candidate set is generated; The semantic completion candidate set is associated with the filtered high-confidence cross-page semantic units and stored together to obtain the associated cross-page semantic unit set.

9. A personnel file management system based on artificial intelligence as described in claim 1, characterized in that, When the file construction module constructs digital personnel files using filtered cross-page semantic units, it is specifically used for: Extract the logical relationship tags of the filtered cross-page semantic units to obtain a logical relationship graph; Based on the logical relationship graph, the hierarchical structure of digital personnel files is divided; Based on the hierarchical structure and logical relationships, the content order of cross-page semantic units is arranged to obtain a structured digital personnel file.

10. A personnel file management system based on artificial intelligence as described in claim 1, characterized in that, The file construction module is also used for: When generating structured digital personnel files, the unique identification information of the personnel in the file and the current timestamp are extracted to generate an anti-tampering archiving watermark, and the archiving watermark is embedded in the metadata or file image layer of the digital personnel file.

Images