An intelligent method for large-scale data review and verification of traditional Chinese medicine new drugs

Abstract

The application relates to the technical field of medical data processing, and discloses an intelligent method for large-scale data examination and verification of new Chinese medicine, which comprises the following steps: acquiring a new Chinese medicine data source set and mapping the new Chinese medicine data source set into a standardized record set; generating a record abstract signature according to each standardized record and forming a verification snapshot; taking the record abstract signature as a basis to construct an association relationship graph; extracting a detection fingerprint feature vector, generating an entity matching candidate pair, solving a global constraint under a consistency constraint, and obtaining a unique association; when the standardized record set is changed, a verification alarm set is obtained; for each verification alarm, a minimum conflict subgraph that makes the verification alarm true is extracted, and an evidence package is generated; receiving a review feedback, and updating an enabling condition of a verification constraint based on a trigger constraint identifier. The application improves the traceability, reproducibility and alarm reliability of large-scale data examination and reduces review cost.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, and more specifically, to an intelligent method for large-scale data review and verification of new traditional Chinese medicine drugs. Background Technology

[0002] In the research and development and registration application process of new traditional Chinese medicine drugs, data typically comes from laboratory information management systems, electronic experimental records, quality inspection systems, statistical analysis systems, and testing instrument platforms. This includes structured records such as batches, samples, prescriptions, procedures, and test results, as well as data from testing instruments such as spectra. With the increasing number of application batches and the expanding scale of data, the review and verification process places higher demands on the authenticity, completeness, traceability, and consistency of the data.

[0003] In existing technologies, the closest solutions often remain at the level of general frameworks for data governance or data quality monitoring, making it difficult to simultaneously meet the three key requirements of record-level reproducibility and playback, cross-source unique association, and evidence-level verifiability. For example, patent document CN111190881A provides a governance framework for data cleaning and fusion, metadata governance, standardization, and quality management, but lacks a mechanism to bind each record with metadata and rule versions to form a verification snapshot. When version drift occurs in field semantics, code tables, or unit calibers, only verification conclusions with inconsistent calibers at different times can be obtained. CN111916161B uses multiple data sources and establishes mapping relationships through fuzzy matching to achieve conversion. CN114155925B focuses on conversion compatible with multiple versions of source data types. Although this can improve conversion efficiency, in the scenario of new traditional Chinese medicine drugs, where the same physical sample has multiple numbers across systems and test results are associated with spectral files across different libraries, fuzzy matching or type conversion is prone to ambiguous many-to-one or one-to-many relationships. It lacks the ability to solve global constraints under consistency constraints to determine the unique association between batch objects and sample objects, sample objects and test result objects, and test result objects and instrument operation objects. Therefore, in verification scenarios, problems such as low alarm reliability, high review costs, and difficulty in accurately quantifying the scope of impact after rectification may still occur.

[0004] Therefore, it is necessary to design an intelligent method for large-scale data review and verification of new traditional Chinese medicine drugs to solve the problems existing in the current technology. Summary of the Invention

[0005] In view of this, the present invention proposes an intelligent method for large-scale data review and verification of new traditional Chinese medicine drugs, aiming to solve the problems of low alarm reliability, high review cost, and difficulty in accurately quantifying the scope of impact after rectification in the verification scenario.

[0006] This invention proposes an intelligent method for large-scale data review and verification of new traditional Chinese medicine drugs, comprising: Obtain the data source set of new traditional Chinese medicine drugs and map it into a standardized record set; A record digest signature is generated for each standardized record in the standardized record set. The record digest signature is then bound to the corresponding metadata version identifier and rule version identifier to form a verification snapshot. Based on the standardized record set, batch objects, sample objects, test result objects and instrument operation objects are extracted, and a relationship graph is constructed using the record summary signature as the association basis. Data from the testing instrument is obtained from the standardized record set, and the detection fingerprint feature vector is extracted. An evidence vector is constructed based on the identification elements, attribute summaries, traceability weights, and the detection fingerprint feature vector. Entity matching candidate pairs are generated, and global constraint solutions are performed under consistency constraints to obtain unique associations between batch objects and sample objects, sample objects and test result objects, and test result objects and instrument operation objects. The association graph is updated based on the unique associations. When the standardized record set changes, a dependency subgraph is obtained by expanding the relationship graph, and the corresponding verification constraint is selected according to the rule version identifier in the dependency subgraph for verification to obtain a set of verification alarms. For each verification alarm in the verification alarm set, the minimum conflict subgraph that makes the verification alarm valid is extracted from the dependency subgraph, and an evidence package is generated. The evidence package includes at least the trigger constraint identifier, related data differences, version differences, source information, and verification snapshot reference. Receive verification feedback and update the enabling conditions of the verification constraint based on the trigger constraint identifier.

[0007] Furthermore, when obtaining the data source set of new traditional Chinese medicine drugs and mapping it to a standardized record set, the following is included: For each data source in the set of new Chinese medicine drug data sources, the field definition, unit of measurement, code table definition and detection methodology parameters are read, and version description information is formed in a preset order. The metadata version identifier is determined based on the version description information. The mapping rules corresponding to the metadata version identifier are invoked to perform field mapping, unit conversion, and code table conversion on the source record to obtain a standardized record; For each of the standardized records, traceability information is written, which includes the source system identifier, the source record identifier, and the collection time.

[0008] Furthermore, when generating a record digest signature and forming a verification snapshot, the process includes: A set of key fields is determined for each standardized record. The set of key fields includes at least an identifier field corresponding to the batch object, sample object, test result object, or instrument operation object, as well as a field corresponding to the unit of measurement and code conversion result. The field values ​​of the key field set are formatted and serialized content is generated according to a fixed field order. Perform digest calculation on the serialized content to obtain the record digest signature; The record digest signature, along with the metadata version identifier, rule version identifier, and traceability information corresponding to the standardized record, are written into the verification snapshot.

[0009] Furthermore, when constructing the relationship graph, the following is included: According to the preset object extraction rules, the object type is identified from each standardized record and an object identifier is generated. The object identifier is determined by the identifier field in the standardized record. Standardized records with the same object identifier are grouped into the same object, and the corresponding record summary signature is written into the object evidence reference set. Based on the association fields in the standardized records, association edges are generated between batch objects and sample objects, sample objects and test result objects, and test result objects and instrument operation objects, respectively. Each association edge is written into an edge evidence reference set, which includes the record summary signature that supports the generation of the association edge. When generating the association edge, association edges that do not meet the cardinality constraint or time order constraint are marked as candidate associations and retained in the association graph.

[0010] Furthermore, when acquiring detection instrument data and extracting the fingerprint feature vector, the process includes: Read the instrument operation record and test result record from the data of the testing instrument, and extract the testing methodology parameters; Peak identification is performed on the original detection signal corresponding to the instrument operation record to obtain a peak feature set; the peak feature set is aligned based on the detection methodology parameters, and the aligned peak feature set is encoded into a detection fingerprint feature vector in a fixed order; The detected fingerprint feature vectors are written into the association information of the instrument operation object and the detection result object, respectively.

[0011] Furthermore, when constructing the evidence vector and generating entity matching candidate pairs, the following steps are included: Extract the identifier elements and attribute summaries of the objects to be matched, and obtain the source weights and detection fingerprint feature vectors corresponding to the objects to be matched; The identification elements, attribute summaries, and detection fingerprint feature vectors are compared for consistency to obtain identification consistency results, attribute consistency results, and fingerprint consistency results; the identification consistency results, attribute consistency results, and fingerprint consistency results are then combined with the source tracing weight to form an evidence vector; The entity matching candidate pairs are generated based on the matching score of the evidence vector and the consistency condition of the detection methodology parameters.

[0012] Furthermore, when solving for global constraints and updating the association graph under consistency constraints, the process includes: Based on the entity matching candidate pairs, a candidate association set is constructed, which covers candidate associations from batch objects to sample objects, from sample objects to test result objects, and from test result objects to instrument operation objects. The consistency constraints are applied to the candidate association set, and the consistency constraints include at least uniqueness constraints, cardinality constraints, temporal order constraints, and detection methodology consistency constraints. Under the consistency constraint, the candidate association combination with the largest total matching score is selected as the unique association; the unique association is written into the association graph and the corresponding candidate association is replaced.

