A method and system for compiling and quality node-based data source graphs for physical examinations.

By generating immutable examination source snapshots and parsing the skeleton of the source graph of the examined objects, configuring field addresses and content hashes, performing concept mapping and unit conversion, and verifying quality rules, the accuracy problems of field mapping and quality control in multi-departmental physical examination data are solved, and hierarchical verification of complex patterns and hierarchical representation of data quality problems are realized.

CN122816681BActive Publication Date: 2026-11-10XUNKANG INFORMATION TECH (SHENZHEN) CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611327458.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-08-31
Publication Date
2026-11-10
Estimated Expiration
2046-08-31

AI Technical Summary

Technical Problem

Existing technologies struggle to balance field mapping, concept layer candidate coverage, and quality control in complex heterogeneous data processing, resulting in insufficient accuracy of the quality status view.

Method used

By generating an immutable inspection source snapshot, parsing the source graph skeleton of the inspected object, configuring field addresses and content hashes, performing concept mapping and unit conversion, and verifying quality rules, an initial source graph and quality status view of the inspected object are formed, ensuring the hierarchical representation and interpretable matching of data quality issues.

Benefits of technology

It enables hierarchical validation of complex patterns and layered representation of data quality issues while maintaining the traceability of field sources, thereby improving the accuracy and interpretability of the quality status view and reducing the possibility of unauthorized mapping.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122816681B_ABST
    Figure CN122816681B_ABST
Patent Text Reader

Abstract

This invention discloses a method and system for compiling a source graph and assigning quality nodes to physical examination data. The method generates an invariant physical examination source snapshot and a skeleton of the subject's source graph based on collected multi-departmental structured physical examination data. It configures field addresses, content hashes, source versions, and node identifiers for result fields, resulting in a standardized set of observation nodes containing both original and standardized values. It then performs summary detail merging, quality rule verification, and deterministic discovery generation, establishing source relationships, constraint relationships, blocking relationships, or contradictory relationships to form an initial source graph and quality status view of the subject. Hierarchical pattern compilation, bounded one-hop mapping trajectory, and quality shape coverage assessment can be combined with the main process to improve the compatibility, interpretability, and accuracy of quality status mapping for complex fields.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of medical information data processing and heterogeneous data pattern compilation technology, and in particular to a method and system for performing field-level source anchoring, standard concept mapping, quality constraint verification, and source graph construction on collected multi-departmental physical examination structured data. Background Technology

[0002] The physical examination information processing system typically receives structured data from different departments, including laboratory tests, imaging, functional examinations, and specialist examinations. Because different institutions differ in their interface models, field hierarchies, item naming, unit representation, and summary detail organization, data source tracing, item standardization, and quality status representation must be completed before the data is provided to the final review, quality control, or report generation stages.

[0003] Existing solutions typically use fixed field mapping tables to convert local fields into standard fields and rely on project dictionaries to unify project names, units, and reference ranges. For situations such as missing fields, abandoned projects, inconsistencies between summaries and details, or unavailable sources, the system generally generates error logs or manual verification lists. Fixed mapping tables are suitable for structurally stable one-to-one field relationships, but when faced with multi-table, multi-level, or composite fields whose meanings are jointly defined by multiple attributes, they easily create a large direct comparison space between all fields and all standard concepts; furthermore, the overall similarity results struggle to explain the specific basis for retaining or excluding each candidate.

[0004] In existing technologies, source graph processing involves parsing nested physical examination data into examined subjects, cycles, departments, examination groups, and original observation nodes. Field addresses and content hashes are stored for observation nodes, and missing, abandoned, unavailable, and conflicting data are converted into quality issue nodes. Summary records are then associated with detailed observations. While this method can preserve field-level sources and form an initial source graph, its concept mapping and summary-to-detail merging may still rely on one-time composite matching; its quality rules are also mostly discrete inspection items, lacking a unified representation of the rule's scope and uncovered structures.

[0005] Therefore, existing technologies have at least the following limitations: On the one hand, field mapping in complex heterogeneous patterns is difficult to simultaneously take into account candidate coverage at the concept layer, compatibility of hard attributes at the field layer, and the reproducibility of the mapping process. Soft similarity may also exceed necessary constraints such as unit dimension, sample type, or detection method. On the other hand, existing quality control results usually only indicate whether the verification has passed or failed, making it difficult to distinguish whether the data has indeed passed the inspection, whether the data inspection failed, or whether the corresponding structure was not inspected by the rules at all, thus affecting the accuracy of the quality status view. Summary of the Invention

[0006] The technical problem to be solved by this invention is to provide a method and system for compiling and quality node-based data source graphs for physical examinations, which can support hierarchical verification and bounded interpretable matching for complex patterns, and can hierarchically represent data quality problems and insufficient rule coverage, while maintaining the traceability of the source of physical examination fields and ensuring that the original information is not covered.

[0007] To address the aforementioned technical problems, this invention provides a method for compiling and quality nodeizing a source graph of physical examination data, comprising: generating an invariant physical examination source snapshot based on the collected multi-departmental structured physical examination data, and parsing the invariant physical examination source snapshot to obtain a source graph skeleton of the examined object; configuring field addresses, content hashes, source versions, and node identifiers for the result fields based on the source graph skeleton of the examined object, obtaining an original set of observation nodes, performing concept mapping, unit conversion, and reference range structuring on the original set of observation nodes, obtaining a standardized set of observation nodes with both original and standardized values; performing summary detail merging and quality rule verification based on the standardized set of observation nodes, obtaining the source relationship between the summary and the detail and a set of quality problem nodes; generating a deterministic discovery set that does not exceed the semantics of the original results based on the standardized set of observation nodes, and establishing restrictive, blocking, or contradictory relationships by combining the source relationships and the set of quality problem nodes to form an initial source graph and quality status view of the examined object.

[0008] Further, the step of performing concept mapping, unit conversion, and reference range structuring on the original set of observation nodes to obtain a standardized set of observation nodes containing both original and standardized values ​​includes: rolling up field fragments in the original set of observation nodes into local concept clusters according to parent path, department, inspection group, field level, and detection method to obtain a set of local concept clusters; performing candidate recall on the concept cluster descriptions in the set of local concept clusters with standard item families, institutional item dictionaries, and existing mapping templates to obtain a set of standard concept candidates; drilling down the candidate concepts in the set of standard concept candidates to the corresponding result fields and performing hard constraint verification on the fields to obtain a set of hard constraint-passing candidate sets; performing semantic matching and sorting within the set of hard constraint-passing candidate sets, generating the standardized set of observation nodes when the sorting result can uniquely determine the candidates, and generating mapping conflict nodes when the sorting result cannot uniquely determine the candidates.

[0009] Further, the step of performing summary detail merging based on the standardized observation node set includes: generating a trajectory plan containing query stages, maximum hop count, and upper limit of candidates per hop based on the field nodes to be matched and the target relation signature; sequentially performing structural adjacency one-hop query, conceptual adjacency one-hop query, and attribute constraint one-hop query according to the trajectory plan to obtain a local candidate sequence; applying allowable relation signatures and exclusion rules to the local candidate sequence hop by hop, and recording the reasons for retaining, excluding, and terminating the candidates to obtain a finite one-hop mapping trajectory; verifying whether the trajectory endpoint of the finite one-hop mapping trajectory satisfies the target relation signature to obtain a summary detail candidate correspondence with the finite one-hop mapping trajectory.

[0010] Further, the step of performing quality rule verification based on the standardized observation node set to obtain a quality problem node set includes: compiling the quality dimensions of integrity, consistency, uniqueness, temporal sufficiency, and source availability into a versioned quality shape set based on the interface pattern version and quality rules; configuring an applicability selector, node shape, attribute constraints, severity level, and failure output template for each quality shape in the versioned quality shape set; determining applicable nodes from the standardized observation node set using the applicability selector; performing attribute verification, cardinality verification, type verification, relationship verification, and cross-node consistency verification on the applicable nodes; generating node-by-node satisfaction evidence for applicable nodes that pass verification, and generating node-by-node failure evidence for applicable nodes that fail verification; and generating the quality problem node set based on the node-by-node failure evidence and the failure output template.

[0011] Further, the node-by-node satisfaction evidence and the node-by-node failure evidence are aggregated according to quality dimension, field path, quality shape version, and severity level to obtain a quality coverage matrix. The quality coverage matrix records verification pass status, verification failure status, and uncovered status, respectively. Based on the uncovered status, field types or relation types not covered by any applicable quality shape are identified. For the field types or relation types not covered by any applicable quality shape, verification coverage insufficiency status nodes are generated. These verification coverage insufficiency status nodes are stored hierarchically with the quality problem nodes generated based on the node-by-node failure evidence. According to the organization's policy, the verification coverage insufficiency status nodes trigger manual configuration requests or downgrade processing flags, and the default hard blocking relationship is prohibited for these nodes. The quality coverage matrix and the verification coverage insufficiency status nodes are written into the quality status view.

[0012] The present invention also provides a system for compiling and quality nodeizing source graphs of physical examination data, including a source snapshot parsing module, an observation node standardization module, a merging and verification module, and a source graph forming module. Each module is configured to perform the corresponding data processing operations in the aforementioned method.

