A pathogenic microorganism laboratory registration and approval management system

By constructing a pathogenic microorganism laboratory registration and approval management system, and utilizing standard field syntax graphs and document evidence graphs, combined with candidate matching and adaptive update algorithms, the problems of field omission and mismatch in the pathogenic microorganism laboratory registration system were solved, achieving accuracy in field attribution and stability in identification results.

CN123118032APending Publication Date: 2026-11-17BEIJING HONGCHENG INNOVATION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611484088.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-22
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

The existing pathogenic microorganism laboratory registration system suffers from problems such as field omission, mismatch, and incorrect attribution when processing application documents of different formats. Furthermore, it lacks a stable mechanism for updating identification results, which affects the effectiveness of subsequent processing.

Method used

A pathogenic microorganism laboratory registration and approval management system was constructed. Through field mapping module, document mapping module, bidirectional attribution module, consistency module, and data processing module, standard field syntax graph and document evidence graph were used, combined with candidate matching algorithm, propagation algorithm, and adaptive update algorithm to achieve accurate field attribution and stable identification results.

Benefits of technology

It improves the accuracy of field attribution, reduces the probability of misattribution and omission of synonymous fields, reduces the impact of changes in application templates on backend approval processing, and ensures the stability and reliability of recognition results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123118032A_ABST
    Figure CN123118032A_ABST
Patent Text Reader

Abstract

This invention relates to the field of computer data processing technology, specifically to a pathogenic microorganism laboratory registration and approval management system. The system includes a field mapping module, a document mapping module, a two-way attribution module, a consistency module, a field determination module, a data processing module, and a registration and approval module. This invention constructs standard field syntax graphs and document evidence graphs, combining candidate screening, two-way propagation, counterfactual substitution, posterior fusion, and unified sorting. This ensures that field attribution is constrained by semantics, type, location, context, duplicate information, business role, and field cardinality, distinguishing near-synonymous fields across multiple roles, reducing mismatches and missed identifications. Based on adaptive updates of candidate concentration, discriminability, evidence consistency, counterfactual stability, and data source reliability, it suppresses the accumulation of erroneous identifications across documents and forms standardized application data traceable to the original page number, cell, or text block, supporting rule matching and anomaly reverse lookup.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of computer data processing technology, specifically to a pathogenic microorganism laboratory registration and approval management system. Background Technology

[0002] Computer data processing technology is typically used to collect, parse, identify, transform, associate, store, and output data from different sources and in different formats. Through text processing, document structure analysis, data type identification, field mapping, and data standardization, it transforms scattered or unstructured information in text documents, spreadsheets, formatted documents, and image documents into computer-processable data with a unified data structure, so as to facilitate subsequent information retrieval, data verification, business rule matching, and information management system calls.

[0003] As the registration of pathogenic microorganism laboratories gradually adopts electronic submission, the Word, Excel, PDF, and image files submitted by different applicants vary in terms of field names, arrangement order, page position, table structure, and chapter structure. Existing technologies typically extract application fields through optical character recognition, document structure parsing, regular expressions, field name similarity, or classification models. However, these methods primarily determine field attribution based on the text, position, local structure, or preset rules within the current document, failing to perform a comprehensive verification of the order, co-occurrence, region, role, mutual exclusion, and cardinality relationships between standard fields across the entire document. When identical or similar field expressions such as "responsible person" and "responsible personnel" correspond to different business roles such as laboratory head and biosafety head, it is easy to cause field omissions, mismatched field values, and incorrect attribution of fields with similar roles. Furthermore, existing systems lack a mechanism to adjust the frequency of field knowledge updates based on the stability of the recognition results, and low-reliability recognition results may have a lasting impact on the processing of subsequent application documents. Therefore, there is an urgent need to construct a pathogenic microorganism laboratory registration approval and management system. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a pathogenic microorganism laboratory registration and approval management system.

[0005] To achieve the above objectives, the present invention provides the following technical solution: This invention provides a pathogenic microorganism laboratory registration and approval management system, the system comprising: The field graphing module is used to obtain standard declaration fields and historical declaration data, calculate field association parameters of the relationship between standard declaration fields and mapping parameters between the original field expressions and standard declaration fields based on the historical declaration data, and construct a standard field syntax graph based on the standard declaration fields and field association parameters. The document graphing module is used to acquire the filing application documents to be processed, then perform structural parsing on the filing application documents to form an information unit sequence, and calculate the document association parameters of each information unit based on the information unit sequence, and construct a document evidence graph based on the parameters; The bidirectional attribution module is used to select target information units sequentially in the document evidence graph, and use a candidate matching algorithm combined with mapping parameters to filter multiple candidate standard fields. Based on the document evidence graph and the standard field syntax graph, the forward and backward propagation algorithms are used sequentially to calculate the bidirectional attribution parameters of each candidate standard field. The consistency module is used to take each candidate standard field as the hypothetical attribution field of the target information unit, and to use the counterfactual consistency algorithm to calculate the relational consistency parameter of each hypothetical attribution field. The field determination module is used to fuse the bidirectional attribution parameters and relation consistency parameters using a posterior fusion algorithm to obtain the final attribution parameters of each candidate standard field. Then, the candidate standard fields are sorted according to the final attribution parameters, and the candidate standard field with the first position in the sort is selected as the standard declaration field of the target information unit. The data processing module is used to associate the field values ​​of each information unit in the information unit sequence with the standard declaration fields selected by the field determination module for that information unit, and to collect them to form a standardized declaration dataset for the filing declaration documents. It also uses an adaptive update algorithm to adjust the field association parameters and mapping parameters according to the final attribution parameters of each information unit, thereby forming an updated standard field syntax diagram and field mapping parameter set.