[0013] Furthermore, when the standardized record set changes and a verification alarm set is obtained within the dependency subgraph, it includes: Based on the source tracing information, a set of change records is determined. The set of change records includes the record digest signature corresponding to the newly added source record identifier, and the record digest signature that has the same source system identifier and source record identifier as the previous verification snapshot but has a different record digest signature. In the relationship graph, retrieve the object evidence reference set and edge evidence reference set that reference the change record set to determine the seed object set and the seed association edge set; Starting from the set of seed objects and the set of seed associated edges, the dependency subgraph is obtained by expanding outward along the associated edges of the association graph to a preset number of layers. Within the dependency subgraph, the corresponding verification constraint is selected according to the rule version identifier, and the objects referencing the change record set and their associated edges are verified to obtain a set of verification alarms. For each verification alarm, a trigger constraint identifier, an associated record summary signature, and the corresponding metadata version identifier and rule version identifier are written.

[0014] Furthermore, when extracting the minimum conflict subgraph from the dependency subgraph and generating the evidence package, the process includes: For each verification alarm, the set of objects and the set of associated edges related to the verification constraint are determined in the dependency subgraph according to the trigger constraint identifier, and the set of objects and the set of associated edges are used to form a conflict subgraph; While ensuring the verification alarm is valid, gradually eliminate objects or associated edges in the conflict subgraph that do not affect the validity of the verification alarm to obtain the minimum conflict subgraph that makes the verification alarm valid. Generate an evidence package, which includes at least the trigger constraint identifier, the minimum conflict subgraph, the record digest signature referenced by the minimum conflict subgraph, relevant data differences extracted from the standardized record set, version differences determined by the metadata version identifier, the tracing information, and the verification snapshot reference.

[0015] Furthermore, when receiving review feedback and updating the verification enable conditions, this includes: Receive the review result corresponding to the evidence package, the review result including an anomaly establishment indicator or an anomaly non-establishment indicator; When the exception does not meet the identifier, the verification constraint corresponding to the trigger constraint identifier is read, and the metadata version identifier, rule version identifier, detection methodology parameters, and object type involved in the minimum conflict subgraph are determined according to the verification snapshot reference. The metadata version identifier, rule version identifier, detection methodology parameters, and object type are written into the activation condition of the verification constraint, so that the verification constraint is only enabled when the metadata version identifier, rule version identifier, detection methodology parameters, and object type are consistent. When the exception meets the identifier, the activation condition of the verification constraint remains unchanged.

[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: By taking multi-source heterogeneous data of new traditional Chinese medicine drugs as the object, data from different systems and testing instrument platforms are uniformly mapped into a standardized record set. Then, a record summary signature is generated for each standardized record and bound to metadata version identifiers and rule version identifiers to form a verification snapshot. This allows any verification conclusion to be repeatedly played back under the same version caliber, fundamentally reducing the risk of inconsistencies caused by field semantics, code table, and unit caliber drift. Subsequently, based on the record summary signature, batch objects, sample objects, test result objects, and instrument operation objects are extracted and a relationship graph with evidence citations is constructed, solidifying the originally scattered link relationships across multiple systems into traceable object associations. A detection fingerprint feature vector is extracted from the standardized record set, and the identification elements, attribute summaries, traceability weights, and detection fingerprint features are combined to form an evidence vector. Global constraint solving is performed under consistency constraints, thereby addressing the issues of multiple numbering and cross-database issues. Even in complex scenarios, a unique association between batch objects and instrument operation objects can still be obtained, reducing mismatches and false alarms caused by ambiguous many-to-one or one-to-many matching. When data changes, the dependency subgraph is expanded through the relationship graph and the corresponding verification constraint is selected according to the rule version identifier. Incremental verification is only performed on the affected range, which ensures processing efficiency in large-scale scenarios and avoids alarm propagation caused by full recalculation. For each verification alarm, the minimum conflict subgraph that makes the alarm valid is extracted and an evidence package containing trigger constraint identifier, data difference, version difference, traceability information and verification snapshot reference is generated. This upgrades the alarm from "result prompt" to "explainable, traceable and verifiable" verification evidence, thereby reducing review costs and improving rectification and location efficiency. Finally, the review feedback is used to update the activation conditions of verification constraints, so that the rules can adaptively converge under specific version caliber and methodology conditions, continuously reducing false alarms and improving alarm reliability and review and verification pass rate. Attached Figure Description

[0017] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart of an intelligent method for large-scale data review and verification of new traditional Chinese medicine drugs provided in an embodiment of the present invention. Detailed Implementation

[0018] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0019] For this, please refer to Figure 1 As shown, this application proposes an intelligent method for large-scale data review and verification of new traditional Chinese medicine drugs, including: S100: Obtain the data source set of new Chinese medicine drugs and map it to a standardized record set.

[0020] S200: Generate a record digest signature based on each standardized record in the standardized record set, and bind the record digest signature to the corresponding metadata version identifier and rule version identifier to form a verification snapshot.

[0021] S300: Based on a standardized record set, batch objects, sample objects, test result objects, and instrument operation objects are extracted, and a relationship diagram is constructed using the record summary signature as the basis for association.

[0022] S400: Obtain testing instrument data from the standardized record set and extract the testing fingerprint feature vector. Construct evidence vectors based on identification elements, attribute summaries, traceability weights, and testing fingerprint feature vectors. Generate entity matching candidate pairs and solve global constraints under consistency constraints to obtain unique associations from batch objects to sample objects, sample objects to test result objects, and test result objects to instrument operation objects. Update the association graph based on the unique associations.

[0023] S500: When the standardized record set changes, the dependency subgraph is expanded on the relationship graph, and the corresponding verification constraint is selected according to the rule version identifier within the dependency subgraph for verification to obtain the verification alarm set.

[0024] S600: For each verification alarm in the verification alarm set, extract the minimum conflict subgraph that makes the verification alarm valid from the dependency subgraph, and generate an evidence package. The evidence package includes at least the trigger constraint identifier, relevant data differences, version differences, traceability information, and verification snapshot reference.

[0025] S700: Receives verification feedback and updates the enable conditions of the verification constraint based on the trigger constraint identifier.

[0026] Specifically, in some embodiments of this application, the data source set for new traditional Chinese medicine drugs refers to a collection of multi-source data associated with the same batch of applications, including at least sample ledgers and inspection records from the laboratory information management system, process and prescription records from the electronic experimental record system, test result records from the quality inspection system, summary table records from the statistical analysis system, and instrument operation records and original test signal files from the testing instrument platform. Mapping to a standardized record set refers to the set of records formed by mapping the source records from each data source to fields, converting units of measurement, and converting code tables according to a unified data model. Each standardized record contains traceability information, including the source system identifier, source record identifier, and collection time. The metadata version identifier is a version identifier determined based on version description information formed in a preset order according to field definitions, units of measurement, code table definitions, and testing methodology parameters, used to characterize the data caliber used in this standardized record. The rule version identifier is used to characterize the version of the verification constraints used when performing verification on this standardized record. The record summary signature is a summary value obtained by normalizing the format of the key fields of each standardized record, serializing them in a fixed field order, and then performing summary calculation. It uniquely indicates the content status of the standardized record under a specific data caliber. The verification snapshot saves the binding relationship between the record summary signature and the metadata version identifier, rule version identifier, and traceability information, enabling repeated playback of the verification process and results under the same caliber. The association graph includes batch objects, sample objects, detection result objects, and instrument operation objects, as well as the association edges between them. Record summary signatures supporting the association relationships are written into the object evidence reference set and the edge evidence reference set. The detection fingerprint feature vector is obtained by obtaining the peak feature set from the original detection signal through peak identification, performing alignment processing based on the detection methodology parameters, and encoding it in a fixed order. The evidence vector is used to comprehensively identify element consistency, attribute summary consistency, detection fingerprint consistency, and traceability weight, supporting the generation of entity matching candidate pairs and the global constraint solution under consistency constraints. Consistency constraints include at least uniqueness constraints, cardinality constraints, time order constraints, and detection methodology consistency constraints. The dependency subgraph is obtained by expanding outward from the seed object set and seed edge set of the reference change record set in the relationship graph, and is used to limit the scope of incremental verification. The minimum conflict subgraph is the minimum set of causes obtained by trimming the dependency subgraph while ensuring that the verification alarm is valid. The evidence package includes at least the trigger constraint identifier, relevant data differences, version differences, traceability information, and verification snapshot reference, and is used for review and rectification. The enabling conditions are used to limit the scope of application of the verification constraints in subsequent verifications.