[0013] Compared with existing technologies, this invention has at least the following beneficial effects. The immutable source snapshot, field address, content hash, and original value preservation mechanism enable the standardization results to be traced back to specific source fields and prevent derived attributes from overwriting original information. Concept cluster rollup first recalls standard item families at a coarse-grained level, then drills down under hard constraints to verify compatibility at the field granularity. Soft scoring only passes candidate ranking for hard constraints, thus reducing the possibility of unauthorized mapping. The bounded one-hop mapping trajectory illustrates the mapping process with a local, continuous, and cacheable sequence of evidence; the query phase, maximum hop count, and candidate upper limit jointly limit the traversal scale. Versioned quality shapes transform abstract quality dimensions into node-by-node executable constraints; the quality coverage matrix further distinguishes between verified passes, verified failures, and uncovered states; nodes with insufficient verification coverage and quality problem nodes are expressed hierarchically, preventing unchecked structures from being mistakenly identified as data errors. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the main process of the method for compiling and quality node-based analysis of physical examination data source graphs provided in this embodiment of the invention;

[0015] Figure 2 A flowchart illustrating the hierarchical pattern compilation steps provided in an embodiment of the present invention;

[0016] Figure 3 A flowchart illustrating the bounded one-hop trajectory matching steps provided in an embodiment of the present invention;

[0017] Figure 4 A flowchart illustrating the quality shape compilation and node-by-node verification steps provided in an embodiment of the present invention;

[0018] Figure 5 This is a schematic diagram of the module structure of the physical examination data source graph compilation and quality nodeization system provided in the embodiments of the present invention. Detailed Implementation

[0019] The present invention will be further described below with reference to the accompanying drawings and embodiments. It should be understood that the following embodiments are used to illustrate the technical concept of the present invention and are not intended to limit the scope of protection of the present invention. The embodiments process structured physical examination data that has already been collected, and output data source diagrams, quality status, and deterministic data objects for subsequent information processing. They do not output medical conclusions or treatment suggestions for specific individuals, nor do they include operations that directly affect living organisms.

[0020] Example 1

[0021] like Figure 1 As shown in the figure, this embodiment provides a method for compiling and quality node-based analysis of physical examination data source graphs, which is performed according to the following steps.

[0022] S1. Generate an invariant sample source snapshot based on the collected multi-departmental physical examination structured data, and parse the invariant sample source snapshot to obtain the skeleton of the source map of the examined object.

[0023] The multi-departmental structured physical examination data can be collected and stored in the temporary data area after acquisition by data interfaces such as laboratory tests, imaging, functional examinations, and specialist examinations. This step performs normalized serialization on the input data corresponding to the same examinee and the same physical examination cycle, then calculates the overall summary from the serialization results, and writes the overall summary, interface mode version, examinee identifier, physical examination cycle identifier, and snapshot generation time into the snapshot header. Once the snapshot is confirmed, it is set to read-only; subsequent mapping, conversion, or quality verification will generate derived nodes, without writing back the snapshot content. The overall summary is recalculated before the compilation task begins; if the summaries match, the snapshot is read; if the summaries do not match, the current compilation task is stopped and a source snapshot integrity event is generated.

[0024] Specifically, an immutable source snapshot is a data object that solidifies both the source data content and the version context required to interpret that content. Normalized serialization at least unifies the order of object members, character encoding, numeric decimal representation, null value representation, and date / time representation; for set elements, they are sorted according to stable identifiers declared in the interface pattern, preserving the original order and setting an unstable order flag when no stable identifier exists. By eliminating representational differences unrelated to data semantics before digest computation, it is possible to avoid generating different overall digests from the same source data simply due to different arrangements of object members.

[0025] In this embodiment of the invention, snapshot generation employs an idempotent commit mechanism. The system constructs the snapshot idempotent key using the inspected object identifier, the health check cycle identifier, the interface mode version, and the overall summary. When two compilation requests have the same snapshot idempotent key, only one immutable health check source snapshot is retained, and the same snapshot identifier is returned to both compilation requests. When the inspected object identifier and health check cycle identifier are the same but the overall summary is different, the system does not overwrite the existing snapshot but instead forms a new snapshot version and connects the two snapshots through a version evolution relationship. This method can both suppress duplicate nodes caused by repeated transmissions and preserve complete evidence of the source data before and after revision.

[0026] It's important to note that the source snapshot integrity event and subsequent quality issue nodes serve different purposes. The source snapshot integrity event indicates that the compilation input failed the overall consistency check before entering the source graph; therefore, the corresponding compilation task does not generate further observation nodes. Quality issue nodes, on the other hand, indicate that the data already in the source graph has quality issues in terms of field content, relationships, or applicable rules. These two are separated by different event types and processing queues to prevent transmission corruption from being mistakenly recorded as the content of a specific result field.

[0027] During parsing, the hierarchical description corresponding to the interface mode version is used to identify the case level, cycle level, department level, examination group level, summary level, and detail level. Each level object is converted into a graph node, and parent-child edges are established between levels. Fields such as page status, display color, client cache key, and interactive button status that are not within the scope of physical examination information processing are isolated according to a whitelist. The resulting examination object source graph skeleton not only limits the mounting position of subsequent result fields but also saves the path entry point for each node to return to the source snapshot.

[0028] S2. Based on the source graph skeleton of the inspected object, configure the field address, content hash, source version and node identifier for the result field to obtain the original observation node set. Perform concept mapping, unit conversion and reference range structuring on the original observation node set to obtain the standardized observation node set with both original values ​​and standardized values.

[0029] A field address can be represented as an ordered path from the snapshot root node to the result field. Each segment in the path includes a hierarchy name, collection element location information, and field name. The content hash is calculated using the field address, original value, original unit, and source version. Node identifiers remain unique within the same examined object and examination period. For records whose positions in the array may change, the source record identifier or project combined identifier is used first as the collection element location information. Only when a stable identifier is lacking is the array sequence number used, and the corresponding instability is written into the node attribute.

[0030] Concept mapping is used to attach standard concept identifiers, standard names, and concept versions to the original observation nodes. Unit conversion is performed only when the original and target units belong to the same unit dimension and the conversion relationship can be determined; reference range structuring converts text or separated fields into lower limits, upper limits, boundary open / closed states, applicable conditions, and source versions. The original value, original unit, and original reference range are always retained in the original attribute area, while the normalized value, standard unit, and structured reference range are written to the derived attribute area. Derived attributes carry the version of the mapping template that generated the attribute and the version of the conversion rule, thus allowing the processing basis to be traced along the node attributes.

[0031] Specifically, each original observation node can be a node record containing an identity area, a source area, an original attribute area, a derived attribute area, and a processing status area. The identity area stores the node identifier and node type; the source area stores the snapshot identifier, field address, content hash, and source version; the original attribute area stores the original value, original unit, and original reference range; the derived attribute area stores the standard concept, standardized value, standard unit, and structured reference range; and the processing status area stores the mapping status, unit conversion status, quality verification status, and corresponding evidence index. Each area is set with write permissions, and only the derived attribute area and the processing status area allow the compilation task to append new versions, thus ensuring that the original information is not overwritten from a data structure perspective.

[0032] Preferably, the unit conversion rules, in addition to the conversion relationship, also record the input unit dimension, output unit dimension, numerical precision, rounding method, and applicable version of the rule. Before conversion, the unit dimensions are compared. If the dimensions are inconsistent, numerical calculations are not performed, and evidence of unit incompatibility is generated. If the dimensions are consistent but the original unit is missing, the original value is retained, and the unit conversion status is set to insufficient evidence. For values ​​undergoing conversion, the system can simultaneously save the value before conversion, the value after conversion, and the identifier of the conversion rule used, enabling subsequent modules to determine whether differences in display precision originate from the conversion process.

[0033] Taking an exemplary field as an example, the field address can be represented as "cycle node / laboratory node / examination group node / result record identifier / result value field". The content hash is generated by the field address, the original string, the original unit, and the source version. If the mapping template maps the result value field to a certain standard concept and completes the same-dimensional unit conversion, the standard concept and the conversion result are written to the derived attribute area, while the source query can still locate the original string in the snapshot along the field address. If the mapping template is updated subsequently, the system adds a derived attribute version without modifying the existing content hash, so the outputs of different rule versions can be compared on the same field source.

[0034] S3. Based on the standardized set of observation nodes, perform summary detail merging and quality rule verification to obtain the source relationship between the summary and the details, as well as the set of quality problem nodes.

[0035] The abstract-to-detail merging process first constructs matching keys for abstract nodes and detailed observation nodes. Matching keys can include departmental, project joint identifiers, standard concepts, detection methods, sites, lateralization, and standardized results. If the abstract is a combined expression of multiple detailed items, a source relationship is established between the abstract node and multiple detailed observation nodes. If the abstract and details have the same semantics and consistent results, the abstract is retained as the source of display to suppress duplicate findings. If there are irresolvable differences in results, methods, or sites, merging is not forced; instead, conflicting abstract-to-detail nodes are generated and the evidence from both sides is linked.

[0036] The objects of quality rule verification include not only individual standardized observation nodes, but also summary nodes, detailed observation nodes, and their corresponding relationships. Missing items, abandoned projects, unavailable sources, insufficient units, homonymous methods, and conflicting results are converted into quality issue nodes with attributes such as issue type, evidence address, severity level, scope of application, and handling recommendations. Quality issue nodes do not delete the original observation nodes, but rather affect whether the corresponding nodes can participate in subsequent processing through restrictive relationships, blocking relationships, or contradictory relationships.

[0037] Furthermore, when the same standardized observation node is associated with multiple quality issue nodes, the system does not use a single severity level to cover the remaining quality states, but rather retains each quality issue node and its respective evidence. The available state of a node for subsequent processing is determined by relation priority: when a blocking relationship exists, the node is prohibited from generating deterministic findings; when no blocking relationship exists but a limiting relationship exists, the node is allowed to enter the quality status view with the limiting reason; when only suggestive quality issue nodes exist, the node can continue to participate in subsequent processing but must retain suggestive evidence. Conflicting relationships are used to express that two source assertions cannot be true simultaneously; they do not directly replace blocking relationships, and whether a blocking relationship is established is determined by the severity level of the quality issue node connected to the conflicting relationship.

[0038] It is worth noting that the lifecycle of a quality issue node is tied to the rule version that generated it. During recompilation, the system first determines whether existing verification results can be reused based on the original observation node content hash and rule version. If the source content remains unchanged but the rule version changes, only the affected quality rules are re-executed, and the old quality issue node is marked as the corresponding historical compilation version. The affected scope can be determined by the "rule-field path-node type" dependency index, thereby avoiding a full verification of the entire immutable detection source snapshot every time a rule is adjusted.