[0006] Furthermore, the field association parameters include sequential association parameters, co-occurrence association parameters, region association parameters, role association parameters, mutual exclusion association parameters, and cardinality association parameters; The field graphing module calculates the field association parameters based on the arrangement status, co-occurrence status, regional affiliation status, role status, mutual exclusion status, and field quantity status of the standard declaration fields in the historical declaration data. It then constructs a standard field syntax graph using the standard declaration fields as nodes and the field association parameters as edge parameters connecting the standard declaration field nodes.

[0007] Furthermore, the document association parameters include sequential association parameters, structural association parameters, regional association parameters, and cross-location association parameters; The document graphing module calculates the document association parameters based on the reading order, document structure, regional affiliation, and duplicate information of each information unit in the filing application document, and constructs the document evidence graph using each information unit as a node and the document association parameters as edge parameters connecting the information unit nodes.

[0008] Furthermore, the bidirectional home module includes: The candidate filtering unit is used to calculate the local matching parameters between the target information unit and each standard declaration field based on the field semantic features, field value features, document structure features and mapping parameters of the target information unit, and to filter multiple candidate standard fields according to the local matching parameters. The forward propagation unit is used to calculate the forward attribution parameters of each candidate standard field based on the preceding information unit of the target information unit, document association parameters, and field association parameters. The backward propagation unit is used to modify the forward attribution parameters based on the subsequent information units of the target information unit, document association parameters, and field association parameters, using the forward attribution parameters as initial parameters, to form bidirectional attribution parameters for each candidate standard field.

[0009] Furthermore, the consistency module includes: The hypothetical mapping unit is used to sequentially establish hypothetical mapping relationships between each candidate standard field and the target information unit. The deviation calculation unit is used to calculate the relationship deviation parameter between the association relationship in the document evidence diagram and the field association relationship in the standard field syntax diagram, based on each hypothetical mapping relationship; The parameter generation unit is used to generate relation consistency parameters for each assumed attribution field based on the relation deviation parameters.

[0010] Furthermore, the field determination module includes: The parameter transformation unit is used to normalize the relationship consistency parameters of each hypothesized attribution field to form a consistency probability. The fusion computing unit is used to fuse the bidirectional attribution parameters and consistency probabilities of each candidate standard field to form the final attribution parameters; The field sorting unit is used to sort the candidate standard fields in descending order according to the final attribution parameter, and to take the candidate standard field with the first position in the sort as the standard declaration field of the target information unit.

[0011] Furthermore, the relational deviation parameters include field sequence deviation parameters, document structure deviation parameters, region deviation parameters, duplicate information deviation parameters, role deviation parameters, and field cardinality deviation parameters; The deviation calculation unit calculates the deviation parameters of each relationship respectively, and merges the deviation parameters of each relationship to obtain the comprehensive deviation parameter of the assumed attribution field. The parameter generation unit generates the relationship consistency parameter based on the comprehensive deviation parameter.

[0012] Furthermore, the adaptive update algorithm includes: Calculate candidate distribution parameters and candidate differentiation parameters based on the final attribution parameters of each candidate standard field, calculate bidirectional consistency parameters based on the forward attribution parameters and bidirectional attribution parameters, obtain hypothetical stability parameters based on the relationship consistency parameters of each hypothetical attribution field, and obtain source reliability parameters based on the data source of the filing application documents. The candidate distribution parameters, candidate differentiation parameters, bidirectional consistency parameters, assumed stability parameters, and source reliability parameters are integrated to form update adjustment parameters, and the update magnitude of field association parameters and mapping parameters is adjusted according to the update adjustment parameters.

[0013] Furthermore, the filing application documents include one or more of the following: text documents, spreadsheets, formatted documents, and image documents; The document mapping module extracts the text content, table structure, page position and field value information from the filing application document, and converts the extracted results into information units with a unified data structure according to the document reading order, forming an information unit sequence; The standardized declaration dataset includes standard declaration fields, field values, original field expressions, original file locations, and final attribution parameters.

[0014] Furthermore, the system also includes a filing and approval module; The filing and approval module is used to match the standardized application dataset with preset filing rules, generate filing and approval records based on the matching results, and extract standardized application data that does not meet the preset filing rules to form anomalies. The data processing module is also used to associate and store the filing approval records and anomalies with the standardized application dataset, and then use the updated standard field syntax diagram and field mapping parameter set for the processing of the next filing application document.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention constructs a standard field syntax graph and a document evidence graph to represent the theoretical relationships between standard declaration fields in the filing process and the actual relationships between information units in the current declaration document, respectively. It then sequentially performs candidate screening, forward propagation, backward correction, counterfactual substitution, posterior fusion, and unified sorting. This ensures that field attribution is simultaneously constrained by field semantics, data type, regional location, preceding and following field relationships, cross-location duplicate information, business roles, and field cardinality. For original field expressions such as "responsible personnel," which may correspond to multiple personnel roles, the system can determine field attribution based on the impact of each candidate standard field on the consistency of relationships throughout the entire declaration document, reducing the probability of incorrect attribution of synonymous fields, mismatched field values, and missed field identification. Furthermore, all information units undergo the same processing chain, enabling the field identification results to be traced step-by-step based on local matching parameters, forward attribution parameters, bidirectional attribution parameters, relationship consistency parameters, and final attribution parameters.