[0027] Specifically, in one optional implementation, entity matching candidate pairs are transformed into a weighted set of candidate associations, where each candidate association carries its evidence vector identifier, matching score, and constraint label derived from consistency constraints. Then, an optimization problem is constructed in the form of an object-object bipartite or multipartite graph: at the hard constraint level, uniqueness constraints, cardinality constraints, temporal order constraints, and consistency constraints of the detection methodology are encoded as inviolable constraints; at the objective function level, the optimization objective is to maximize the sum of matching scores of candidate associations, and the optimal solution set satisfying the hard constraints is obtained using maximum weighted matching, constraint satisfaction solving, or integer programming solvers. When multiple equally optimal solutions exist, disambiguation selection is further performed according to preset deterministic rules. These deterministic rules include at least "prioritizing associations generated by data sources with higher traceability weights," "prioritizing candidate associations with more record digest signature citations," and "prioritizing association combinations with smaller time spans and more stable temporal sequences," thereby ensuring consistent unique association results are output and the association graph is stably updated under the same verification snapshot and rule version identifier.

[0028] Specifically, in one optional implementation, the preset number of layers in the dependent subgraph is determined by the object type and associated edge type involved in the verification constraint entry corresponding to the trigger constraint identifier: when the verification constraint only involves the field consistency between the sample object and the detection result object, the preset number of layers can be one-hop or two-hop extension covering the sample object to the detection result object. When the verification constraint involves the integrity or timing consistency of the link from the batch object to the instrument operation object, the preset number of layers can be the full-link extension covering the batch object—sample object—detection result object—instrument operation object. Meanwhile, to avoid excessive diffusion, a maximum layer limit and early stopping conditions can be set during the expansion process. The early stopping conditions include at least "the newly added objects expanded to no longer reference the change record set" and "the associated edges expanded to no longer satisfy the edge type filtering conditions related to the trigger constraint identifier," thereby controlling the computational scale of incremental verification and reducing the diffusion of irrelevant alarms while ensuring that the affected scope is not overlooked.

[0029] Specifically, taking a specific batch of applications as an example, step S100: The sample ledger, testing items, and testing records corresponding to this batch of applications are read from the laboratory information management system; the prescription composition and process records are read from the electronic experimental record system; and the instrument operation records and original testing signal files associated with the testing records are read from the testing instrument platform. Subsequently, based on a unified data model, field mapping, unit conversion, and code table conversion are performed on the aforementioned source records to form a standardized record set. For each standardized record, the source system identifier, source record identifier, and acquisition time are written. Step S200: The key field set is extracted from each standardized record, and format normalization and fixed-order serialization are completed to obtain the record summary signature. Simultaneously, version description information is formed based on field definitions, units of measurement, code table definitions, and testing methodology parameters to determine the metadata version identifier. The rule version identifier is determined based on the currently used verification constraint version. The record summary signature, metadata version identifier, rule version identifier, and traceability information are jointly written into the verification snapshot. Step S300: Identify batch objects, sample objects, test result objects, and instrument operation objects from the standardized record set. Group standardized records with the same object identifier into the same object and write the corresponding record summary signature into the object evidence reference set. Establish association edges between batch objects and sample objects, sample objects and test result objects, and test result objects and instrument operation objects based on the association fields in the standardized records. Write the record summary signature supporting the association edge into the edge evidence reference set to form an association graph. Step S400: Obtain test instrument data from the standardized record set. Perform peak identification on the original test signal to obtain a peak feature set. Align the peak feature set based on the test methodology parameters and encode it into a test fingerprint feature vector. Write the test fingerprint feature vector into the association information between the test result object and the instrument operation object. Extract identifier elements and attribute summaries from the objects to be matched, and construct evidence vectors by combining the traceability weights and test fingerprint feature vectors to generate entity matching candidate pairs. Under consistency constraints, perform global constraint solving on the candidate association set, select the candidate association combination that satisfies the consistency constraints and has the largest total matching score as the unique association, and update the association graph with this unique association. Subsequently, when a source system modifies the record content corresponding to the same source record identifier and forms a new standardized record, step S500: align the source system identifier with the source record identifier based on the traceability information, and determine that the record summary signature is inconsistent with the previous verification snapshot, thereby including the record summary signature in the change record set. In the relationship graph, retrieve the object evidence reference set and edge evidence reference set that reference the change record set, determine the seed object set and seed association edge set, and expand outward along the association edges to a preset number of layers to obtain the dependency subgraph.Within the dependency subgraph, select the corresponding verification constraint based on the rule version identifier, perform verification on the objects and their associated edges in the referenced change record set, obtain the verification alarm set, and write the trigger constraint identifier, associated record summary signature, and version identifier. For each verification alarm, step S600: Based on the trigger constraint identifier, determine the relevant object set and associated edge set to form a conflict subgraph in the dependency subgraph. While maintaining the validity of the verification alarm, gradually eliminate objects or associated edges that do not affect the validity of the alarm to obtain the minimum conflict subgraph and generate an evidence package. The evidence package contains the trigger constraint identifier, the minimum conflict subgraph, relevant data differences, version differences, traceability information, and verification snapshot reference. Step S700: Receive the review result corresponding to the evidence package. When the review result is an anomaly invalidation identifier, locate the corresponding verification constraint based on the trigger constraint identifier, and determine the metadata version identifier, rule version identifier, detection methodology parameters, and object type involved in the minimum conflict subgraph at the time of triggering based on the verification snapshot reference. Write the above information into the activation conditions of the verification constraint so that the verification constraint is only enabled when the metadata version identifier, rule version identifier, detection methodology parameters, and object type are consistent. When the verification result indicates an anomaly, the conditions for enabling the verification constraint remain unchanged.

[0030] Specifically, in one optional implementation, to improve the determinism of extracting the minimum conflict subgraph, the conflict localization range is first converged from the dependency subgraph to the set of objects and associated edges directly related to the verification constraint, forming a conflict subgraph. Each object / associated edge in the conflict subgraph is then associated with its evidence reference (record digest signature set), the shortest reach distance to the set of change records, and a contribution label to the alarm predicate. Subsequently, a preset elimination order is used, prioritizing objects or associated edges with "weak evidence references, greater distance from changes, and lower contribution to the alarm predicate." After each elimination, the verification constraint is quickly re-examined to determine if the verification alarm still holds. If it still holds, the elimination is confirmed; otherwise, it is rolled back. Through this "rollbackable greedy elimination" until no more items can be eliminated, the minimum conflict subgraph that makes the verification alarm hold is obtained, thus stably converging the alarm cause to the minimum evidence set and facilitating review and rectification localization.

[0031] Understandably, compared to methods that only perform general data cleaning or rule verification, this solution can bind each standardized record to its metadata version identifier and rule version identifier at the record level in scenarios involving long-term evolution of multi-source data, forming a replayable verification snapshot. This ensures the reproducibility and traceability of verification conclusions. Simultaneously, it constructs a relationship graph using record digest signatures as evidence references and introduces global constraint solving under detection fingerprint feature vectors and consistency constraints. This ensures that the link relationships between batch objects, sample objects, detection result objects, and instrument operation objects converge to a unique relationship even with multiple IDs and cross-database data, reducing mismatches and false alarms caused by ambiguous relationships. Furthermore, combined with incremental verification of dependency subgraphs and evidence package output of minimum conflict subgraphs, each alarm has an interpretable and verifiable set of causes and can accurately pinpoint the scope of rectification. Coupled with the convergence and update of activation conditions through review feedback, this continuously reduces review costs and improves alarm reliability and review efficiency.

[0032] In some embodiments of this application, obtaining a set of data sources for new traditional Chinese medicine (TCM) drugs and mapping it to a standardized record set includes: reading the field definitions, units of measurement, code table definitions, and detection methodology parameters for each data source in the TCM new drug data source set, forming version description information in a preset order, and determining the metadata version identifier based on the version description information. The mapping rules corresponding to the metadata version identifier are then invoked to perform field mapping, unit conversion, and code table conversion on the source records to obtain standardized records. Traceability information is written to each standardized record, including the source system identifier, source record identifier, and collection time.

[0033] Specifically, in some embodiments of this application, field definitions describe the name, data type, value range, and whether each field in the source record is required. Units of measurement describe the original units of numerical fields and their conversion relationship with target standard units. Code table definitions describe the encoding set of classification fields and its meaning mapping. Detection methodology parameters describe the set of parameters related to the data interpretation caliber of the detection instrument, including at least the detection method name, detection conditions, sampling frequency or sampling interval, and key parameters required for peak alignment. A preset order refers to arranging field definitions, units of measurement, code table definitions, and detection methodology parameters in a fixed concatenation order to form version description information, thereby ensuring consistent metadata version identifiers under the same data caliber. The metadata version identifier uniquely represents the data caliber status of a data source at a specific point in time and serves as an index for selecting mapping rules. Mapping rules define the correspondence between source fields and fields in the unified data model, unit conversion rules, and code table conversion rules, and correspond one-to-one with the metadata version identifier. The critical association identifier integrity check is used to determine whether the source record possesses the minimum necessary identifier set to participate in subsequent object extraction and association graph construction. The batch identifier, sample identifier, test result identifier, and instrument operation identifier are used to support the identification and interrelationship of batch objects, sample objects, test result objects, and instrument operation objects, respectively. When any critical association identifier is missing, the source record can be retained in the original data area for traceability, but it will not be included in the standardized record set to avoid introducing uncertain associations or generating unexplainable alarms in subsequent association solving and verification.