[0039] S4. Based on the standardized set of observation nodes, generate a deterministic discovery set that does not exceed the semantics of the original results. Combine the source relationship and the set of quality problem nodes to establish a restrictive relationship, a blocking relationship, or a contradictory relationship, forming an initial source map and a quality status view of the inspected object.

[0040] Deterministic findings are generated by pre-configured deterministic comparison operators. These operators perform equivalence, interval, enumeration attribution, sign consistency, or unit compatibility checks only on the structured results, and write the check conditions, input node identifiers, and rule versions into the discovery node. For observation nodes lacking necessary units, with unclear applicable reference ranges, or restricted by hard blocking relationships, deterministic findings that cross the evidence boundary are not generated; instead, the corresponding quality status is preserved. These deterministic findings are data processing results and do not represent medical conclusions or treatment recommendations for specific individuals.

[0041] The initial source graph of the inspected object includes source graph skeleton nodes, original observation nodes, standardized observation nodes, summary nodes, deterministic discovery nodes, and quality problem nodes, and includes parent-child edges, derived edges, source edges, restrictive edges, blocking edges, and contradictory edges. The quality status view is a queryable view projected from the initial source graph of the inspected object according to the inspected object, examination cycle, department, problem type, and severity level. Subsequent summary or report generation modules can read data according to relation types without losing field-level source evidence.

[0042] In one implementation, the initial source graph of the inspected object undergoes graph integrity verification before submission. This graph integrity verification confirms at least the following: each standardized observation node has an original observation node source; each deterministic discovery node can trace back to at least one standardized observation node along its source edge; each quality issue node has an evidence field address or relational evidence; each blocking and restrictive relation points to a specific data node; and all nodes belong to the same inspected object and the boundary of the inspection cycle. If any of these necessary conditions are not met, the current graph version is not released to subsequent modules; instead, a graph compilation integrity event is generated.

[0043] Optionally, the system uses a versioned graph commit method to form the initial source graph of the inspected object. The compilation task first generates nodes and relationships in the temporary graph, and then switches the current graph version pointer once after integrity verification; therefore, the subsequent reading module will not observe the intermediate state where only a portion of the nodes have been generated. When recompilation only involves a certain field or rule version, the affected subgraph can be updated based on the field content hash and dependency index, and unchanged nodes can be reused. Regardless of whether a full commit or subgraph update is used, the current graph version records the corresponding snapshot identifier and version vector to ensure that the result can be reconstructed.

[0044] Example 2

[0045] like Figure 2 As shown, this embodiment provides a hierarchical compilation method based on concept rollup and field drill-down, building upon embodiment 1.

[0046] S2.1. Roll up the field fragments in the original set of observation nodes into local concept clusters according to the parent path, department, inspection group, field level and detection method to obtain a set of local concept clusters.

[0047] Field fragments are bounded by a common parent path. Within these boundaries, the compiler extracts the field name, value type, unit, sample type, detection method, location, side, and roles of adjacent fields, and groups semantically related fields that can collectively describe a check item into local concept clusters. Instead of merging field contents into a single indivisible value, the roll-up process generates concept cluster descriptions pointing to the original field nodes; therefore, each field within a cluster retains its independent field address and content hash. For fields with the same name but different detection methods, the detection method is used as a clustering condition to avoid premature mixing during the roll-up phase.

[0048] S2.2. Perform candidate recall by comparing the concept cluster descriptions in the local concept cluster set with the standard project family, the institutional project dictionary and the existing mapping template to obtain the standard concept candidate set.

[0049] Candidate recall first locates standard project families using core field names from departments, inspection teams, and concept clusters. Then, it supplements these with synonyms, historical field paths, and interface pattern version compatibility using the organization's project dictionary and existing mapping templates. This stage aims for coverage and can retain multiple concept candidates, but each candidate must have at least a project family identifier compatible with the concept cluster and the necessary attribute signatures for drill-down. The standard concept candidate set includes the candidate record's recall source, recall rule version, and corresponding concept cluster identifier.

[0050] S2.3. Drill down the candidate concepts in the standard concept candidate set to the corresponding result fields, perform field hard constraint verification, and obtain the hard constraint passing candidate set.

[0051] The drill-down phase restores the necessary attribute signatures of candidate concepts to specific result fields, verifying value type, unit dimension, sample type, detection method, location, and laterality. Hard constraints are divided into necessary hard constraints and conditional hard constraints: a candidate is excluded if any one of the necessary hard constraints fails; conditional hard constraints are only verified if the source snapshot contains the corresponding attribute. For attributes not provided in the source field but required by the candidate concept, the system marks the candidate as insufficient evidence and does not consider it passed. Thus, similar names at the concept level cannot replace medical attribute compatibility at the field level.

[0052] S2.4. Perform semantic matching sorting within the candidate set under the hard constraints. Generate the standardized observation node set when the sorting result can uniquely determine the candidate, and generate mapping conflict nodes when the sorting result cannot uniquely determine the candidate.

[0053] In one implementation, the semantic matching score is calculated according to equation (1):

[0054] Equation (1)

[0055] in, Representing candidate concepts semantic matching score, This indicates the sequence number of the candidate concept currently participating in the sorting. This indicates the semantic similarity between the concept cluster description and the candidate concept names and definitions. This indicates the consistency between the parent path of the field and the existing mapping path. This indicates the compatibility between the current interface mode version and the version applicable to the candidate mapping template. , and represents the non-negative ranking weights of the three scores, and the sum of the three ranking weights is 1. Equation (1) is only used within the candidate set that passes the hard constraint and is not used to recover candidates that have been excluded by the hard constraint.

[0056] If the highest semantic matching score corresponds to a unique candidate, and the difference between the highest and second-highest semantic matching scores reaches a preset discrimination threshold, then the highest-scoring candidate is selected and a standardized observation node is generated. If the discrimination threshold is not reached, there are ties for the highest score, or no candidate passes the hard constraints, then a mapping conflict node is generated. The discrimination threshold can be configured between 0 and 1 based on historical manual review results; the higher the threshold, the more conservative the automatic mapping. In other embodiments, the semantic matching score can also be achieved by rule-based item scoring or the output of a calibrated classification model, but the execution order of hard constraint screening first and soft scoring ranking second remains unchanged.

[0057] This hierarchical model first narrows the search scope of standard concepts by concept clusters, and then returns to the field granularity to verify necessary attributes, thus maintaining both coarse semantic granularity candidate coverage and the traceability of field-level evidence. Since soft scoring does not recover candidates that fail due to hard constraints, it reduces the likelihood of candidates with similar names but incompatible unit dimensions, sample types, or detection methods being automatically written into the source graph.

[0058] In this embodiment of the invention, the concept cluster description is essentially a high-order pattern fragment that preserves pointers to member fields, rather than an irreversible merging of member field values. The concept cluster description may include a cluster identifier, a common parent path, core name terms, a set of candidate value types, a set of candidate unit dimensions, sample type, detection method, location, lateralization, and a list of member field addresses. Candidate recall reads the coarse-grained semantics and structural attributes in the concept cluster description, while drill-down verification reads the original attributes pointed to by the member field addresses. The two stages therefore share the same source anchor but have different matching granularities.

[0059] Concept rollup, employing local boundaries, can also suppress erroneous combinations across check groups. If two field names are similar but differ in parent path, detection method, or sample type, the compiler first places them into different local concept clusters; only when the standard item family definition explicitly allows cross-field combinations are common candidates established during the candidate recall phase, rather than being directly merged during the rollup phase. Conversely, for composite structures where an item is jointly expressed by "result value field, unit field, method field, and location field," these fields, due to having a common parent path and satisfying complementary role relationships, can be rolled up into the same concept cluster, enabling candidate recall to utilize complete compositional semantics.

[0060] Taking a composite field as an example, a certain inspection group simultaneously contains four fields: name "Result," unit (preset unit), method (method M), and location (location L). If only the "Result" field name is compared, multiple standard item families might be recalled. The concept cluster description formed after rolling up simultaneously carries the unit dimension, method M, and location L. The drill-down phase then verifies each of the four original fields. Even if a candidate concept has the highest name similarity, if the method required by the candidate concept is incompatible with method M, the candidate is still excluded in the hard constraint phase. This example illustrates that rolling up expands the contextualized candidate recall information, while drill-down re-constrains candidate conclusions to specific, locatable fields. Together, they form an evidence loop from the concept layer to the field layer.

[0061] It should be understood that semantic matching ranking can be achieved using dictionary similarity, calibrated text encoding similarity, or multi-rule scoring. Regardless of the scoring method used, the system saves the sub-scores, ranking weight versions, and discrimination threshold versions, and limits the ranking input to hard constraints passing the candidate set. In this way, even if the soft scoring implementation is replaced, the hard constraint evidence, mapping conflict nodes, and field sources remain stable, preventing algorithm replacement from changing the interpretation boundaries of the original data.

[0062] Example 3

[0063] This embodiment further illustrates the implementation of field hard constraint validation.

[0064] S2.3.1. Extract necessary attribute signatures from the conceptual definitions of candidate concepts.

[0065] Required attribute signatures can be generated from machine-readable definitions of standard item families, organization-approved mapping templates, or the intersection of both. Each attribute includes an attribute name, allowed value ranges, whether it is required, default handling, and definition version.

[0066] S2.3.2. The necessary attribute signature is decomposed into value type constraint, unit dimension constraint, sample type constraint, detection method constraint, location constraint and side constraint.

[0067] The split constraints are represented using a unified constraint record, which includes at least the constraint type, comparison operator, allowed set or allowed dimension, applicable conditions, and failure reason code. Unit dimension constraints compare physical dimensions rather than unit strings, thus allowing units of the same dimension to be compatible after conversion.

[0068] S2.3.3. Read the original attribute corresponding to the necessary attribute signature according to the field address of the result field.

[0069] The read operation directly locates the original fields in the immutable source snapshot, without replacing the original evidence with derived normalized values. If an attribute is provided by a secondary field under the same parent path, the address of the secondary field is also recorded.