[0016] By employing a continuous adaptive update method, considering factors such as the concentration of candidate distribution, the differentiation between the first two candidates, the consistency of evidence before and after, the stability of counterfactual evidence, and the reliability of data sources, the update magnitude of field mapping parameters and field association parameters is controlled. This ensures that stable identification results generate relatively large, constrained updates to field knowledge, while unstable identification results generate only small updates, thereby reducing the accumulation and self-reinforcement of misidentifications during cross-document learning. By forming a standardized application dataset containing standard application fields, field values, original field expressions, original document locations, and calculation parameters at each stage, the filing and approval module can match unified fields with preset filing rules and can reverse locate page numbers, cells, or text blocks in the original document from approval records or anomalies, thus reducing the impact of application template changes on backend approval processing. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the construction process of the multidimensional task dependency graph of the present invention. Detailed Implementation

[0018] 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1This invention provides a pathogenic microorganism laboratory registration and approval management system. The system is deployed on a computer server, which includes a processor, a memory, a file interface, and a database. The memory stores computer programs executed by the processor. When the processor executes the computer programs, it implements a field mapping module, a document mapping module, a two-way attribution module, a consistency module, a field judgment module, a data processing module, and a registration and approval module. The file interface is used to receive one or more registration application documents, including text documents, spreadsheets, formatted documents, and image documents. The database is used to store field knowledge status, application processing data, and registration and approval results.

[0020] The system sequentially executes the following steps: standard field syntax graph construction, filing application document parsing, document evidence graph construction, candidate screening, forward propagation, backward propagation, counterfactual consistency calculation, posterior fusion, field determination, standardized output, adaptive update, filing approval, and knowledge status submission. All information units to be identified undergo the same field identification steps, without switching identification paths based on local matching results, conflict states, or confidence levels. In this embodiment, , , and For standard declaration field indexing, , and For information unit indexing, the fusion weights of the same group are non-negative and sum to 1. Sample weights and update adjustment parameters are calculated according to their definitions. Candidate scores and states are calculated only in the corresponding candidate sets. The bidirectional attribution parameters, final attribution parameters, and soft mapping probabilities outside the candidate sets are set to 0. The system specifically includes: The field graphing module is used to obtain standard declaration fields and historical declaration data, calculate field association parameters of the relationship between standard declaration fields and mapping parameters between the original field expressions and standard declaration fields based on the historical declaration data, and construct a standard field syntax graph based on the standard declaration fields and field association parameters. Specifically, the field mapping module establishes a standard declaration field set. ,in, The total number of standard declaration fields is [number]. Each standard declaration field must include at least the following: affiliated institution, laboratory name, laboratory address, laboratory director, biosafety officer, contact number, laboratory biosafety level, name of pathogenic microorganism, category of pathogenic microorganism, type of experimental activity, experimental personnel, personnel qualifications, facilities and equipment, and safety protection conditions. Represented using a field feature model .in, For standard field names, For semantic description, A confirmed collection of historical expressions. For the allowed field value data types, This is the historical regional distribution vector. For the cardinality distribution of the field, For business roles or domain attributes, The largest component in the middle corresponds to the typical semantic region of this field.

[0021] For example, both the Laboratory Head and the Biosafety Head are allowed to have PERSON type field values, but the former is mainly distributed in the basic laboratory information area, while the latter is mainly distributed in the biosafety management information area, corresponding to the laboratory operation management role and the biosafety management role, respectively. The historical expression set of "Laboratory Head" can include "Head," "Laboratory Responsible Person," and "Responsible Person," and the same expression can establish mapping parameters with multiple standard declaration fields, without directly determining its affiliation.

[0022] The field mapping module reads verified historical declaration data and constructs a standard field syntax graph based on sequential association parameters, co-occurrence association parameters, regional association parameters, role association parameters, mutually exclusive association parameters, and cardinality association parameters. , For the standard declaration field node set, The set of edges carries field association parameters. The first five types of association parameters are used as edge parameters between field nodes, and the cardinality association parameters are used as self-connection edge parameters of the corresponding field nodes.

[0023] Historical documents Sample weights This is the product of source reliability and manual verification completeness. Source reliability is calculated using the source verification method described later. Manual verification completeness is the ratio of the number of verified information units to the total number of information units to be verified in the file; both are within the range [0, 1]. For a certain type of historical relationship event, the weighted occurrence count is the sum of the sample weights of each file multiplied by the event occurrence indicator value; the weighted exposure count is the sum of the sample weights of each file multiplied by the verifiable indicator value of the event. The relationship probability is calculated by dividing the weighted occurrence count plus the product of the prior strength and the prior probability by the weighted exposure count plus the prior strength. The prior strength is greater than 0, and the prior probability is within the range [0, 1]. When calculating the relationship strength, the actual relationship strength within the range [0, 1] is used instead of the occurrence indicator value, and the denominator still uses the relationship exposure count.

[0024] The sequential association parameters each store the first-order probability. Second-order probability and third-order probability This corresponds to the probability that the target field appears and forms a valid sequential relationship after one, two, or three preceding standard declaration fields are given. The co-occurrence association parameter indicates the degree to which two fields co-occur; the region association parameter indicates the consistency of the region paths of two fields; the role association parameter indicates the compatibility of personnel role combinations; the mutual exclusion association parameter indicates the degree to which the same original information should not be simultaneously attributed to two fields; and the cardinality association parameter indicates the probability that a standard declaration field has a number of different unique field values ​​in a declaration object.

[0025] Original field expression With standard declaration fields Initial mapping parameters between The numerator is determined according to the weighted frequency with prior knowledge. After confirmation, the mapping was... Weighted number of times The denominator is Weighted occurrence count ,in Each confirmed original field expression is counted as only one standard declaration field, and the sum of the mapping parameters for all standard declaration fields corresponding to the same expression is 1.