[0034] Specifically, in one embodiment, three data sources are accessed: a laboratory information management system, a quality inspection system, and a testing instrument platform. First, for each data source, the field definitions, units of measurement, and code table definitions registered in its configuration are read. Specifically, the field definitions of the laboratory information management system include at least batch identifier, sample identifier, test item, and acquisition time fields; the field definitions of the quality inspection system include at least test result identifier, test value, test unit, and judgment field; and the field definitions of the testing instrument platform include at least instrument operation identifier, instrument operation time, and original test signal reference field. Simultaneously, the corresponding testing methodology parameters for the testing instrument platform are read. Then, the field definitions, units of measurement, code table definitions, and testing methodology parameters are organized into version description information in a preset order, and a summary calculation is performed on the version description information to obtain a metadata version identifier. Based on this metadata version identifier, the corresponding mapping rules are retrieved from the mapping rule base, and field mapping is performed on each source record of each data source, filling the source fields into the target fields of the unified data model. For test value fields with unit differences, unit conversion is performed to convert them to the standard units specified by the unified data model. The classification fields are converted into standard codes as defined by the unified data model to obtain standardized records. Before generating each standardized record, a key association identifier integrity check is performed on the source record. The check rule is that the source record must be able to resolve to the batch identifier, sample identifier, test result identifier, and instrument operation identifier. If a quality inspection system record is missing a test result identifier, or a testing instrument platform record is missing an instrument operation identifier, the source record is marked as an incomplete identifier record and prevented from entering the standardized record set. For source records that pass the integrity check, traceability information is written to the generated standardized record. The traceability information includes at least the source system identifier, source record identifier, and collection time. The standardized record is then added to the standardized record set for subsequent record summary signature generation, object extraction, and association graph construction.

[0035] Understandably, introducing a one-to-one correspondence between metadata version identifiers and mapping rules during the mapping to a standardized record set stage ensures that the standardization process has identifiable and reproducible version boundaries for changes in field definitions, units, and code tables, thereby preventing inconsistent interpretations of records from the same source under different definitions. Simultaneously, implementing integrity checks on key association identifiers before data entry removes data that cannot support object associations from the standardized record set in advance, reducing ambiguous associations and unexplainable alarms during subsequent global association solving and verification processes, lowering review costs, and improving the reliability and traceability of verification results.

[0036] In some embodiments of this application, generating a record summary signature and forming a verification snapshot includes: determining a set of key fields for each standardized record; the set of key fields includes at least an identifier field corresponding to a batch object, sample object, test result object, or instrument operation object, and a field corresponding to the unit of measurement and code conversion result; normalizing the format of the field values ​​in the set of key fields and generating serialized content in a fixed field order; performing a summary calculation on the serialized content to obtain a record summary signature; and writing the record summary signature, along with the metadata version identifier, rule version identifier, and traceability information corresponding to the standardized record, into the verification snapshot.

[0037] Specifically, the key field set is used to limit the minimum range of fields involved in generating the record summary signature, ensuring that the record summary signature can stably represent the state of the record content and remain consistent with the association evidence required for subsequent object extraction and association graph construction. The identifier field includes at least one of the following: batch identifier, sample identifier, test result identifier, and instrument operation identifier, used to represent the object affiliation corresponding to the standardized record. Fields corresponding to the unit of measurement and code table conversion results refer to the numerical fields after unit conversion and the classification fields after code table conversion, used to reflect changes in data caliber in the record summary signature. Format normalization processing is used to eliminate pseudo-differences caused by different writing styles under the same semantics, including at least the uniformity of precision and null value standardization for numerical fields, the standardization of spaces and capitalization for text fields, and the uniformity of time zones and formats for date and time fields. Fixed field order is used to ensure that the same field set generates consistent serialized content at different computing nodes or different time points. Serialized content is the concatenation result of field name and field value pairs output by the key field set in a fixed field order. Verification snapshots are used to save the binding relationship between record digest signatures and metadata version identifiers, rule version identifiers, and traceability information, so that the input state and reference chain when the record participates in verification can be reproduced later based on the same version caliber.

[0038] Specifically, in one embodiment, a test result record is selected from the standardized record set as the target standardized record. A set of key fields is determined based on the record's object affiliation. This set includes batch identifier, sample identifier, test result identifier, instrument operation identifier, test item code, standard unit field for test value, test value field, and judgment result code field. The standard unit field for test value and the judgment result code field are derived from the output results of unit conversion and code table conversion, respectively. The key field set is then subjected to format normalization: the test value field is rounded according to a preset precision rule and unified to a fixed decimal place; the date and time field is converted to a unified time zone and formatted as a unified date and time string; leading and trailing spaces are removed from text fields, and empty values ​​are uniformly replaced with preset empty value markers; and coded fields are unified to the target code table encoding. Subsequently, field name and field value pairs are output in a fixed field order and concatenated to generate serialized content. A digest calculation is performed on this serialized content to obtain the record digest signature. Then, read the metadata version identifier, rule version identifier, and traceability information from the standardized record, write the record summary signature, metadata version identifier, rule version identifier, source system identifier, source record identifier, and collection time into the verification snapshot, and establish a reference relationship between the verification snapshot and the standardized record for subsequent dependent subgraph location, evidence package generation, and review playback.

[0039] Specifically, in one optional implementation, the evidence package is stored in the form of a structured object, including at least: a trigger constraint identifier, a minimum conflict subgraph structure (object list, associated edge list and their evidence references), a set of record digest signatures for the minimum conflict subgraph references, relevant data differences (difference field names, previous values, subsequent values ​​and objects / edges to which the differences belong), version differences (field definition differences / unit of measurement differences / code table definition differences and their corresponding metadata version identifier parsing results), tracing information (source system identifier, source record identifier, collection time and necessary audit markers), and a verification snapshot reference (pointing to the current verification snapshot entry for playback). The verification snapshot reference serves as a "caliber anchor" to ensure that reviewers can reproduce alarm triggering and cause location under the same metadata version identifier and rule version identifier during playback, avoiding the problem of "inconsistent interpretation of the same evidence package at different points in time" due to subsequent version drift.

[0040] Understandably, by combining a key field set, format standardization, and a fixed field order, record digest signatures can maintain stability and consistency across multiple systems, time periods, and computing nodes, avoiding false changes caused by writing differences. Simultaneously, incorporating the results of unit conversion and code table conversion into the key field set ensures that record digest signatures accurately reflect changes in data caliber. Combined with metadata version identifiers and rule version identifiers written to verification snapshots, each verification can be repeatedly replayed under the same version caliber and accurately traced back to the source record, thereby improving the reproducibility, traceability, and alarm interpretation capabilities of large-scale audits and verifications.

[0041] In some embodiments of this application, constructing the association graph includes: identifying object types and generating object identifiers from each standardized record according to preset object extraction rules, wherein the object identifiers are determined by the identifier field in the standardized record; merging standardized records with the same object identifier into the same object, and writing the corresponding record summary signature into the object evidence reference set; generating association edges between batch objects and sample objects, sample objects and test result objects, and test result objects and instrument operation objects based on the association fields in the standardized records; writing each association edge into an edge evidence reference set, which includes the record summary signature supporting the generation of the association edge; and marking association edges that do not satisfy cardinality constraints or time order constraints as candidate associations and retaining them in the association graph when generating association edges.

[0042] Specifically, preset object extraction rules are used to classify standardized records into predefined object types and generate object identifiers. Object types include at least batch objects, sample objects, test result objects, and instrument operation objects. The object type is determined by the record type field, business event field, or data source type field of the standardized record, and the object identifier is determined by the identifier field in the standardized record. The identifier field includes at least one or more of the batch identifier, sample identifier, test result identifier, and instrument operation identifier. The object evidence reference set records the evidence sources that form the object, specifically the set of record summary signatures corresponding to the standardized record merged into the object, thus allowing the object to point back to the minimum set of records supporting its existence and attributes. The association field describes the business associations between objects, such as the consistency of the sample identifier in batch records and sample records, the sample identifier carried in the test result record, and the test result identifier or sample identifier carried in the instrument operation record. The cardinality constraint limits the multiplicity boundaries of the association edges, including at least a one-to-many relationship between batch objects and sample objects, a one-to-many relationship between sample objects and test result objects, and a many-to-one or one-to-one relationship between test result objects and instrument operation objects. Temporal order constraints are used to define the temporal logical boundaries of the objects at both ends of an associated edge. At a minimum, the running time of the instrument running object must not be later than the generation time of its corresponding detection result object, and the generation time of the detection result object must not be earlier than the acquisition time of the sample object. Candidate associations are used to retain associated edges that do not meet cardinality or temporal order constraints but may still be repaired in subsequent consistency constraint solutions. These can be distinguished from determined associations so that the basis for association repair can be explained during subsequent global constraint solutions and evidence package generation.