[0070] S2.3.4. Apply the value type constraint, unit dimension constraint, sample type constraint, detection method constraint, location constraint, and side constraint to each of the original attributes.

[0071] The validator compares each constraint record item by item. Value type validation precedes unit dimension validation; the corresponding hard constraints are only executed if the source schema contains fields for sample type, detection method, location, or lateralization. Missing necessary fields are recorded as insufficient evidence.

[0072] S2.3.5. Record the pass status, failure reason code, and evidence field address for each constraint.

[0073] Records from the same validation are linked together through candidate concept identifiers, concept definition versions, original observation node identifiers, and validation batch identifiers to form a replayable hard constraint evidence package.

[0074] S2.3.6. Only add candidate concepts that have all necessary hard constraints in the passed state to the hard constraint passed candidate set.

[0075] The default handling of hard constraints is explicitly configured by the concept definition, and semantic matching scores are not allowed to replace the failure results of necessary hard constraints. The hard constraints, through the candidate set, are therefore the sole input range for subsequent soft ranking.

[0076] S2.3.7. Generate attribute incompatible mapping conflict nodes when the hard constraint passes through the candidate set which is empty, and generate candidate non-unique mapping conflict nodes when the hard constraint passes through the candidate set which contains at least two indistinguishable candidate concepts.

[0077] The two types of mapping conflict nodes use different problem types. Attribute incompatible mapping conflict nodes store the necessary constraints and corresponding evidence addresses for the failure; candidate non-unique mapping conflict nodes store the indistinguishable candidate list and sub-scores. Neither type of node automatically writes standard concept identifiers; these can be processed by the mapping template maintenance module or the manual review module.

[0078] Specifically, hard constraint validation uses a three-state record: pass, fail, and insufficient evidence. A pass indicates that the source field possesses the corresponding attribute and satisfies the necessary attribute signature of the candidate concept; a fail indicates that the source field possesses the corresponding attribute but is incompatible with the necessary attribute signature; insufficient evidence indicates that the candidate concept requires a certain necessary attribute, but the source field does not provide sufficient information to make a compatibility judgment. For necessary hard constraints, neither failure nor insufficient evidence is included in the hard constraint pass candidate set, but they generate different reason codes for subsequent determination of whether to revise the mapping template or supplement the interface field.

[0079] In one implementation scenario, two candidate concepts have the same name and unit dimension, where the first candidate requires sample type A, and the second candidate requires sample type B. If the source field explicitly records sample type A, the first candidate passes and the second candidate fails. If the source field does not provide a sample type, both candidates are marked as insufficient evidence, and a non-unique candidate mapping conflict node is generated, instead of selecting one based solely on name score. Therefore, the system retains "incompatible" and "cannot confirm incompatibility" separately in the evidence, preventing unknown attributes from being assumed to be satisfied.

[0080] Furthermore, the manual review results can generate a new version of the existing mapping template, but without directly modifying the standardized observation nodes in the current compilation task. The review module records the conflict node identifier, selected candidate, review basis, and new template version; subsequently, the new compilation task uses the new template version to re-execute hard constraint verification. This process ensures that manual selection also undergoes a versioned compilation and evidence generation process, preventing the appearance of manually defined standard concepts lacking rule support in the source graph.

[0081] Example 4

[0082] like Figure 3 As shown, this embodiment provides an interpretable summary detail merging method based on a one-hop query trajectory, based on embodiment 1.

[0083] S3.1. Based on the field nodes to be matched and the target relationship signature, generate a trajectory plan that includes the query stage, the maximum number of hops, and the upper limit of candidates per hop.

[0084] The target relation signature describes the start-point type, allowed relation sequences, end-point type, and required end-point attributes for a valid trajectory. For summaries to details, the start-point type is a summary node, and the end-point type is a detail observation node; for local fields to standard concepts, the start-point type is an original observation node, and the end-point type is a standard concept node. The trajectory plan predetermines the phase order of structural adjacency, concept adjacency, and attribute verification based on the target signature, and configures the maximum number of hops and candidate limits for each phase.

[0085] S3.2. Perform structural adjacency one-hop query, concept adjacency one-hop query, and attribute constraint one-hop query in sequence according to the trajectory plan to obtain a local candidate sequence.

[0086] Structural adjacency queries only access adjacent nodes that are within the same department, inspection group, parent path, or have a clear summary / detail relationship; conceptual adjacency queries only access standard concept, item family, or item dictionary nodes that are directly connected to the current node; attribute constraint queries only access attribute nodes such as detection method, sample type, unit dimension, location, and lateral. Each query uses the currently selected node as the starting point for the next query and does not perform full-graph expansion that deviates from the trajectory plan.

[0087] S3.3. Apply allowable relation signatures and exclusion rules to the local candidate sequence hop by hop, and record the reasons for retaining the candidate, the reasons for excluding it, and the reasons for terminating it to obtain a finite one-hop mapping trajectory.

[0088] The query size can be constrained by equation (2):

[0089] Equation (2)

[0090] in, This represents the total number of candidate nodes actually checked for a single finite one-hop mapping trajectory. This represents the trajectory number of the current finite one-hop mapped trajectory. This indicates the maximum number of hops allowed in the trajectory plan. This represents the jump count variable. Indicates the first The upper limit of candidates allowed to be retained during the jump. This is achieved by limiting each candidate individually. and This prevents consecutive one-hop queries from degenerating into uncontrolled multi-hop traversals. In one example, the maximum number of hops for the summary detail trajectory is set to 4, and the upper limit for each hop candidate in the structural adjacency phase and the conceptual adjacency phase is set to 20 and 10, respectively; the above values ​​are for illustrative purposes only and can be configured according to the interface pattern scale and compilation latency requirements.

[0091] Each operation, whether retained or excluded, records the starting node, relationship type, candidate set, rule version, reason code, and selected result. When the maximum hop count is reached, the current candidate set is empty, or the endpoint satisfies the target signature, a corresponding termination reason is generated, and the query stops. Because local evidence is saved at each step, the audit module can progressively review the mapping basis along the trajectory from the starting point without re-executing a full similarity search.

[0092] Specifically, the target relation signature not only specifies the trajectory endpoint type but also encodes the allowed relation types and stage order into a finite state plan. The trajectory executor accepts only one-hop query results allowed by the current state at each stage. If the node type or relation type in the query result does not match the current state, the reason for exclusion is immediately recorded, without extending the corresponding node to the next hop starting point. Therefore, although consecutive one-hop queries can span multiple local relations, each state transition is restricted by the target relation signature, preventing arbitrary graph traversal.

[0093] The local candidate sequence can adopt a record structure of "trajectory identifier, jump number, starting node identifier, relationship type, candidate node identifier set, retained node identifier set, exclusion reason set, and plan version". The original attributes of candidate nodes are not copied into the trajectory record, but are referenced through node identifiers; this reduces trajectory storage and ensures that the attributes read during auditing are consistent with the corresponding source graph version. If the source graph version changes, the trajectory record still points to the graph version at the time of its generation and is not mixed with nodes of the same name in the new graph version.

[0094] Taking the trajectory from the summary node to the detailed observation node as an example, the first hop can access the same inspection group from the summary node along the parent-child edge; the second hop can access the detailed observation node from the inspection group; the third hop can access the standard concept associated with the detailed observation node; and the fourth hop can verify the method, location, and lateral attributes. If the number of candidates for the second hop exceeds the candidate limit, the system first removes nodes with inconsistent examination cycles or departments, and then prunes them according to structural adjacency distance and stable node identifiers. If the third hop has confirmed that the endpoint node meets the target signature, it terminates early without executing an unnecessary fourth hop. The trajectory ultimately retains the reasons for each pruning and early termination, allowing the formation process of the same correspondence to be reviewed step by step.

[0095] Compared to a one-time multi-condition query, the finite state plan breaks down composite matching into multiple local operations with explicit inputs and outputs. When a local operation fails, the system can determine whether the failure occurred during the structural adjacency, conceptual adjacency, or attribute verification stage; therefore, rule maintainers can adjust relation signatures or supplementary fields for specific stages without having to guess the cause based solely on an overall similarity result. Simultaneously, the upper limit of candidates per hop and the maximum number of hops limit the candidate checking scale within the trajectory plan, allowing compilation latency to be configured according to the interface pattern scale.

[0096] S3.4. Verify whether the endpoint of the finite one-hop mapping trajectory satisfies the target relation signature, and obtain the summary details candidate correspondence with the finite one-hop mapping trajectory.

[0097] Endpoint verification includes signature items such as node type, standard concept, detection method, location, side, and source version. When all necessary signature items pass, the trajectory identifier is written into the candidate correspondence of the summary details; if necessary signature items fail or the trajectory terminates prematurely, no source relationship is generated, but the trajectory and termination reason are handled by the conflict node generation logic. In other implementations, one-hop queries can be implemented using graph database adjacency queries, in-memory adjacency list queries, or relational database index queries, as long as stage boundedness, step-by-step recording, and endpoint signature verification are maintained.

[0098] Example 5

[0099] This embodiment further illustrates the termination, recording, and caching mechanism of the finite one-hop mapping trajectory.

[0100] S3.3.1. The execution order of the structure query stage, semantic query stage, and attribute verification stage is determined according to the target relation signature.

[0101] The execution order of stages is saved as part of the trajectory plan along with the target relation signature versioning. If no candidates are generated in the previous stage, the next stage will not proceed; the selected nodes generated in the previous stage constitute the starting point for a one-hop query in the next stage.

[0102] S3.3.2. When the upper limit of candidates per hop is reached in any query stage, candidates are retained according to the number and structural adjacency distance that satisfy the hard constraints.

[0103] Candidate pruning first compares the number of hard constraints satisfied, then compares the structural adjacency distances to the starting field node; if they still cannot be distinguished, stable node identifiers are used for sorting to ensure that results can be repeated under the same input and the same version.

[0104] S3.3.3. Terminate the query when the number of hops executed reaches the maximum number of hops, the current candidate set is empty, or the trajectory endpoint satisfies the target relationship signature.