[0026] The field association parameter is the relationship probability obtained from historical statistics, the field compatibility parameter is the parameter after unifying the direction and scale, and the relationship channel parameter is the parameter used by the propagation algorithm. The four types of relationship channel parameters are calculated according to the following formula (1), specifically: ; (1) in, , , and These are co-occurrence probability, region association probability, role compatibility probability, and mutual exclusion probability, respectively. , For the same-group fusion weights, the parameters of each channel are located in [-1, 1]. Role compatibility contributes positively, and mutual exclusion contributes negatively. The compatibility parameters of the second-order and third-order order fields are respectively... and As a higher-order attribute of the order relation, the field cardinality distribution does not enter the pairwise propagation channel, but is used in the counterfactual cardinality bias.

[0027] The document graphing module is used to acquire the filing application documents to be processed, then perform structural parsing on the filing application documents to form a sequence of information units, and calculate the document association parameters of each information unit based on the information unit sequence, and construct a document evidence graph based on the parameters; Specifically, the document mapping module acquires the filing application documents to be processed. For text documents, it extracts paragraphs, headings, tables, and merged cell structures; for spreadsheets, it extracts worksheets, cell coordinates, table headers, and merged areas; for text-based layout documents, it extracts text blocks and page coordinates; for scanned layout documents and image documents, it combines optical character recognition and layout analysis to extract text blocks and table structures.

[0028] The parsing results are converted into a sequence of information units according to the reading order. Information unit Represented as ,in, For identification, For the original field expression, For field values, Page number or coordinate. This is the feature vector of the current semantic region. This describes the structural relationship between fields and field values. Number them according to the reading order. For information from the source documents and attachments, This is the data type for field values. Fixed-pattern mobile phone numbers, dates, grade codes, areas, and quantities are identified using regular expressions and domain dictionaries. Personnel names, unit names, and pathogen names are identified using a named entity recognition model. For example, "Zhang San" corresponds to PERSON, and "BSL-2" corresponds to LAB_LEVEL.

[0029] Regional features are encoded using a unified regional category dictionary. and Using the same dimension, It is formed by the category corresponding to the current title path and the page area. The field model is formed by the weighted frequency of occurrence of the field in each region category in the historical files, and after normalization, it is written into the field model.

[0030] A reading scope includes consecutive paragraphs within the same chapter, tables with a unified header, or consecutive lists. Tables are ordered according to row priority or column priority as indicated by the header. Consecutive tables that span multiple pages but maintain the same header belong to the same reading scope. The main text and attachments are established separately for reading scope. When there is a clear citation relationship, cross-positional association is established. Sequential relationships are not directly established based on the adjacency of the documents.

[0031] Document Evidence Image Using information units as nodes and sequence, structure, region, and cross-positional relationships as edges, the document association parameters are calculated according to the following formula (2): ; ; ; (2) in, , , and These represent the same key-value structure, the same table row, the same table column controlled by the same header, and the same paragraph, respectively. A value of 1 is obtained if these conditions are met, and a value of 0 is obtained otherwise. The number of common prefix levels for regional paths. , The number of path levels; Character similarity; Indicates that the data type of the field value is consistent with the value; The degree of correlation between source locations; the two sets of fusion weights are normalized respectively. When The time zone parameter is set to 0; The value is 1 between the main text and explicitly cited attachments, 0.8 between page breaks in the same file, and 0.5 when there is no direct citation but the documents belong to the same application. All document association parameters are located in [0, 1]. Evidence edges store the relationship type, strength, and nodes at both ends.

[0032] The bidirectional attribution module is used to select target information units sequentially in the document evidence graph, and use a candidate matching algorithm combined with mapping parameters to filter multiple candidate standard fields. Based on the document evidence graph and the standard field syntax graph, the forward and backward propagation algorithms are used to calculate the bidirectional attribution parameters of each candidate standard field. Specifically, the bidirectional attribution module includes a candidate selection unit, a forward propagation unit, and a backward propagation unit. The candidate selection unit calculates the target information unit. With standard declaration fields The local matching parameters are calculated using the following formula: (3) in, , , , , These are the semantic, data type, region, structure, and expression matching parameters, all located in [0, 1]. to For group fusion weights; semantic matching is , It is obtained by encoding the original field expression, chapter name, and adjacent field expressions. Obtained through encoding of standard field names and semantic descriptions; region matching is... The cosine value corresponding to the zero vector is conventionally set to 0; it is set to 1 when the data types are the same, and to 1 when they belong to a preset compatible type. If incompatible, the value is 0; structural matching is output by the model with key direction, row / column position, merge status, heading level, and adjacent field types as input; expression matching is the maximum character similarity, maximum semantic similarity, and... in the historical expression set. The weighted sum of the three terms, with non-negative weights and a sum of 1.

[0033] Candidate filtering units by Select a fixed number in descending order The candidate set is formed by these fields. , Based on the validation data, a smaller number of candidates is selected while still meeting the preset candidate omission rate. After the entire candidate set for the current file is generated, forward and backward propagation of the entire sequence is performed.

[0034] set up For candidate distributions that have completed forward computation, The three most recent sequential information units were , , ,but: ; ; ; (4) in, For a set of orders with sufficient preorder context, This is the average weight of the consecutive sequential edges in this context. , , , For order-based fusion weights, when no preceding context exists. If the order is 0, and there are fewer than three preceding nodes, only the available order is merged.

[0035] Includes located in Nodes that previously had structural, regional, or cross-location evidence edges with the node, and nodes with only sequential edges, are not repeatedly added to the aggregation term. The forward parameters are calculated according to the following formula (5), specifically: ; (5) in, , It is a non-negative adjustment parameter. , , The weights are non-negative and sum to 1; no further weighting is required. ; The number of nodes participating in the summation and the number of candidate combinations. When empty Take 0, Proceed in the order of reading.

[0036] Backpropagation is calculated from back to front according to the reading order. Let... For the distribution of the completed forward computation, The next three sequential nodes are , , ,but: ; ; ; (6) Among them, the second order express Third order express All readers should maintain their original reading direction. , and These are the available post-order weights, the renormalized order weights, and the average order edge weights, respectively, when there is no post-order context. Take 0.