[0043] Specifically, in one embodiment, a relationship graph is constructed using multiple records in a standardized record set. First, object types are identified according to preset object extraction rules: when a standardized record contains a batch identifier and the record type field represents a batch ledger, a batch object is generated, and the batch identifier is used as the object identifier. When a standardized record contains a sample identifier and the record type field represents a sample ledger, a sample object is generated, and the sample identifier is used as the object identifier. When a standardized record contains a test result identifier and the record type field represents an inspection result, a test result object is generated, and the test result identifier is used as the object identifier. When a standardized record contains an instrument operation identifier and the record type field represents an instrument operation, an instrument operation object is generated, and the instrument operation identifier is used as the object identifier. Second, standardized records with the same object identifier are merged into the same object. For example, multiple sample ledger records and process records carrying the same sample identifier are merged into the same sample object, and the record summary signatures corresponding to these records are written into the object evidence reference set of the sample object. Then, relationship edges are generated based on the relationship fields: the batch identifier field carried in the sample object is read, and relationship edges are generated between the batch object and the sample object. The sample identifier field carried in the detection result object is read, and an association edge is generated between the sample object and the detection result object. Similarly, the detection result identifier field or sample identifier field carried in the instrument operation object is read, and an association edge is generated between the detection result object and the instrument operation object. When generating each association edge, the record summary signature of the standardized record supporting the generation of that association edge is simultaneously written into the edge evidence reference set, ensuring that each edge has a traceable evidence reference. Finally, constraint checks are performed on the newly generated association edges: if the same detection result object corresponds to multiple instrument operation objects exceeding a preset cardinality boundary, or if the generation time of the detection result object is earlier than the acquisition time of the sample object, then the association edge is marked as a candidate association and retained in the association graph. For association edges that satisfy both the cardinality constraint and the time order constraint, they are marked as definitive associations and retained in the association graph for subsequent dependency subgraph expansion and incremental verification.

[0044] Understandably, introducing object evidence reference sets and edge evidence reference sets when constructing the relationship graph allows both objects and related edges to point back to the record digest signatures that support their generation. This elevates the relationship between objects from a traditional field concatenation relationship to a verifiable and documentable chain of evidence. Simultaneously, related edges that do not meet cardinality or temporal order constraints are retained as candidate relationships. This allows subsequent global constraint solving to repair relationships while preserving evidence, avoiding information loss due to direct deletion, and reducing unexplainable mismatches and false alarms caused by common issues such as multiple numbering across systems and inconsistent temporal sequences. This improves the stability and traceability of large-scale review and verification.

[0045] In some embodiments of this application, acquiring detection instrument data and extracting detection fingerprint feature vectors includes: reading instrument operation records and detection result records from the detection instrument data, and extracting detection methodology parameters. Peak identification is performed on the original detection signal corresponding to the instrument operation record to obtain a peak feature set. The peak feature set is aligned based on the detection methodology parameters, and the aligned peak feature set is encoded into a detection fingerprint feature vector in a fixed order. The detection fingerprint feature vector is written into the association information of the instrument operation object and the detection result object, respectively.

[0046] In some embodiments of this application, constructing the evidence vector and generating entity matching candidate pairs includes: extracting the identifier elements and attribute summaries of the objects to be matched, and obtaining the source weights and detection fingerprint feature vectors corresponding to the objects to be matched. Consistency comparisons are performed on the identifier elements, attribute summaries, and detection fingerprint feature vectors to obtain identifier consistency results, attribute consistency results, and fingerprint consistency results. The identifier consistency results, attribute consistency results, and fingerprint consistency results are merged with the source weights to form the evidence vector. Entity matching candidate pairs are generated based on the matching scores of the evidence vectors and the consistency conditions of the detection methodology parameters.

[0047] In some embodiments of this application, when solving global constraints and updating the association graph under consistency constraints, the process includes: constructing a candidate association set based on entity matching candidate pairs, the candidate association set covering candidate associations from batch objects to sample objects, from sample objects to detection result objects, and from detection result objects to instrument operation objects. Consistency constraints are applied to the candidate association set, including at least uniqueness constraints, cardinality constraints, temporal order constraints, and methodological consistency constraints. Under the consistency constraints, the candidate association combination with the largest total matching score is selected as the unique association. The unique association is written into the association graph and replaces the corresponding candidate association.

[0048] Specifically, the detection instrument data includes at least the instrument operation record, the detection result record, and the raw detection signal associated with the instrument operation record. The raw detection signal can be a spectrum, chromatographic curve, or other signal sequence that varies with time or wavelength. Detection method parameters are used to define the interpretation and alignment scope of the raw detection signal, and at least include the detection method name, detection conditions, sampling interval, target peak range, or alignment window parameters. Peak identification is used to locate peak positions and extract peak features in the raw detection signal. Peak features include at least one of peak position, peak height, or peak area. Alignment processing is used to eliminate peak position drift between different batches or different instrument operations, ensuring that peaks corresponding to the same component fall into a consistent alignment position. Fixed-order coding is used to output the aligned peak feature set according to a preset peak order or preset alignment position, thereby obtaining a comparable detection fingerprint feature vector. Identification elements are used to characterize the explicit identity information of the object, and at least include at least one of batch identifier, sample identifier, detection result identifier, instrument operation identifier, and test item code. Attribute summaries are used to characterize the set of key attributes of the object, and at least include at least one of acquisition time, detection time, test item, detection value range, or judgment result. Source tracing weights are used to quantify the credibility or priority of data sources, determined at least based on source system identifiers and source record identifiers, and can also be determined in conjunction with the freshness of the collection time or the data receipt status. Evidence vectors are used to merge identifier consistency results, attribute consistency results, fingerprint consistency results, and source tracing weights into a unified matching evidence carrier. The consistency condition for detection methodology parameters limits the use of fingerprint consistency results for matching score calculation only when the detection method name and key methodology parameters are consistent. The candidate association set is a set of candidate object associations derived from entity matching candidate pairs. Uniqueness constraints, cardinality constraints, temporal order constraints, and detection methodology consistency constraints are used to filter globally consistent association combinations from the candidate association set. Evidence vector reference information records the evidence vector identifiers supporting the existence of candidate associations, allowing for referencing specific matching evidence during subsequent generation of minimum conflict subgraphs and evidence packages.

[0049] Specifically, the traceability weight uses configurable discrete levels or normalized values ​​to quantify the data credibility and priority corresponding to different source system identifiers. It only participates in the matching and ranking as a "score weighting item" in the evidence vector and does not replace the hard judgment of consistency constraints. For example, the source system identifier can be mapped to a basic credibility level (e.g., quality inspection system is higher than manual entry system, raw signal from testing instrument platform is higher than summary table), and the basic credibility level is adjusted by combining the source record receipt status, collection time freshness, or audit pass markers to finally obtain the traceability weight. During evidence vector synthesis, the identifier consistency result, attribute consistency result, and fingerprint consistency result output comparable sub-scores or consistency / inconsistency labels respectively. If the detection methodology parameters do not meet the consistency conditions, the fingerprint consistency result is set as an unparticipated item, thereby avoiding false alarms and mismatches caused by fingerprint incomparability when methodological standards are inconsistent.