[0105] The three termination conditions generate reason codes for reaching the boundary, no available candidates, and the target being met, respectively. Reaching the boundary does not equate to a matching failure; the system can mark the corresponding trajectory as insufficient evidence and place it in the manual review queue.

[0106] S3.3.4. For each query, specify the starting node, relation type, candidate set, selected node, and reason code.

[0107] Query records can also carry the trajectory plan version, execution time, and rule version, but without changing the listed core fields. The candidate set is stored using a list of node identifiers to avoid copying the original values.

[0108] S3.3.5. Link the records from each query in the execution order to form the finite one-hop mapping trajectory.

[0109] When linking, a preceding record identifier and sequence number are configured for each record. The first record points to the field node to be matched, and the last record points to the end point of the trajectory or the termination event, thus forming a finite sequence that is not repeatable.

[0110] S3.3.6. Generate field fingerprints and version vectors based on the content hash, field address, interface mode version, and mapping template version of the field to be matched.

[0111] Field fingerprints are used to identify field content and location, while version vectors are used to identify the patterns and templates used to interpret the field. Any change to any component generates a different cache key to avoid reusing expired traces across versions.

[0112] S3.3.7. Save the finite one-hop mapping trajectory using the field fingerprint and version vector as cache keys, and reuse the finite one-hop mapping trajectory when the cache keys are consistent.

[0113] Even after a cache hit, the existence of the trajectory endpoint node and the consistency of the target signature version are still verified. If a cache miss or verification fails, the trajectory plan is re-executed. This mechanism allows for the reuse of structurally identical audit paths for the same fields in batches of inspected object data, while maintaining result isolation after interface pattern updates.

[0114] In this embodiment of the invention, the trajectory cache can be divided into a plan cache and a result cache. The plan cache stores the query stage, maximum number of hops, upper limit of candidates per hop, and allowed relation sequence generated by the target relation signature, and is applicable to the same target signature and interface mode version. The result cache stores the finite one-hop mapping trajectory corresponding to the fingerprint of a specific field and the version vector. Changes to the plan cache will not modify the already generated result cache; only new compilation tasks will regenerate the trajectory using the new plan version. By separating the two types of caches, a stable query plan can be reused while maintaining the reproducibility of historical results.

[0115] Preferably, cache writing employs an atomic process of verification before publication. After trajectory execution is complete, the system first verifies that the trajectory sequence numbers are consecutive, the starting node of each query is consistent with the previously selected node, and the endpoint signature satisfies the same status as the termination reason. Then, the result cache is set to a reusable state. Trajectories that fail verification are only retained as execution logs and are not provided as cache hits for other compilation tasks. When multiple tasks generate the same cache key simultaneously, a cache key-level mutual exclusion or comparison-before-write mechanism can be used to publish only one verified trajectory version.

[0116] Optionally, cache invalidation does not employ a time-dependent overall clearing approach. Instead, it compares the versions of components within the cache key. Changes in field content hashes indicate changes to the source field content, changes in field addresses indicate changes to the structural location, changes in interface pattern versions indicate changes to the structural interpretation, and changes in mapping template versions indicate changes to candidates or constraints. Any change invalidates the corresponding result cache. This allows for precise reuse of unaffected trajectories and prevents old trajectories from being misused with new field structures.

[0117] Example 6

[0118] like Figure 4 As shown, this embodiment provides a constraint verification method based on mass shape compilation, based on embodiment 1.

[0119] S3Q.1. Based on interface pattern versioning and quality rules, the integrity, consistency, uniqueness, temporal sufficiency, and source availability quality dimensions are compiled into a set of versioned quality shapes.

[0120] The compiler first breaks down quality rules in natural language or configuration table form into rule scope, verification objects, verification conditions, severity level, and failure output. Then, it organizes rules with the same interface mode version and applicable to the same node type into quality shapes. The integrity dimension constrains the existence of required fields or relationships; the consistency dimension constrains the compatibility of summaries and details, results and anomaly flags, or numerical values ​​and units; the uniqueness dimension constrains the cardinality of the project's joint identifier within a specified level; the temporal sufficiency dimension constrains the relationship between the result time, collection time, and health check period; and the source availability dimension constrains whether the source snapshot, field address, and source version can be located.

[0121] Each compilation generates a quality shape set version and saves the interface pattern version, quality rule version, and compiler version. For rules that are not applicable to the pattern version, they are not silently considered passed, but excluded from the current versioned quality shape set, and the coverage evaluation logic determines whether the corresponding structure is in an uncovered state.

[0122] S3Q.2. Configure the applicable selector, node shape, attribute constraints, severity level, and failure output template for each quality shape in the versioned quality shape set.

[0123] The selector selects the object to be verified based on node type, department, inspection group, standard concept, field path pattern, or source version; the node shape describes the attributes and relationship structure that the target node is allowed to have; attribute constraints record attribute paths, comparison operators, expected types, allowed cardinality, and cross-node references; severity levels are used to distinguish between hints, restrictions, and blockages; the failure output template specifies the quality problem type, evidence fields, recommended actions, and the type of relationship established with the target node. Each component has a stability identifier, and the verification evidence can point back to a specific quality shape and its version.

[0124] S3Q.3. Use the applicable selector to determine the applicable nodes from the standardized set of observation nodes.

[0125] The selector operates within the structural boundaries of the source graph skeleton. A standardized observation node can be selected by multiple quality shapes, thus undergoing verification on different quality dimensions. Nodes not selected by the selector do not generate satisfaction or failure evidence; their uncovered state is recorded by the subsequent coverage matrix. The selector execution result saves the target node identifier, selection criteria, and quality shape identifier.

[0126] S3Q.4. Perform attribute verification, cardinality verification, type verification, relationship verification, and cross-node consistency verification on the applicable nodes.

[0127] Attribute validation checks if the target attribute exists and if its value meets the allowed range; cardinality validation checks the number of values ​​or relationships along the same path; type validation checks the value type and unit dimension; relationship validation checks the existence and direction of source edges, parent-child edges, or summary detail edges; cross-node consistency validation checks the results, methods, locations, sides, or temporal relationships of nodes with explicit associations. The validator only uses applicable nodes, quality shapes, and explicitly referenced adjacent nodes, and does not introduce data from outside the source graph.

[0128] S3Q.5. Generate node-by-node satisfaction evidence for applicable nodes that pass verification, and generate node-by-node failure evidence for applicable nodes that fail verification.

[0129] Node-by-node satisfaction evidence includes at least the target node identifier, quality shape identifier, constraint identifier, pass status, verification time, and evidence field address; node-by-node failure evidence also includes the actual value digest, expected constraints, failure reason code, and severity level. For actual values ​​containing sensitive content, only the content hash or a minimized value digest may be saved, while the original value is still accessible within the controlled environment via the field address.

[0130] S3Q.6. Generate the set of quality problem nodes based on the node-by-node failure evidence and the failure output template.

[0131] Multiple failure pieces of evidence generated by the same target node under the same quality shape can be merged into a single quality problem node, which retains references to each failure piece of evidence. Failure pieces of evidence from different quality dimensions do not overlap. Quality problem nodes with a severity level of blocking are established with a blocking relationship, and quality problem nodes with a severity level of limiting are established with a limiting relationship. Quality problem nodes with a warning level are only included in the quality status view.

[0132] In one implementation, the quality dimension The node coverage rate is calculated according to equation (3):

[0133] Equation (3)

[0134] in, Indicates quality dimension Node coverage Indicates the current quality dimension identifier. This represents the number of nodes that, under the current interface mode version, are selected by at least one of the quality shapes and generate either satisfying or failing evidence. This represents the total number of nodes that should undergo verification based on the interface pattern definition. When When the value is zero, this quality dimension is marked as inapplicable and no node coverage is calculated. Equation (3) is used to describe the verification capability coverage, not to determine whether the data content of a single node is correct.

[0135] Because quality shapes translate quality dimensions into enforceable local constraints on specific nodes and relationships, verification results can simultaneously carry rule-based evidence and field-based proof. When interface patterns or quality rules change, a new set of quality shapes can be formed through versioning and recompilation, without needing to alter the original observation nodes and field-level source anchoring structures.

[0136] In this embodiment of the invention, quality shape compilation includes rule parsing, selector normalization, constraint decomposition, dependency establishment, and executable plan generation. Rule parsing breaks down the applicable objects, conditions, and failure outputs in the quality rules into structured entries; selector normalization converts field path aliases and node type aliases into unified identifiers in the interface pattern; constraint decomposition breaks down composite conditions into attribute, cardinality, type, relation, and cross-node consistency constraints; dependency establishment records the node attributes and relation types read for each constraint; and executable plan generation arranges local verification operations according to the dependency order. The compilation result can be stored as a machine-executable intermediate representation of the quality shape, thereby decoupling rule configuration from runtime verification.

[0137] The executable intermediate representation may include a quality shape identifier, an applicable interface mode version range, an applicable selector, a target node type, a constraint operation sequence, a set of read paths, a severity level, a failure output template, and a shape version. The runtime verifier does not reinterpret the natural language rules; instead, it directly consumes the executable intermediate representation and writes the shape identifier, constraint operation sequence number, and read path into the evidence. Thus, the execution meaning of the same quality rule remains definite within a specific version, and the verification result can pinpoint the specific failed constraint operation.

[0138] Specifically, cross-node consistency constraints limit the comparison objects through explicit relational paths. For example, when performing consistency verification on summary nodes and detailed observation nodes, the quality shape first requires that they be located in the same examination cycle and examination group boundary, then obtains the comparison nodes along the summary-detail candidate relationship, and subsequently compares the standard concepts, methods, locations, aspects, and normalization results. If the relational path does not exist, evidence of relational verification failure is generated; if the relationship exists but the attributes are inconsistent, evidence of cross-node consistency failure is generated. These two types of failure correspond to structural gaps and content conflicts, respectively, avoiding the use of the same error code to mask different causes.