[0037] Includes located in Subsequently, for nodes with structural, regional, or cross-location evidence edges, the backward parameters and bidirectional results are calculated according to the following formula (7): ; ; ; (7) in, It is a non-negative adjustment parameter. The number of combinations participating in the summation. When empty Set to 0; , , Weights for fusion within the same group; This represents an unnormalized bidirectional candidate state. The bidirectional attribution parameters are normalized; forward aggregation, backward aggregation, and probability normalization are all implemented using numerically stable log-sum-exp.

[0038] The consistency module is used to take each candidate standard field as the hypothetical attribution field of the target information unit, and to use the counterfactual consistency algorithm to calculate the relational consistency parameters of each hypothetical attribution field. Specifically, the consistency module includes a hypothesis mapping unit, a deviation calculation unit, and a parameter generation unit. The hypothesis mapping unit is configured according to... The candidate list is formed in descending order, and each candidate is then used as... The assumed attribution field, the counterfactual soft mapping is defined as: ; ; (8) in, Traverse all standard declaration fields; the target node uses hard assumptions, while the remaining nodes retain soft mappings. Temporary mappings are revoked after a candidate calculation is completed. The following deviation calculation... Both refer to the current situation. .

[0039] The field sequence evidence set contains Within the scope of reading First-order, second-order, and third-order continuous windows; any adjacent nodes within a window have a sequential edge; continuous paragraphs and tables spanning multiple pages can form a multi-page window; text and attachments without a clear reading continuity do not form a sequential window; structural evidence set. Included with The set of region nodes represents the key-value pairs, nodes in the same row, column, paragraph, list item, and merged cell relationships of the endpoints. Include And nodes that share non-root chapters, parent and child chapters, or table areas with it, weight The value is 1 for chapters at the same bottom level, 0.8 for parent and child chapters, and 0.5 for adjacent child chapters.

[0040] Duplicate information set This includes edges whose field values ​​are identical and of the same type after normalization, and whose explicit references or derivations have been resolved to identify the same entity. Normalization includes whitespace removal, punctuation standardization, alias normalization, and unit conversion. The same name does not necessarily equate to the same role. Concurrent roles are allowed to be stored separately. (Role set) Included with Edges that are endpoints and involve personnel types or role candidates, and satisfy the following conditions: same personnel table, same personnel, common role slot, or competing for the same single-value role, store name, slot and concurrent status, and use the equivalent group of all field values ​​of the current application object for cardinality calculation.

[0041] set up For the process of Ordered window set, For windows, To determine the average edge weights in sequence, then: ; (9) in, , For the corresponding order normalization parameters, For a set of computable orders, Normalize its weights, without a window The value is 0; the first, second, and third probabilities are respectively... , and .

[0042] The remaining pairwise and regional deviations are calculated according to the following formula (10), specifically: ; ; ; (10) in, For border rights, For structure type, The probability of matching the corresponding historical structure; The value is 1 if the identities in the same field are the same or if multiple roles are allowed, otherwise it is 0. Prioritize assigning 0 based on the rule allowing concurrent appointments, then assign 1 for different names competing for the same single-value role, even if... No exceptions, in other cases take When the evidence set is empty, the corresponding deviation is 0.

[0043] set up For the set of equivalent groups of field values, If the union of the current candidate field and fields that have a saved cardinality or mutual exclusion relationship with the candidate field is fixed in all counterfactual calculations at this node, then: ; ; ; (11) in, To predict the number of unique values, Adopted and Using the same truncation method, non-integer exponents are interpolated linearly using adjacent probabilities, while overflow probabilities are used when the exponent exceeds the highest integer interval. For normalization parameters, to Weights within the same group All six types of bias are located in [0, 1]. This is the consistency parameter for unnormalized relations.

[0044] The field determination module is used to fuse bidirectional attribution parameters and relation consistency parameters using a posterior fusion algorithm to obtain the final attribution parameters of each candidate standard field. Then, the candidate standard fields are sorted according to the final attribution parameters, and the candidate standard field with the first position in the sort is selected as the standard declaration field of the target information unit. Specifically, the field determination module includes a parameter conversion unit, a fusion calculation unit, and a field sorting unit. The parameter conversion unit normalizes... The fusion computing unit integrates bidirectional attribution parameters and consistency probability, and the specific calculation formula is as follows: ; (12) in, The consistency probability after one normalization. , For group fusion weights, For numerical stability parameters; field sorting units are based on The fields are sorted in descending order to determine the first standard declaration field and other candidates and their probabilities are retained. The support level of the current field and the context, as well as the reasonableness of the relationship after substituting counterfactual facts, are used together to determine the field.

[0045] The data processing module is used to associate the field values ​​of each information unit in the information unit sequence with the standard declaration fields selected by the field determination module for that information unit, and to compile them into a standardized declaration dataset for the filing declaration documents. It also uses an adaptive update algorithm to adjust the field association parameters and mapping parameters according to the final attribution parameters of each information unit, thereby forming an updated standard field syntax diagram and field mapping parameter set. Specifically, the data processing module associates the field values ​​of information units with the determined standard declaration fields to form a standardized declaration dataset. Each record includes at least the standard declaration field identifier, name, standard field value, original field expression, original field value, source file, original file location, and final attribution parameter. It also saves local, forward, backward, bidirectional, deviation, consistency, and update parameters. Multi-value fields retain multiple values ​​and their sources, while single-value fields retain the valid value corresponding to the highest final attribution parameter. Other relevant information units are saved as associated evidence.