[0050] Specifically, in one embodiment, the following process is performed for multiple samples and corresponding test results under the same application batch. First, instrument operation records and test result records are read from the testing instrument data, along with the testing methodology parameters. Then, peak identification is performed on the original test signal corresponding to each instrument operation record to obtain a peak feature set. Alignment processing is then performed on the peak feature set based on the alignment window defined in the testing methodology parameters to obtain an aligned peak feature set. Next, the aligned peak position and peak intensity information are output according to a preset peak order, encoded to generate a detection fingerprint feature vector, and this detection fingerprint feature vector is written into the association information of the instrument operation object corresponding to the instrument operation record and the detection result object corresponding to the test result record. Then, evidence construction is performed on the pair of objects to be matched: the identifier elements and attribute summaries of the objects to be matched are extracted, and their source information is read to determine the source weight. Simultaneously, their detection fingerprint feature vectors are read. Consistency comparison is performed on the identifier elements, and the identifier consistency result is output. Consistency comparison is performed on the attribute summaries, and the attribute consistency result is output. Consistency comparison is performed on the detection fingerprint feature vectors, and the fingerprint consistency result is output. When the detection methodology parameters meet the consistency conditions, the fingerprint consistency results, identifier consistency results, attribute consistency results, and traceability weights are merged to form an evidence vector, and a matching score is calculated. Based on object pairs whose matching scores reach the candidate threshold, entity matching candidate pairs are generated, and a candidate association set covering batch objects to sample objects, sample objects to detection result objects, and detection result objects to instrument operation objects is constructed. Subsequently, consistency constraints are applied to the candidate association set: for each type of association, at least a uniqueness constraint is required, ensuring that the same detection result object is associated with only one instrument operation object, and the same sample object is associated with only one batch object. Simultaneously, a cardinality constraint is required to meet a preset multiplicity range, and a time sequence constraint is required, ensuring that the instrument operation object's running time is no later than the detection result object's generation time and the detection result object's generation time is no earlier than the sample object's collection time. Furthermore, a detection methodology consistency constraint is required, ensuring that the detection methodology parameters of the detection result object and the instrument operation object are consistent. Under the premise of satisfying the consistency constraints, the candidate association combination with the largest sum of matching scores is selected from the candidate association set as the unique association, written into the association graph, and replacing the corresponding candidate association. For candidate associations that are not selected and violate consistency constraints, they are retained as candidate associations, and their evidence vector reference information is recorded so that when extracting the minimum conflict subgraph and generating evidence packages for subsequent verification alarms, we can trace back why the candidate association was excluded.

[0051] Understandably, transforming the original detection signal into a comparable detection fingerprint feature vector, and introducing fingerprint consistency evidence into entity matching under the premise of consistent detection methodology parameters, improves object matching in scenarios such as cross-system multi-numbering and cross-database file association from relying on text numbering to joint verification of numbering and detection fingerprint, reducing the probability of mismatch. Simultaneously, by merging identifiers, attributes, fingerprints, and source traceability credibility into an evidence vector, and performing global constraint solving under constraints such as uniqueness, cardinality, temporal order, and consistency of detection methodology, the final association result is not only a locally optimal match but also a globally consistent unique association, enhancing the stability and interpretability of the association graph. Furthermore, retaining candidate associations that violate constraints and recording evidence vector reference information allows subsequent anomaly diagnosis to trace the evidence and reasons for excluded associations, thereby reducing review costs and improving evidence issuance capabilities.

[0052] In some embodiments of this application, when the standardized record set changes and a verification alarm set is obtained within the dependency subgraph, the process includes: determining the changed record set based on source information. The changed record set includes a record digest signature corresponding to a newly added source record identifier, and a record digest signature that has the same source system identifier and source record identifier as the previous verification snapshot but a different record digest signature. The process involves retrieving the object evidence reference set and edge evidence reference set referencing the changed record set from the relationship graph, and determining the seed object set and seed associated edge set. Starting from the seed object set and seed associated edge set, the process extends outward along the associated edges of the relationship graph to a preset number of layers to obtain the dependency subgraph. Within the dependency subgraph, the corresponding verification constraint is selected according to the rule version identifier, and the objects referencing the changed record set and their associated edges are verified to obtain the verification alarm set. For each verification alarm, a trigger constraint identifier, the associated record digest signature, and the corresponding metadata version identifier and rule version identifier are written.

[0053] Specifically, a change in the standardized record set refers to a situation where, under the constraints of the same source system identifier and the same source record identifier, a new source record is added or the content of a source record is modified, resulting in a discrepancy between the record summary signature of the corresponding standardized record and the previous verification snapshot. The changed record set is used to encompass the range of records requiring incremental verification. The record summary signature corresponding to a newly added source record identifier covers the risk introduced by new data, while record summary signatures with the same source system identifier and source record identifier but different record summary signatures cover the risk of revisions to the same source record. The object evidence reference set and the edge evidence reference set are used to establish traceable reference relationships between objects or associated edges and the record summary signatures that support their generation, thus enabling rapid location of affected objects and associated edges from the relationship graph. The seed object set and the seed associated edge set refer to the minimum affected sets referencing the changed record set, serving as the starting point for dependent subgraph expansion. The preset layer number is used to limit the expansion radius to control the incremental verification range. The preset layer number can be set according to the object hierarchy structure to cover the complete link of batch objects, sample objects, test result objects, and instrument operation objects. The rule version identifier selects the verification constraint to ensure that incremental verification uses the same set of verification constraints as the current verification when different rule versions coexist or when rules are iterated and updated. The trigger constraint identifier is used to uniquely locate the verification constraint entry that triggers the alarm, which facilitates subsequent extraction of the minimum conflict subgraph, generation of evidence packages, and updating of the activation conditions for review feedback.

[0054] Specifically, in one embodiment, a full verification has been completed, and a snapshot of the previous verification has been generated. Subsequently, the quality inspection system revises a source record for a specific test result, with changes to the test value field or judgment field, and adds a new source record for the same batch of test results. After receiving incremental data, the system first aligns the source system identifier and source record identifier based on the traceability information: for a newly added source record identifier, a corresponding standardized record is generated, and its record summary signature is calculated and added to the change record set. For a revised source record identifier, the system reads the historical record summary signature corresponding to the same source system identifier and the same source record identifier from the previous verification snapshot and compares it with the newly generated record summary signature. If they are inconsistent, the current record summary signature is added to the change record set. Then, the system searches the relationship graph for the object evidence reference set and the edge evidence reference set that reference the change record set: for example, if the object evidence reference set of a test result object contains the record summary signature from the change record set, the test result object is added to the seed object set. If the edge evidence reference set of the association edge between the sample object and the test result object contains the record summary signature, the association edge is added to the seed association edge set. Starting with the seed object set and the seed associated edge set, the dependency subgraph is expanded outward along the associated edges of the relationship graph to a preset number of layers. When the preset number of layers is set to cover the entire link, the dependency subgraph will at least include the sample objects, batch objects, and instrument operation objects connected to the detection result objects, as well as their associated edges. Within the dependency subgraph, corresponding verification constraints are selected based on the rule version identifier, and verification is performed on the objects referencing the change record set and their associated edges. For example, consistency verification between sample objects and detection result objects, methodological consistency verification between detection result objects and instrument operation objects, and time sequence verification are performed to obtain a set of verification alarms. Each verification alarm is written with a trigger constraint identifier, an associated record summary signature, and the corresponding metadata version identifier and rule version identifier, so that the alarm triggering process can be replayed and the affected records can be located under the same caliber in the future.

[0055] Specifically, in one optional implementation, validation constraint entries are stored in the rule base indexed by rule version identifiers. Each validation constraint contains at least two parts: "enabling condition" and "validation condition." The enabling condition limits the applicable scope of the validation constraint's execution, while the validation condition limits the judgment logic for alarm validity after enabling. When the received review result is an anomaly invalidation flag, the validation snapshot reference in the evidence package points back to the current validation snapshot entry. The metadata version identifier, rule version identifier, and related tracing information bound when the validation constraint was triggered are read. Combined with the detection methodology parameters related to the alarm and the object types involved in the minimum conflict subgraph, the enabling condition is updated convergently, ensuring that the validation constraint only participates in subsequent validation when the metadata version identifier, rule version identifier, detection methodology parameters, and object types are consistent. When the review result is an anomaly validation flag, the enabling condition is not modified, allowing it to continue executing within its original applicable scope. This achieves a closed-loop optimization of "reducing false alarms through enabling conditions without changing the rule semantics."

[0056] Understandably, aligning source identifiers and source record identifiers with traceability information, and then determining changes based on differences in record summary signatures, provides a clear and reproducible criterion for change identification. Simultaneously, using object evidence reference sets and edge evidence reference sets to locate seed object sets and seed associated edge sets in the relationship graph, and expanding to a predetermined number of layers to form a dependency subgraph, the system executes verification constraints consistent with the rule version identifier only within the dependency subgraph. This precisely limits the verification scope to the affected links, avoiding the computational overhead and alarm propagation caused by full recalculation. Furthermore, triggering constraint identifiers and version identifiers for each alarm record ensures consistent caliber and traceable basis for subsequent alarm interpretation, evidence package generation, and review closure, reducing verification costs and improving alarm credibility.