[0139] Taking an exemplary interface pattern as an example, a result field in the pattern definition requires a result value, zero or one unit field, and at least one source edge. The integrity quality shape selects the result field node using an applicable selector. Cardinality constraints verify the number of result values, type constraints verify the type of result values, and relation constraints verify the source edges. When a result value exists and is type-compatible, and a source edge exists but a unit field is not provided, the integrity dimension can generate evidence of satisfaction, while the unit sufficiency dimension generates evidence of failure. This evidence is written to different quality dimensions, indicating that the same node can pass in one dimension and fail in another, without compressing the node into a single, general state.

[0140] It should be understood that both node-by-node satisfaction evidence and node-by-node failure evidence are part of the compilation result. Satisfaction evidence proves that the corresponding node has indeed been checked for a specific quality shape, rather than being presumed to have passed simply because there is no error log; failure evidence is used to generate nodes with quality problems. It is precisely because both passing and failing have explicit evidence that the subsequent quality coverage matrix can distinguish between "checked and passed", "checked but failed", and "no applicable checks".

[0141] Example 7

[0142] This embodiment further illustrates the generation of the quality coverage matrix and the state nodes with insufficient verification coverage.

[0143] Step Q1. Aggregate the node-by-node satisfaction evidence and the node-by-node failure evidence according to the quality dimension, field path, quality shape version, and severity level to obtain the quality coverage matrix.

[0144] The rows of the quality coverage matrix can correspond to field types or relation types in the interface schema, and the columns can correspond to quality dimensions and quality shape versions. The matrix cells store the number of applicable nodes, the number of verified nodes, the number of passes, the number of failures, and the evidence index, so that the coverage results can be traced back to specific quality shapes and field paths.

[0145] Step Q2. Record the verification pass status, verification failure status, and uncovered status in the quality coverage matrix respectively.

[0146] A verification pass status is recorded when a node is selected by an applicable quality shape and all applicable constraints pass; a verification failure status is recorded when at least one failure criterion exists; an uncovered status is recorded when the interface pattern defines that a corresponding field or relation should be verified for a certain quality dimension, but no applicable quality shape selects the node. These three statuses are stored mutually exclusively to avoid simultaneously representing both pass and unchecked results with an empty result.

[0147] Step Q3. Identify the field type or relation type that is not covered by any applicable quality shape based on the uncovered state.

[0148] During identification, the interface pattern definition is used as the baseline for the expected structure. The field types and relation types appearing in the current pattern version are compared with the union of the quality shape applicable selectors. The corresponding structure is considered to be in an uncovered state only if the expected validation is not covered by the union of applicable selectors.

[0149] Step Q4. Generate a validation coverage inadequate state node for the field type or relation type that is not covered by any applicable quality shape.

[0150] The "Insufficient Coverage" status node stores the interface mode version, field type or relationship type, uncovered quality dimension, detection time, and suggested configuration entry. This node indicates the verification capability status and does not declare any errors in the corresponding field content.

[0151] Step Q5. Store the nodes with insufficient verification coverage and the quality problem nodes generated based on the node-by-node failure evidence in a hierarchical manner.

[0152] In a graph structure, nodes with quality issues are connected to target data nodes via restrictive, blocking, or contradictory edges; nodes with insufficient verification coverage are connected to interface schema nodes, field type nodes, or relation type nodes via coverage status edges. Different relation types enable graph computation to clearly distinguish between data quality and verification capabilities.

[0153] Step Q6. According to the organization's policy, trigger a manual configuration request or a downgrade processing flag for the node with insufficient verification coverage, and prohibit the node with insufficient verification coverage from triggering a hard blocking relationship by default.

[0154] Institutional strategies can select from prompts, manual configuration, or downgrade processing based on quality dimensions, departments, field types, and duration of inadequate coverage. Downgrade processing will cause the corresponding data to carry an "insufficiently validated" tag in subsequent stages, but will not establish a hard block relationship without evidence of data failure. Only when the institutional strategy has other explicit configurations and retains the strategy version can the processing level of inadequate coverage be upgraded.

[0155] Step Q7. Write the quality coverage matrix and the verification coverage insufficiency status node into the quality status view.

[0156] The written quality status view can query three categories of status: data passed inspection, data inspection failed, and structure not inspected. This allows rule maintainers to locate field paths lacking quality shape, and the data usage module to understand the verification capability boundaries when reading standardized observation nodes.

[0157] Specifically, the quality coverage matrix is ​​not a simple count of the number of nodes with quality issues. Instead, it uses the expected verification objects defined by the interface pattern as the denominator and the nodes that actually generate evidence of satisfaction or failure as the covered objects. Optional fields in the interface pattern only enter the expected verification object set when the current snapshot actually occurs; required fields, even if missing in a snapshot, can generate evidence of failure based on the integrity quality shape of the parent node. This calculation method ensures that coverage reflects whether the rule reaches the structure that should be verified, rather than being affected by the number of empty or optional fields.

[0158] In this embodiment of the invention, the three coverage states are determined according to the existence of evidence and change with the version vector. For a given field type, quality dimension, and quality shape version, a verification pass state is recorded when at least one applicable quality shape exists and all applicable verifications generate satisfactory evidence; a verification failure state is recorded when failed evidence exists; and an uncovered state is recorded when the interface pattern declares that verification should be performed but no applicable shape exists. After the quality shape version is updated, the old state is retained in the historical quality state view, and the new version recalculates the state, without overwriting the old version evidence with the new state.

[0159] Taking the scenario of a new interface field appearing for the first time as an example, the new interface field can be parsed into an original observation node, but the applicable selectors for existing quality shapes do not cover the corresponding field type. The system records this field type as uncovered in the quality coverage matrix and generates a node with insufficient verification coverage. This node triggers a quality rule configuration request and allows standardized observation nodes to carry an insufficient verification flag into subsequent controlled processing. After the new quality shape is released, the new compilation task generates satisfaction evidence or failure evidence for this field, and the quality status changes from uncovered to verification passed or verification failed. The uncovered status in historical compilation tasks can still be reviewed.

[0160] It is worth noting that the institutional strategy only consumes nodes with insufficient verification coverage, without changing the facts expressed by those nodes. Manual configuration requests are used to prompt rule maintainers to supplement applicable quality shapes; degradation processing flags are used to prompt data-using modules to narrow the scope of application of the corresponding data; hard blocking relationships are not established by default because the lack of verification rules only proves insufficient verification capability, not that the field content is incorrect. Only when a quality shape actually produces evidence of failure will corresponding quality problem nodes and restriction or blocking relationships be generated based on the severity level.

[0161] Through the aforementioned two-tiered state structure, quality maintenance and data processing can operate separately around different objects. The quality maintenance module aggregates and verifies insufficient coverage state nodes by interface mode nodes, field types, and quality dimensions to identify rule configuration gaps. The data processing module reads the content quality status by standardized observation nodes and quality issue nodes to determine whether a node can continue to participate in deterministic discovery generation. The two processing paths are related but not intertwined in the quality status view, thus preventing rule blind spots from propagating into data errors.

[0162] Example 8

[0163] This embodiment further illustrates the formation of the nonvariable source snapshot and the source map skeleton of the inspected object.

[0164] S1.1. Calculate an overall summary of the structured physical examination data from the multidisciplinary departments.

[0165] Before calculating the overall summary, the data is deterministically serialized according to the field order, character encoding, null value representation, and numerical format specified in the interface mode version. For transmitted metadata that does not affect the meaning of the physical examination data, it can be excluded first according to the mode definition before calculating the summary.

[0166] S1.2. Bind the overall summary with the interface mode version, the inspected object identifier and the physical examination cycle identifier to form the immutable physical examination source snapshot.

[0167] The binding information is written to the snapshot header and stored together with the snapshot body. The inspected object identifier can be a de-identified identifier generated internally by the system; the physical examination cycle identifier is used to distinguish data from different batches of the same inspected object, avoiding the incorrect merging of cross-cycle fields.

[0168] S1.3. After verifying the overall summary of the immutable source snapshot, parse the case level, cycle level, department level, examination group level, summary level, and detail level.

[0169] The parser selects the hierarchical parsing rules based on the interface pattern version. When an optional level allowed by the pattern is missing, it retains the empty level state; when encountering undeclared structural changes, it generates a source structure exception node instead of guessing the meaning of the fields.

[0170] S1.4. Isolate client fields that do not participate in physical examination data processing based on a preset field whitelist.

[0171] The default field whitelist records the field types and path patterns that are allowed to enter the source graph. Client fields not accepted by the whitelist enter the quarantine area, where only the quarantine reason and content summary are retained, and they do not participate in concept mapping, quality verification, or deterministic discovery generation.

[0172] S1.5. Convert the hierarchical parsing results into graph skeleton nodes with parent-child edges to obtain the source graph skeleton of the inspected object.

[0173] Each graph skeleton node is configured with node type, parent node identifier, source path, and source version. Only one skeleton node is generated for the same source object; when multiple access paths exist, each access path is appended as a source attribute without copying the node.

[0174] S2.1. When performing concept mapping, unit conversion, and reference range structuring on the original set of observation nodes, the standardized attributes are written as derived attributes into the corresponding original observation nodes, and the derived attributes are prohibited from overwriting the original values ​​and original units.

[0175] Derived attributes and original attributes are set in separate namespaces. By default, the reading module returns the original attribute, derived attribute, and derivation basis simultaneously. When the standardization rules are updated, a new version of the derived attribute is generated, and the old version can still be accessed according to the task retention policy, thus supporting version review of the source graph.

[0176] Specifically, the preset field whitelist can be composed of the allowed path set and the allowed node type set in the interface mode version. When the parser encounters a field not declared in the allowed path set, it first determines whether the field belongs to a known client field type; if it belongs to a known client field type, it is written to an isolated record; otherwise, it is written to an unknown structure record. Both types of records store the field path and content summary, but do not save them to the original observation node set, thereby preventing page display status or temporary interactive fields from accidentally entering the concept mapping space.