[0046] The candidate information entropy, bidirectional consistency parameter, and source reliability parameter are calculated according to the following formula (13): ; ; (13) in, , , Calculated using the natural logarithm, by convention ; As the source project weight, To apply the indicated value, The verification result is [0, 1]. Items include the identity of the uploading entity, hash integrity, layout integrity, character recognition confidence, and electronic signature. Inapplicable items are removed from both the numerator and denominator. Weights are determined based on historical verification data. When the sample size is insufficient, weights are equal. When the denominator is 0, a weight of 0.5 is used for calculation, and missing markers are retained. The relevant update amount is recalculated after valid source verification is completed. Knowledge updates are not submitted until missing markers are eliminated.

[0047] Candidate distinguishing parameters for The difference between the first and second probabilities, assuming a stable parameter. for The difference between the first and second probabilities is used to perform a reliable update according to the following formula (14): ; ; (14) in, This is the lower limit of the single update factor. Based on the learning rate, all standard declaration fields are updated, excluding candidate fields. Take 0, Continuous calculation, without setting a confidence threshold branch.

[0048] Relationship Exposure Value A value of either 0 or 1 indicates whether the relationship can be tested. Located in [0, 1], representing the actual strength, the sequential test requires the existence of a direct directed edge within the same reading range. Pick The co-occurrence test requires that there be identifiable chapters, tables, or lists belonging to the same declared object, within the same common scope. Pick Otherwise, take 0; the region test requires two valid non-root region paths. Pick Role verification requires that both nodes involve personnel types or role candidates, and that compatibility be clearly defined. Take 1; take 0 when different names compete for a single role; take 0.5 when there is insufficient basis. Role basis prioritizes reading the original expression and business rules, and does not replace the current final result.

[0049] First-order statistics are updated according to formula (15): ; ; (15) in, For a valid directed node pair, For historical decay parameters, Prior strength. Second and third order strengths use continuous windows. , or Window weight Precedence field context The increment is The exposure increment is After performing historical attenuation and prior fusion respectively, higher-order probabilities are generated, exposing the probability of the target field of the end node and the window strength without multiplying the statistics.

[0050] Co-occurrence, region, and role are statistically analyzed using unordered field pairs. Each node pair is analyzed by... Only counted once. hour , hour The unified update is as follows: ; ; ; (16) in, For co-occurrence, region, or role relationship, For the corresponding set of verifiable node pairs; For the first The prior strength of class relations is summarized in this way, which aggregates the soft mapping contributions in both directions, so that the statistics and parameters remain symmetric.

[0051] Mutual exclusion uses the probability of allowing co-occupancy as an intermediate value. for , If, simultaneously, the representative information units entering the candidate set are from the same source or belong to the same equivalent group, only one representative is counted, then: ; ; (17) in, For manual business mutual exclusion priors, The strength of prior knowledge; as the evidence of shared possession increases, the probability of mutual exclusion decreases, affecting the strength of prior knowledge and... Restrictions are imposed, and unordered field pairs generate symmetric mutually exclusive parameters and are substituted back into equation (1).

[0052] The base interval is set as The last term is greater than or equal to The overflow interval. Use the final... In the alternative (11) Calculate the number of expected unique values Field Sample Weights non-integer Assign based on its distance from adjacent integers. All amounts are included in the overflow interval, and the allocation ratio is denoted as follows: ,but: ; (18) in, When the relevant node is empty Set to 0. Write the new cardinal distribution. The data processing module then recalculates the order of each level and the associated parameters of the remaining fields, the relational channel parameters, the field model, and the standard field syntax diagram to form a knowledge state to be submitted, while the knowledge version used in the current file remains unchanged.

[0053] Offline training samples consist of verified information units, standard field labels, and graphical evidence, grouped into training, validation, and testing data according to the submitting organization and template version. The training candidate set uses a fixed... Each field has a corresponding minimum local matching candidate; if a correct field is missing, the candidate is replaced. At runtime, candidates are still generated according to the normal matching order. Local, forward, bidirectional, and final losses are respectively applied to... , , and Calculate the cross-entropy; the sorting loss is... The probability calibration loss is The squared error of the unique hot label, the total loss is the weighted sum of six losses. .

[0054] The semantic encoder, structure matching model, and fusion weights are trained using gradient descent, and the weights are reassembled using... Parameterization fixes the candidate set, hypothesis fields, and business indication values ​​in a single calculation, without backpropagation of their generation process; soft mapping and continuous scoring preserve gradients, and the truncation function is calculated using the usual derivative within its effective interval and zero gradient outside the interval. , Order weights, adjustment parameters , , Prior strength, ordination interval and By validating the data selection, the sequence and cardinality normalization parameters were set to the 95th percentile of historical deviation and not less than [a certain value]. The positive lower bound. Validation metrics include candidate coverage, final assignment loss, and calibration error. The loss decreases by less than [value missing] for five consecutive rounds of validation. Training will stop when the time is right. Model weights are only adjusted during offline training or version upgrades.

[0055] The filing and approval module is used to match the standardized application dataset with the preset filing rules, generate filing and approval records based on the matching results, and extract standardized application data that does not meet the preset filing rules to form anomalies. The filing and approval module reads the standardized application dataset, converting laboratory level, pathogenic microorganism category, experimental activity type, personnel qualifications, facilities and equipment, and safety protection conditions into approval features. These features are then matched against preset filing rules that include rule identifiers, applicable conditions, required fields, field values, and associated constraints, generating filing and approval records and anomalies. Anomalies are stored with the relevant fields, field values, application object, and source location, and are associated with the standardized application dataset.

[0056] Example rule R-BSL2-02 requires that when the laboratory level is BSL-2 and the pathogen category meets the preset conditions, a biosafety officer, laboratory personnel, personnel qualifications, and safety protection conditions must be present simultaneously, and the correspondence between qualifications and activity types must be verified. This rule only describes the system's processing method and does not represent a specific administrative standard. If personnel qualifications are missing, a "Required Standard Declaration Field Missing" exception will be generated, linking the laboratory name, personnel, and source records.