[0057] In some embodiments of this application, the process of extracting the minimum conflict subgraph from the dependency subgraph and generating an evidence package includes: for each verification alarm, determining the set of objects and the set of associated edges related to the verification constraint in the dependency subgraph based on the trigger constraint identifier, and constructing a conflict subgraph using the set of objects and the set of associated edges. While maintaining the validity of the verification alarm, gradually eliminating objects or associated edges in the conflict subgraph that do not affect the validity of the verification alarm to obtain the minimum conflict subgraph that makes the verification alarm valid. An evidence package is generated, which includes at least the trigger constraint identifier, the minimum conflict subgraph, the record digest signature referenced by the minimum conflict subgraph, relevant data differences extracted from the standardized record set, version differences determined by the metadata version identifier, source information, and a verification snapshot reference.

[0058] Specifically, the trigger constraint identifier is used to uniquely locate the verification constraint entry that triggers the verification alarm and to determine the object type and associated edge type associated with the verification constraint, thereby limiting the scope of conflict localization in the dependency subgraph. The conflict subgraph is a subgraph within the dependency subgraph consisting of the set of objects and associated edges associated with the verification constraint corresponding to the trigger constraint identifier, containing all candidate causal elements that can cause the verification alarm to be established. The minimum conflict subgraph is the smallest set of causes obtained by removing objects or associated edges that do not affect the establishment of the alarm from the conflict subgraph, while ensuring the verification alarm is still established; it is used to converge the alarm cause to the minimum range of evidence. Related data differences are used to characterize the inconsistencies in key fields between the standardized records referenced by the minimum conflict subgraph, including at least the difference field name, the previous value, and the subsequent value. Version differences are used to characterize the changes in scope that caused the differences, including at least one of field definition differences, unit of measurement differences, or code table definition differences, and are obtained by parsing the version description information corresponding to the metadata version identifier. The verification snapshot reference is used to indicate the snapshot entry of the record summary signature and metadata version identifier, rule version identifier and source information bound when the alarm was generated, so that the evidence package can replay the alarm triggering process and reproduce the cause location under the same version caliber.

[0059] Specifically, in one embodiment, after executing a verification constraint within the dependency subgraph, a verification alarm is generated. The verification constraint corresponding to the trigger constraint identifier is that the inspection item code of the detection result object is inconsistent with the detection methodology parameters of the instrument operation object. First, based on the trigger constraint identifier, objects and associated edges related to the verification constraint are filtered within the dependency subgraph: the object set includes at least the detection result object involved in the alarm and its associated instrument operation object, and if necessary, also includes sample objects associated with the detection result object to provide time and traceability context. The associated edge set includes at least the associated edges between the detection result object and the instrument operation object, and a set of edge evidence references supporting the generation of these associated edges. A conflict subgraph is constructed using the object set and the associated edge set, and a step-by-step elimination process is performed within the conflict subgraph: first, an attempt is made to eliminate sample objects and their associated edges, and then it is re-determined whether the alarm still holds. If the alarm still holds, it is confirmed that the sample object does not affect the alarm's validity and it is removed from the conflict subgraph. Next, attempts are made to remove other test result objects or other instrument operation objects unrelated to the alarm. After each removal, it is re-evaluated whether the verification constraint corresponding to the trigger constraint identifier is still violated, until no more objects or associated edges can be removed and the alarm is still established, resulting in a minimum conflict subgraph. Then, an evidence package is generated: the trigger constraint identifier and the minimum conflict subgraph are written into the evidence package, and the referenced record summary signatures are extracted from the object evidence reference set and edge evidence reference set of the minimum conflict subgraph. Then, the standardized records corresponding to the record summary signatures are read from the standardized record set, and relevant data differences are extracted, such as the inconsistency between the test item coding field value of the test result object and the test methodology parameter field value of the instrument operation object. Simultaneously, based on the metadata version identifier of the standardized records involved in the minimum conflict subgraph, the corresponding version description information is parsed, and version differences are extracted, such as code table definition updates causing the same test item to have different codes in different versions. Furthermore, the traceability information of the relevant standardized records is written into the evidence package, and a verification snapshot reference is written to point to the snapshot entry corresponding to the time the alarm was generated, facilitating the reviewer to replay the alarm triggering process and cause localization process under the same metadata version identifier and rule version identifier caliber.

[0060] Understandably, by using the trigger constraint identifier as the entry point, the alarm localization scope is converged from the entire dependency subgraph to the conflict subgraph directly related to the verification constraint. By gradually eliminating irrelevant objects and related edges while maintaining the alarm's validity, the smallest conflict subgraph is obtained. This allows the alarm cause to be presented in the form of a minimal evidence set, avoiding the common problem of an excessively large impact range but unclear causes. Simultaneously, the evidence package includes record summary signatures, data differences, version differences, traceability information, and verification snapshot references. This ensures that the alarm is not only interpretable and traceable but can also be reproduced and verified under the same criteria, thereby reducing review costs, improving the accuracy of rectification localization, and enhancing the ability to provide evidence during the review and verification process.

[0061] In some embodiments of this application, receiving review feedback and updating the verification enable conditions includes: receiving a review result corresponding to the evidence package, the review result including an anomaly establishment flag or an anomaly non-establishment flag. In the case of an anomaly non-establishment flag, the verification constraint corresponding to the trigger constraint flag is read, and the metadata version flag, rule version flag, detection methodology parameters, and object type involved in the minimum conflict subgraph are determined based on the verification snapshot reference. The metadata version flag, rule version flag, detection methodology parameters, and object type are written into the verification constraint enable conditions, so that the verification constraint is enabled only when the metadata version flag, rule version flag, detection methodology parameters, and object type are consistent. In the case of an anomaly establishment flag, the verification constraint enable conditions remain unchanged.

[0062] Specifically, the review result refers to the conclusion data formed by the reviewer after verifying the verification alarm based on the evidence package. It includes at least an anomaly establishment indicator or an anomaly non-establishment indicator to distinguish whether the alarm is a genuine anomaly. The trigger constraint identifier is used to uniquely locate the verification constraint entry that generated the alarm. The verification constraint refers to a specific entry in the set of verification rule entries indexed by the rule version identifier. Each verification constraint includes an activation condition and a verification condition. The activation condition is used to limit the data scope and object scenario in which the verification constraint participates in the execution, and the verification condition is used to limit how to determine whether the alarm is established after activation. The verification snapshot reference is used to point back to the verification snapshot entry when the alarm was generated, so that the metadata version identifier, rule version identifier, and traceability information and record summary signature binding relationship used in the current verification can be read. The detection methodology parameters are used to characterize the caliber parameters of fingerprint comparison and methodology consistency verification. The object type is used to characterize the range of object categories involved in the minimum conflict subgraph, such as detection result objects, instrument operation objects, sample objects, or batch objects. By writing metadata version identifiers, rule version identifiers, detection methodology parameters, and object types into the enable conditions, certain verification constraints that are prone to false alarms under specific versions or methodologies can be limited to a more precise scope of application, thereby preventing the verification constraint from triggering invalid alarms in mismatched scenarios or object scenarios.

[0063] Specifically, in one embodiment, after verification within the dependency subgraph, a verification alarm is generated, and a corresponding evidence package is formed. Upon opening the evidence package, the reviewer discovers that the root cause of the alarm is a difference in the recording method of a certain detection methodology parameter across different instrument models. This difference is not yet covered under the current rule version identifier, causing the alarm to be an invalid anomaly. Upon receiving the review results, the corresponding verification constraint entry is first located based on the trigger constraint identifier in the evidence package, and its existing enabling conditions are read. Then, based on the verification snapshot reference in the evidence package, the verification snapshot entry for the current verification is referenced, and the metadata version identifier and rule version identifier bound when the verification constraint is triggered are read. The detection methodology parameter is then read from the standardized records related to the alarm. Simultaneously, the minimum conflict subgraph is read, determining that the object types involved in this minimum conflict subgraph are detection result objects and instrument operation objects. The aforementioned metadata version identifier, rule version identifier, detection methodology parameters, and object type are written into the enable conditions of this validation constraint entry, forming a limited enable condition. This ensures that the validation constraint will only participate in subsequent validations if the metadata version identifier, rule version identifier, detection methodology parameters, and object type are consistent. When objects with different metadata version identifiers or different detection methodology parameters appear in subsequent validation processes, the enable condition is determined not to be met, and the validation constraint is skipped to avoid repeatedly triggering the same type of false alarm. If the review result indicates an anomaly is established, the enable conditions of this validation constraint entry are not modified, allowing it to continue to execute within its original applicable scope.