[0177] In this embodiment of the invention, hierarchical parsing employs parent node existence constraints. Before generating any inspection group node, the existence of the corresponding department node and cycle node must be confirmed; before generating summary nodes or detail nodes, the existence of the corresponding inspection group node must be confirmed. If the source data uses flat records and does not explicitly contain intermediate levels, the parser can construct the corresponding skeleton node based on the grouping key declared in the interface pattern, but the grouping key and pattern version used must be recorded as node source attributes. This method constitutes a deterministic reconstruction of the source structure, without guessing the hierarchy based on the result content.

[0178] Taking a scenario where data is transmitted twice within the same health check cycle as an example, the first transmission creates snapshot version V1 and generates the corresponding source graph skeleton of the examined object; the second transmission adds a detailed record, and the overall summary after normalization and serialization changes, resulting in snapshot version V2. The original observation nodes and derived attributes in V1 remain unchanged, while V2 re-parses the newly added record and reuses the compilation results of unchanged fields based on content hashing. Subsequent queries can specify V1 or V2, or compare newly added, deleted, and changed field nodes along the version evolution relationship.

[0179] It should be understood that derived attribute versions and source snapshot versions are two different dimensions. Changes in the source snapshot version indicate changes in the content of the source data; changes in the derived attribute version indicate changes in the mapping template, unit conversion rules, or reference range structured rules used on the same source data. The version vector records both dimensions simultaneously, enabling the system to distinguish between "source data being revised" and "interpretation rules being revised," and accordingly choose to recompile the entire dataset or update only the derived attributes.

[0180] Example 9

[0181] This embodiment further illustrates summary detail merging, quality issue nodeization, and duplicate discovery suppression.

[0182] S3.1a. Construct summary detail matching keys based on department, project joint identifier, standard concept, testing method, location, side and result.

[0183] The item joint identifier in the matching key preferentially uses the stable item identifier provided by the source system; if there is no stable item identifier, it can be constructed from stable fragments of department, inspection group, and field address. The result is only normalized before entering the matching key, without changing the semantics of the original result.

[0184] S3.2a. Generate candidate correspondences for summary details based on the summary detail matching key.

[0185] Keys that are exact matches generate first-priority candidates; keys that are consistent in standard concepts, methods, locations, and sides but lack a joint identifier for the item generate second-priority candidates. The candidate relationship stores the priority and the comparison results of each key component.

[0186] S3.3a. Perform consistency verification, source version verification, and duplication suppression on the candidate correspondence of the summary details.

[0187] Consistency checks are performed based on value type, using methods such as equivalent comparison, unit conversion, or standardized text consistency comparison. Source version verification confirms that the summary and details belong to the same inspected object, inspection cycle, and interface mode version. Duplicate suppression only suppresses duplicate outputs; it does not delete summary or detail observation nodes.

[0188] S3.4a. When the verification passes, generate the source relationship between the summary and the details; when the verification fails, generate the summary-detail conflict node.

[0189] The source relationship record includes the matching key, verification rule version, and evidence node identifier. The summary and detail conflict node record includes the conflicting components, summary evidence address, and detail evidence address, and establishes the conflict relationship.

[0190] S3.5a. Use quality rules to identify missing fields, abandoned projects, unavailable sources, insufficient units, conflicting methods with the same name, and conflicting results, and generate the set of quality problem nodes.

[0191] Project abandonment is represented as a business status node with a clearly defined source field, and should not be confused with missing fields. Source unavailable can be triggered by snapshot digest failure, invalid field address, or uninterpretable source version. Insufficient units indicates a value exists but reliable unit conversion cannot be performed.

[0192] S4.1a. Apply the deterministic comparison operator to the standardized observation nodes that are not restricted or blocked by the set of quality problem nodes to generate the deterministic discovery set.

[0193] Deterministic comparison operators, their applicable value types, input unit dimensions, and result enumerations are all pre-versioned. If the input does not meet the operator's preconditions, the comparison is not performed and the corresponding quality state is not generated; generating discoveries by guessing units or reference ranges is prohibited.

[0194] S4.2a. Merge the summary findings and detailed findings corresponding to the same standardized observation node into a single deterministic discovery node, while preserving the respective source relationships of the summary findings and the detailed findings.

[0195] The merged deterministic discovery node saves the summary source edge and the detail source edge, and configures the display priority and evidence priority separately. This reduces redundant expressions in downstream report generation without losing the field addresses and source versions of the summary and details.

[0196] Specifically, the summary detail matching key uses a fractional recording method, rather than simply storing the concatenated single string. Each component separately records its original value, normalized value, comparison result, and evidence field address; therefore, when a candidate correspondence fails verification, the system can indicate whether the difference originates from the project joint identifier, standard concept, detection method, location, lateralization, or result. For components not provided by the source system, the matching key records a missing status, and whether this missing status allows entry into the candidate is determined by the corresponding merging rule version.

[0197] In one implementation, duplication suppression primarily relies on standardized observation node identifiers and deterministic comparison operator versions. When summary findings and detailed findings point to the same standardized observation node and generate the same result from the same operator version, the system creates a deterministic discovery node and retains both source edges. If the two findings display the same text but correspond to different observation nodes, merging is not performed. Thus, duplication suppression applies to discovery outputs with the same evidentiary basis, rather than deleting data solely based on identical displayed text.

[0198] Taking an exemplary inspection group as an example, the summary node records the normalized result R, and the two detailed observation nodes correspond to items P1 and P2, respectively. The summary-detail matching key indicates that the summary node is consistent with P1 in terms of standard concepts, methods, locations, aspects, and results, while it is only similar in name to P2 but inconsistent in method. The system establishes a source relationship between the summary node and P1 and suppresses duplicate findings, while generating a summary-detail conflict node for the summary node and P2. During subsequent readings, a deterministic discovery node retains source evidence for both the summary and P1, while P2 still exists as an independent observation node carrying a conflict status.

[0199] It should be noted that the semantic boundaries of the deterministic comparison operator are jointly defined by the input type, applicable unit dimension, comparison relation, and output enumeration. The operator only converts the structured results into data findings directly corresponding to the comparison conditions, without deriving individual conclusions beyond the input fields and comparison rules. If the reference range has age group, gender, or other applicable conditions, but the source snapshot lacks corresponding conditions, the system does not select a default reference range to continue the comparison, but instead generates quality problem nodes where the applicable conditions of the reference range are insufficient. This processing places the sufficiency of evidence before the generation of findings, preventing the data processing results from exceeding the semantic range supported by the original fields.

[0200] Furthermore, source relationships and duplication suppression relationships can support reverse tracing. When the data usage module reads deterministic discovery nodes, it can return summary nodes, detailed observation nodes, normalized observation nodes, original observation nodes, and immutable source snapshots along the source edges; when the source field or mapping template is revised, the affected deterministic discovery nodes can also be located through the reverse dependency index. This bidirectional traceability structure enables local revisions to precisely trigger the regeneration of relevant discoveries without affecting irrelevant check groups.

[0201] Example 10

[0202] like Figure 5 As shown, this embodiment provides a system for compiling and quality node-based analysis of physical examination data source graphs.

[0203] The source snapshot parsing module is used to generate an invariant source snapshot based on the collected multi-departmental physical examination structured data, and to parse the invariant source snapshot to obtain the source map skeleton of the examined object.

[0204] The source snapshot parsing module may include a deterministic serialization component, an overall summary component, a snapshot read-only storage component, and a hierarchical parsing component, used to perform the data processing described in step S1 of embodiment 1.

[0205] The observation node standardization module is used to configure field addresses, content hashes, source versions, and node identifiers for result fields based on the source graph skeleton of the inspected object, to obtain an original set of observation nodes. The original set of observation nodes is then subjected to concept mapping, unit conversion, and reference range structuring to obtain a standardized set of observation nodes in which original values ​​and standardized values ​​coexist.

[0206] The observation node standardization module can load the hierarchical pattern compilation component to independently enable the concept rollup, candidate recall, field drill-down, hard constraint filtering, and soft scoring sorting functions described in Embodiments 2 and 3; when the hierarchical pattern compilation component is not enabled, standardization can be performed using a version-confirmed fixed mapping template.

[0207] The merge verification module is used to perform summary detail merging and quality rule verification based on the standardized observation node set, so as to obtain the source relationship between the summary and the detail and the set of quality problem nodes.

[0208] The merge verification module can load the trajectory matching component and the quality shape component respectively. The trajectory matching component is used to perform the bounded one-hop query described in Embodiments 4 and 5; the quality shape component is used to perform the quality shape verification and coverage evaluation described in Embodiments 6 and 7. Each component has an enable switch and version configuration, and can be individually combined with the main process of Embodiment 1.

[0209] The source graph formation module is used to generate a deterministic discovery set that does not exceed the semantics of the original result based on the standardized set of observation nodes, and to establish restrictive, blocking, or contradictory relationships by combining the source relationship and the set of quality problem nodes, thereby forming an initial source graph and quality status view of the inspected object.

[0210] The source graph formation module can be implemented using an attribute graph, a resource description framework graph, or other graph data structures that support nodes and directed relationships. Regardless of the storage method used, the original observation node, derived attribute, source relationship, quality issue node, and coverage status node all use different types of identifiers and do not update the source snapshot by overwriting the original value.

[0211] In one deployment method, the source snapshot parsing module and the observation node standardization module are located in the data compilation service, the merge verification module is located in the quality verification service, and the source graph formation module is located in the graph storage service. Each service transmits node identifiers, version information, and evidence indexes through a controlled data interface. In another deployment method, the modules can run as functional components within the same software process. This invention does not limit the specific deployment topology.

[0212] In this embodiment of the invention, the compilation messages transmitted between modules include at least a task identifier, an object under examination identifier, an examination cycle identifier, a snapshot identifier, a version vector, a set of input node identifiers, and an expected output type. The message does not directly copy the entire source snapshot; the receiving module reads the required data from the controlled storage area based on the snapshot identifier and node identifier. After each module completes processing, it returns the output node identifier set, evidence index, and processing status, ensuring that the module boundaries align with the data object boundaries in the source graph.