[0057] Once standardized data, approval records, and anomalies have passed integrity checks and been successfully stored, and the source parameters are reliable with no missing markers, the status of the fields to be updated is submitted in a transactional manner, and the version is registered for the next file to read. Approval anomalies do not automatically negate field knowledge updates; source or storage issues only delay knowledge submission and do not change the field identification path.

[0058] For example, suppose a document contains "Affiliation: Unit A", "Laboratory Name: Laboratory B", "Laboratory Address: Address C", and "Responsible Person: Zhang San", an attachment contains "Biosafety Officer: Zhang San", and another location contains "Laboratory Head: Li Si". After parsing, establish evidence of sequence, structure, region, duplicate information, and role, and group the same Zhang San in both the text and the attachment into an equivalent group with the same field value.

[0059] Total number of example fields In addition to the aforementioned fourteen types of fields, it includes contact person and legal representative. The candidates, in order, are the biosafety officer, laboratory director, contact person, legal representative, and laboratory personnel. The values ​​are 0.720, 0.780, 0.300, 0.200, and 0.100, respectively. The values ​​are 1.200, 1.300, -0.500, -0.800, and -1.000, respectively. The values ​​are 0.500, -1.442, -2.188, -2.608, and -3.270, respectively. The values ​​are 1.700, -0.142, -2.688, -3.408, and -4.270, respectively. (Take...) , , , Approximately 0.65008, 0.26991, 0.04001, 0.02501, and 0.01500.

[0060] The weights for all six types of deviations are 1 / 6, with corresponding deviations for the biosafety officer being 0.12, 0.10, 0.18, 0.04, 0.08, and 0.16. Approximately 0.11333; the corresponding deviations for the laboratory supervisors were 0.08, 0.10, 0.12, 0.55, 0.50, and 0.63. It is 0.33000; the rest The values ​​are 0.560, 0.611, and 0.652. (Take...) , Approximately 0.87911, 0.10071, 0.01010, 0.00606, and 0.00402; take... , , The values ​​are approximately 0.78663, 0.17156, 0.02091, 0.01281, and 0.00808, and the responsible personnel are identified as the biosafety officer.

[0061] According to the complete , Calculated Approximately 0.82643, Approximately 0.41435, Approximately 0.61507, The value is approximately 0.77840. The verification weights for identity, hash, layout, and character recognition are 0.25, 0.20, 0.20, and 0.20 respectively, resulting in values ​​of 1, 1, 0.95, and 0.85. Therefore, electronic signatures are not applicable. =0.81 / 0.85≈0.95294. Take... , It is approximately 0.22499.

[0062] Before the update, the mapping parameters for the five candidates corresponding to this expression were 0.545, 0.345, 0.040, 0.040, and 0.020, respectively. The remaining eleven fields totaled 0.010 and were evenly distributed. After the update, the five parameters are approximately 0.54772, 0.34305, 0.03979, 0.03969, and 0.01987, respectively, and the remaining fields total approximately 0.00989. The sum of the complete mapping distributions without rounding is 1.

[0063] Sequential update example The original occurrence statistic is 72, the exposure statistic is 100, the prior strength is 10, and the prior probability is 0.05; laboratory address node. 0.80, corresponding The value is 0.95, the sequence intensity is 0.50, the exposure value is 1, the increment is approximately 0.06725, the exposure increment is approximately 0.17099, and the sequence probability changes from 72.5 / 110≈0.65909 to 72.56725 / 110.17099≈0.65868. It changed from approximately 0.31818 to approximately 0.31736.

[0064] The mutual exclusion example uses a prior value of 0.90, an intensity of 20, an original number of shared occupancy attempts of 2, and an exposure count of 10. The current increments are approximately 0.06073 and 0.22499, respectively, and the mutual exclusion probability changes from 0.86667 to approximately 0.86565. The cardinality example uses three intervals: zero exposures, one exposure, and more than two exposures, with original counts of 10, 85, and 5. When the expected number of unique values ​​is 1 and the field weight is 0.22499, the probabilities change from approximately 0.10680, 0.83495, and 0.05825 to approximately 0.10656, 0.83531, and 0.05813.

[0065] Standardized records are maintained, including the biosafety officer, Zhang San, the person responsible for the original expression, the source location, and the final attribution parameters. If the conditions of the example approval rule are met but personnel qualifications are missing, a related anomaly is generated. After the relevant data is successfully stored, a pending knowledge status is submitted for use in the next step of filing and application documents.

[0066] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0067] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A pathogenic microorganism laboratory registration and approval management system, characterized in that the system... include: The field graphing module is used to obtain standard declaration fields and historical declaration data, calculate field association parameters of the relationship between standard declaration fields and mapping parameters between the original field expressions and standard declaration fields based on the historical declaration data, and construct a standard field syntax graph based on the standard declaration fields and field association parameters. The document graphing module is used to acquire the filing application documents to be processed, then perform structural parsing on the filing application documents to form an information unit sequence, and calculate the document association parameters of each information unit based on the information unit sequence, and construct a document evidence graph based on the parameters; The bidirectional attribution module is used to select target information units sequentially in the document evidence graph, and use a candidate matching algorithm combined with mapping parameters to filter multiple candidate standard fields. Based on the document evidence graph and the standard field syntax graph, the forward and backward propagation algorithms are used sequentially to calculate the bidirectional attribution parameters of each candidate standard field. The consistency module is used to take each candidate standard field as the hypothetical attribution field of the target information unit, and to use the counterfactual consistency algorithm to calculate the relational consistency parameter of each hypothetical attribution field. The field determination module is used to fuse the bidirectional attribution parameters and relation consistency parameters using a posterior fusion algorithm to obtain the final attribution parameters of each candidate standard field. Then, the candidate standard fields are sorted according to the final attribution parameters, and the candidate standard field with the first position in the sort is selected as the standard declaration field of the target information unit. The data processing module is used to associate the field values ​​of each information unit in the information unit sequence with the standard declaration fields selected by the field determination module for that information unit, and to collect them to form a standardized declaration dataset for the filing declaration documents. It also uses an adaptive update algorithm to adjust the field association parameters and mapping parameters according to the final attribution parameters of each information unit, thereby forming an updated standard field syntax diagram and field mapping parameter set.