[0064] Understandably, feeding back the review conclusions into the activation conditions of the verification constraints in an actionable manner, and updating the activation conditions based on the trigger constraint identifier to locate specific rule entries, restoring the trigger scope and version environment based on the verification snapshot reference, and determining the object type range based on the minimum conflict subgraph, thus providing clear boundaries for false alarm governance and avoiding the risk of missed alarms caused by simply lowering thresholds or disabling rules. Furthermore, since the activation conditions are bound to the metadata version identifier and the rule version identifier, consistency and traceability can be maintained during rule iterations or scope changes. This allows the verification rules to gradually converge to a stable state with fewer false alarms, explainability, and reproducibility in large-scale data review scenarios, reducing review costs and improving alarm credibility.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. An intelligent method for large-scale data review and verification of new traditional Chinese medicine drugs, characterized in that, include: Obtain the data source set of new traditional Chinese medicine drugs and map it into a standardized record set; A record digest signature is generated for each standardized record in the standardized record set. The record digest signature is then bound to the corresponding metadata version identifier and rule version identifier to form a verification snapshot. Based on the standardized record set, batch objects, sample objects, test result objects and instrument operation objects are extracted, and a relationship graph is constructed using the record summary signature as the association basis. Data from the testing instrument is obtained from the standardized record set, and the detection fingerprint feature vector is extracted. An evidence vector is constructed based on the identification elements, attribute summaries, traceability weights, and the detection fingerprint feature vector. Entity matching candidate pairs are generated, and global constraint solutions are performed under consistency constraints to obtain unique associations between batch objects and sample objects, sample objects and test result objects, and test result objects and instrument operation objects. The association graph is updated based on the unique associations. When the standardized record set changes, a dependency subgraph is obtained by expanding the relationship graph, and the corresponding verification constraint is selected according to the rule version identifier in the dependency subgraph for verification to obtain a set of verification alarms. For each verification alarm in the verification alarm set, the minimum conflict subgraph that makes the verification alarm valid is extracted from the dependency subgraph, and an evidence package is generated. The evidence package includes at least the trigger constraint identifier, related data differences, version differences, source information, and the verification snapshot reference. Receive review feedback and update the enabling conditions of the verification constraint based on the trigger constraint identifier; When constructing a relationship graph, the following is included: According to the preset object extraction rules, the object type is identified from each standardized record and an object identifier is generated. The object identifier is determined by the identifier field in the standardized record. Standardized records with the same object identifier are grouped into the same object, and the corresponding record summary signature is written into the object evidence reference set. Based on the association fields in the standardized records, association edges are generated between batch objects and sample objects, sample objects and test result objects, and test result objects and instrument operation objects, respectively. Each association edge is written into an edge evidence reference set, which includes the record summary signature that supports the generation of the association edge. When generating the association edge, association edges that do not meet the cardinality constraint or time order constraint are marked as candidate associations and retained in the association graph. When acquiring data from the detection instrument and extracting the fingerprint feature vector, the following steps are included: Read the instrument operation record and test result record from the data of the testing instrument, and extract the testing methodology parameters; Peak identification is performed on the original detection signal corresponding to the instrument operation record to obtain a peak feature set; the peak feature set is aligned based on the detection methodology parameters, and the aligned peak feature set is encoded into a detection fingerprint feature vector in a fixed order; The detected fingerprint feature vectors are respectively written into the association information between the instrument operation object and the detection result object; When constructing evidence vectors and generating entity matching candidate pairs, the following steps are included: Extract the identifier elements and attribute summaries of the objects to be matched, and obtain the source weights and detection fingerprint feature vectors corresponding to the objects to be matched; The identification elements, attribute summaries, and detection fingerprint feature vectors are compared for consistency to obtain identification consistency results, attribute consistency results, and fingerprint consistency results; the identification consistency results, attribute consistency results, and fingerprint consistency results are then combined with the source tracing weight to form an evidence vector; The entity matching candidate pairs are generated based on the matching score of the evidence vector and the consistency condition of the detection methodology parameters; When solving global constraints and updating the association graph under consistency constraints, the process includes: Based on the entity matching candidate pairs, a candidate association set is constructed, which covers candidate associations from batch objects to sample objects, from sample objects to test result objects, and from test result objects to instrument operation objects. The consistency constraints are applied to the candidate association set, and the consistency constraints include at least uniqueness constraints, cardinality constraints, temporal order constraints, and detection methodology consistency constraints. Under the consistency constraint, the candidate association combination with the largest total matching score is selected as the unique association; the unique association is written into the association graph and the corresponding candidate association is replaced. When the standardized record set changes and a verification alarm set is obtained within the dependency subgraph, it includes: Based on the source tracing information, a set of change records is determined. The set of change records includes the record digest signature corresponding to the newly added source record identifier, and the record digest signature that has the same source system identifier and source record identifier as the previous verification snapshot but has a different record digest signature. In the relationship graph, retrieve the object evidence reference set and edge evidence reference set that reference the change record set to determine the seed object set and the seed association edge set; Starting from the set of seed objects and the set of seed associated edges, the dependency subgraph is obtained by expanding outward along the associated edges of the association graph to a preset number of layers. Within the dependency subgraph, the corresponding verification constraint is selected according to the rule version identifier, and the object referencing the change record set and its associated edges are verified to obtain a set of verification alarms. For each verification alarm, a trigger constraint identifier, an associated record summary signature, and the corresponding metadata version identifier and rule version identifier are written. When extracting the minimum conflict subgraph from the dependency subgraph and generating the evidence package, the process includes: For each verification alarm, the set of objects and the set of associated edges related to the verification constraint are determined in the dependency subgraph according to the trigger constraint identifier, and the set of objects and the set of associated edges are used to form a conflict subgraph; While ensuring the verification alarm is valid, gradually eliminate objects or associated edges in the conflict subgraph that do not affect the validity of the verification alarm to obtain the minimum conflict subgraph that makes the verification alarm valid. Generate an evidence package, which includes at least the trigger constraint identifier, the minimum conflict subgraph, the record digest signature referenced by the minimum conflict subgraph, relevant data differences extracted from the standardized record set, version differences determined by the metadata version identifier, the tracing information, and the verification snapshot reference; When receiving review feedback and updating the verification enable conditions, including: Receive the review result corresponding to the evidence package, the review result including an anomaly establishment indicator or an anomaly non-establishment indicator; When the exception does not meet the identifier, the verification constraint corresponding to the trigger constraint identifier is read, and the metadata version identifier, rule version identifier, detection methodology parameters, and object type involved in the minimum conflict subgraph are determined according to the verification snapshot reference. The metadata version identifier, rule version identifier, detection methodology parameters, and object type are written into the activation condition of the verification constraint, so that the verification constraint is only enabled when the metadata version identifier, rule version identifier, detection methodology parameters, and object type are consistent. When the exception meets the identifier, the activation condition of the verification constraint remains unchanged. The rule version identifier is used to characterize the version of the verification constraint used when performing verification on the corresponding standardized record, and the verification snapshot saves the binding relationship between the record digest signature and the metadata version identifier, the rule version identifier and the traceability information; For objects and their associated edges that reference the change record set within the dependency subgraph, a corresponding verification snapshot is determined based on the record digest signature in the object evidence reference set or the edge evidence reference set. The rule version identifier bound to the record digest signature in the verification snapshot is read, and the corresponding verification constraint is selected based on the rule version identifier.

2. The intelligent method for large-scale data review and verification of new traditional Chinese medicine drugs according to claim 1, characterized in that, When obtaining a collection of new data sources for traditional Chinese medicine and mapping it to a standardized record set, the following is included: For each data source in the set of new Chinese medicine drug data sources, the field definition, unit of measurement, code table definition and detection methodology parameters are read, and version description information is formed in a preset order. The metadata version identifier is determined based on the version description information. The mapping rules corresponding to the metadata version identifier are invoked to perform field mapping, unit conversion, and code table conversion on the source record to obtain a standardized record; For each of the standardized records, traceability information is written, which includes the source system identifier, the source record identifier, and the collection time.

3. The intelligent method for large-scale data review and verification of new traditional Chinese medicine drugs according to claim 1, characterized in that, When generating a record digest signature and creating a verification snapshot, the process includes: A set of key fields is determined for each standardized record. The set of key fields includes at least an identifier field corresponding to the batch object, sample object, test result object, or instrument operation object, as well as a field corresponding to the unit of measurement and code conversion result. The field values ​​of the key field set are formatted and serialized content is generated according to a fixed field order. Perform digest calculation on the serialized content to obtain the record digest signature; The record digest signature, along with the metadata version identifier, rule version identifier, and traceability information corresponding to the standardized record, are written into the verification snapshot.

Images