2. The pathogenic microorganism laboratory registration and approval management system according to claim 1, characterized in that, The field association parameters include sequential association parameters, co-occurrence association parameters, region association parameters, role association parameters, mutual exclusion association parameters, and cardinality association parameters; The field graphing module calculates the field association parameters based on the arrangement status, co-occurrence status, regional affiliation status, role status, mutual exclusion status, and field quantity status of the standard declaration fields in the historical declaration data. It then constructs a standard field syntax graph using the standard declaration fields as nodes and the field association parameters as edge parameters connecting the standard declaration field nodes.

3. The pathogenic microorganism laboratory registration and approval management system according to claim 1, characterized in that, The document association parameters include sequential association parameters, structural association parameters, regional association parameters, and cross-location association parameters; The document graphing module calculates the document association parameters based on the reading order, document structure, regional affiliation, and duplicate information of each information unit in the filing application document, and constructs the document evidence graph using each information unit as a node and the document association parameters as edge parameters connecting the information unit nodes.

4. The pathogenic microorganism laboratory registration and approval management system according to claim 1, characterized in that, The bidirectional attribution module includes: The candidate filtering unit is used to calculate the local matching parameters between the target information unit and each standard declaration field based on the field semantic features, field value features, document structure features and mapping parameters of the target information unit, and to filter multiple candidate standard fields according to the local matching parameters. The forward propagation unit is used to calculate the forward attribution parameters of each candidate standard field based on the preceding information unit of the target information unit, document association parameters, and field association parameters. The backward propagation unit is used to modify the forward attribution parameters based on the subsequent information units of the target information unit, document association parameters, and field association parameters, using the forward attribution parameters as initial parameters, to form bidirectional attribution parameters for each candidate standard field.

5. The pathogenic microorganism laboratory registration and approval management system according to claim 1, characterized in that, The consistency module includes: The hypothetical mapping unit is used to sequentially establish hypothetical mapping relationships between each candidate standard field and the target information unit. The deviation calculation unit is used to calculate the relationship deviation parameter between the association relationship in the document evidence diagram and the field association relationship in the standard field syntax diagram, based on each hypothetical mapping relationship; The parameter generation unit is used to generate relation consistency parameters for each assumed attribution field based on the relation deviation parameters.

6. The pathogenic microorganism laboratory registration and approval management system according to claim 1, characterized in that, The field determination module includes: The parameter transformation unit is used to normalize the relationship consistency parameters of each hypothesized attribution field to form a consistency probability. The fusion computing unit is used to fuse the bidirectional attribution parameters and consistency probabilities of each candidate standard field to form the final attribution parameters; The field sorting unit is used to sort the candidate standard fields in descending order according to the final attribution parameter, and to take the candidate standard field with the first position in the sort as the standard declaration field of the target information unit.

7. The pathogenic microorganism laboratory registration and approval management system according to claim 5, characterized in that, The relational deviation parameters include field sequence deviation parameters, document structure deviation parameters, region deviation parameters, duplicate information deviation parameters, role deviation parameters, and field cardinality deviation parameters; The deviation calculation unit calculates the deviation parameters of each relationship respectively, and merges the deviation parameters of each relationship to obtain the comprehensive deviation parameter of the assumed attribution field. The parameter generation unit generates the relationship consistency parameter based on the comprehensive deviation parameter.

8. The pathogenic microorganism laboratory registration and approval management system according to claim 4, characterized in that, The adaptive update algorithm includes: Calculate candidate distribution parameters and candidate differentiation parameters based on the final attribution parameters of each candidate standard field, calculate bidirectional consistency parameters based on the forward attribution parameters and bidirectional attribution parameters, obtain hypothetical stability parameters based on the relationship consistency parameters of each hypothetical attribution field, and obtain source reliability parameters based on the data source of the filing application documents. The candidate distribution parameters, candidate differentiation parameters, bidirectional consistency parameters, assumed stability parameters, and source reliability parameters are integrated to form update adjustment parameters, and the update magnitude of field association parameters and mapping parameters is adjusted according to the update adjustment parameters.

9. The pathogenic microorganism laboratory registration and approval management system according to claim 1, characterized in that, The filing application documents include one or more of the following: text documents, spreadsheets, formatted documents, and image documents; The document mapping module extracts the text content, table structure, page position and field value information from the filing application document, and converts the extracted results into information units with a unified data structure according to the document reading order, forming an information unit sequence; The standardized declaration dataset includes standard declaration fields, field values, original field expressions, original file locations, and final attribution parameters.

10. The pathogenic microorganism laboratory registration and approval management system according to claim 1, characterized in that, The system also includes a filing and approval module; The filing and approval module is used to match the standardized application dataset with preset filing rules, generate filing and approval records based on the matching results, and extract standardized application data that does not meet the preset filing rules to form anomalies. The data processing module is also used to associate and store the filing approval records and anomalies with the standardized application dataset, and then use the updated standard field syntax diagram and field mapping parameter set for the processing of the next filing application document.