[0213] Specifically, the system can use a task state machine to coordinate four modules. Task states sequentially include: snapshot confirmed, skeleton formed, observation nodes standardized, merge verification completed, source graph submitted, and task published. Each module executes only if its preceding states and version vectors meet the requirements; when processing fails, the failing module, input node identifier, and error reason are recorded, and the previously submitted graph version remains readable. When retrying a task using the same idempotent key, the successful preceding output is reused, and execution continues from the failed module, avoiding duplicate node generation.

[0214] Preferably, when submitting a graph version, the source graph formation module verifies whether all evidence indices output by the merge verification module are accessible, and verifies whether the endpoints of blocking, restricting, and contradictory relationships belong to the current snapshot boundary. If the verification passes, the system writes graph version metadata and switches the current version pointer in the same submission operation; if the verification fails, the pointer is not switched. Therefore, even if the modules are deployed in a distributed manner, the reader observes a complete and version-consistent source graph, rather than an intermediate state where processing across modules is only halfway complete.

[0215] Optionally, the system can also provide a compilation replay interface. The replay request specifies an immutable inspection source snapshot and version vector. The four modules load the specified version's interface pattern, mapping template, quality shape, and deterministic comparison operator without reading the current mutable configuration, and regenerate the initial source graph of the inspected object. The replay result is compared with the historical graph version in terms of node identification, derived attributes, and evidentiary relationships to confirm the determinism of versioned compilation.

[0216] In a comprehensive implementation scenario, the data interface submits structured physical examination data from multiple departments for a single examination cycle for a given subject. The source snapshot parsing module performs deterministic serialization and generates an immutable physical examination source snapshot, subsequently converting the department, examination group, summary, and detail levels into a subject source graph skeleton. For a composite field jointly expressed by result value, unit, method, and location, the observation node standardization module establishes original observation nodes and generates field addresses and content hashes, then rolls up the field into a local concept cluster.

[0217] The concept cluster candidate recall yields two standard concept candidates with similar names. Field drill-down verification reveals that the first candidate's unit dimension, method, and location are compatible with the source field, while the second candidate's method constraint fails. Therefore, only the first candidate enters the hard constraint pass candidate set. The system generates pass or fail evidence for each constraint, forming a standardized observation node when the first candidate can be uniquely identified, while retaining the reason for excluding the second candidate. The mapping result can thus be used to progressively return from the standard concepts to the specific fields in the concept cluster members and the immutable variant detection source snapshot.

[0218] The merge verification module initiates a bounded one-hop trajectory from the summary node. The trajectory plan first visits the same check group, then the detailed observation node and the standard concept node, and finally verifies the method, location, and side. Each hop candidate does not exceed the trajectory plan's upper limit, and the endpoint satisfies the target signature, generating a source relationship between the summary and the details. The trajectory record saves the starting node, relationship type, candidate set, and retention reason for each hop, allowing the source relationship to be reviewed independently of soft similarity retrieval.

[0219] The quality shape component then selects the standardized observation node and performs consistency verification for result type, unit dimension, source relation, and summary detail, respectively. The node generates evidence of satisfaction in the integrity and source availability dimensions, but a newly added auxiliary relation type in the interface pattern is not yet covered by any quality shape. Therefore, the system generates a node with insufficient verification coverage, instead of recording it as a content error. The quality coverage matrix records both verified and uncovered dimensions. The institutional policy triggers a rule configuration request and causes the corresponding node to carry an insufficient verification flag.

[0220] The source graph formation module applies deterministic comparison operators only to standardized observation nodes that are not blocked, forming deterministic discovery nodes. Then, snapshot nodes, source graph skeleton nodes, original observation nodes, standardized observation nodes, summary nodes, finite one-hop mapping trajectories, quality issue nodes, insufficient verification coverage state nodes, and deterministic discovery nodes are submitted as the same graph version. This integrated scenario demonstrates that hierarchical pattern compilation solves the candidate search and hard attribute compatibility problems, bounded one-hop trajectories provide stepwise evidence of mapping relationships, and the dual-layer output of quality shape and coverage status distinguishes between content verification results and rule capability boundaries. All three share field-level source anchoring without substituting for each other.

[0221] The interface pattern version, mapping template version, quality shape version, quality rule version, and compiler version described in this specification can all be managed using immutable version identifiers. A single compilation task should have a fixed version vector; if a version is updated during task execution, the new version only applies to subsequent tasks or re-initiated compilation tasks, avoiding the mixing of different interpretations in the initial source graph of the same inspected object.

Claims

1. A method for compiling and quality node-based mapping of physical examination data source graphs, characterized in that, include: Based on the collected multi-departmental physical examination structured data, an invariant sample source snapshot is generated, and the invariant sample source snapshot is parsed to obtain the skeleton of the source map of the examined object; Based on the source graph skeleton of the inspected object, the field address, content hash, source version and node identifier are configured for the result field to obtain the original observation node set. Concept mapping, unit conversion and reference range structuring are performed on the original observation node set to obtain the standardized observation node set in which the original value and the standardized value coexist. Based on the standardized observation node set, summary detail merging is performed to obtain the source relationship between the summary and the detail; and a node-by-node quality verification is performed on the standardized observation node set using a versioned quality shape set, generating a quality problem node set based on the verification results; the evidence from the node-by-node quality verification is aggregated to obtain a quality coverage matrix, which records verification pass status, verification failure status, and uncovered status; for the uncovered status, nodes with insufficient verification coverage are generated, and the nodes with insufficient verification coverage are stored hierarchically with the quality problem nodes. The summary detail merging based on the standardized observation node set includes: generating a trajectory plan containing query stages, maximum hop count, and upper limit of candidates per hop based on the field node to be matched and the target relation signature; the trajectory plan limits the execution order of the structural query stage, semantic query stage, and attribute verification stage according to the target relation signature; and executing structural adjacency one-hop query, concept adjacency one-hop query, and attribute constraint one-hop query sequentially according to the trajectory plan. A local candidate sequence is obtained; allowable relation signatures and exclusion rules are applied to the local candidate sequence hop by hop; when the upper limit of candidates per hop is reached at any query stage, candidates are retained according to the hard constraints of quantity and structural adjacency distance; the query is terminated when the number of executed hops reaches the maximum number of hops, the current candidate set is empty, or the trajectory endpoint satisfies the target relation signature, and the reasons for retaining, excluding, and terminating the candidates are recorded; the starting node, relation type, candidate set, selected node, and reason code are recorded for each query; the records of each query are linked into a finite one-hop mapping trajectory according to the execution order; a field fingerprint and version vector are generated based on the content hash, field address, interface mode version, and mapping template version of the field to be matched; the finite one-hop mapping trajectory is saved with the field fingerprint and version vector as the cache key, and the finite one-hop mapping trajectory is reused when the cache key is consistent; the trajectory endpoint of the finite one-hop mapping trajectory is verified to satisfy the target relation signature, and a summary detail candidate correspondence relationship with the finite one-hop mapping trajectory is obtained; Based on the standardized set of observation nodes, a deterministic set of findings is generated that does not exceed the semantics of the original results. By combining the source relationship and the set of quality problem nodes, a restrictive relationship, a blocking relationship, or a contradictory relationship is established to form an initial source map and a quality status view of the inspected object.

2. The method according to claim 1, characterized in that, The process of performing concept mapping, unit conversion, and reference range structuring on the original set of observation nodes to obtain a standardized set of observation nodes containing both original and standardized values ​​includes: According to the parent path, department, inspection group, field level and detection method, the field fragments in the original set of observation nodes are rolled up into local concept clusters to obtain a set of local concept clusters; The concept cluster descriptions in the local concept cluster set are compared with the standard project family, the institutional project dictionary and the existing mapping template to obtain the standard concept candidate set; Drill down the candidate concepts in the standard concept candidate set to the corresponding result fields, perform field hard constraint verification, and obtain the hard constraint passed candidate set; The hard constraints are used to perform semantic matching and sorting within the candidate set. When the sorting result can uniquely determine the candidate, the standardized observation node set is generated. When the sorting result cannot uniquely determine the candidate, the mapping conflict node is generated.

3. The method according to claim 2, characterized in that, The step of drilling down the candidate concepts in the standard concept candidate set to the corresponding result fields, performing field hard constraint verification, and obtaining a hard constraint-passing candidate set includes: Extract necessary attribute signatures from the conceptual definitions of candidate concepts; The necessary attribute signature is broken down into value type constraint, unit dimension constraint, sample type constraint, detection method constraint, location constraint, and side constraint. Read the original attribute corresponding to the necessary attribute signature according to the field address of the result field; The value type constraint, unit dimension constraint, sample type constraint, detection method constraint, location constraint, and lateral constraint are applied to each of the original attributes; Record the pass status, failure reason code, and evidence field address for each constraint; Only candidate concepts whose hard constraints are all passed are added to the hard constraint pass candidate set; When the hard constraint passes through an empty candidate set, an attribute incompatible mapping conflict node is generated; when the hard constraint passes through a candidate set containing at least two indistinguishable candidate concepts, a candidate non-unique mapping conflict node is generated.

4. The method according to claim 1, characterized in that, The process involves: performing summary detail merging based on the standardized observation node set to obtain the source relationship between the summary and the detail; performing node-by-node quality verification on the standardized observation node set using a versioned quality shape set, and generating a quality problem node set based on the verification results; aggregating the evidence from the node-by-node quality verification to obtain a quality coverage matrix, which records verification pass status, verification failure status, and uncovered status; generating insufficient verification coverage status nodes for the uncovered status; and storing the insufficient verification coverage status nodes and quality problem nodes hierarchically, including: Based on interface pattern versions and quality rules, the quality dimensions of integrity, consistency, uniqueness, temporal sufficiency, and source availability are compiled into the versioned quality shape set. Configure an applicable selector, node shape, attribute constraints, severity level, and failure output template for each quality shape in the versioned quality shape set; The applicable nodes are determined from the standardized set of observation nodes using the applicable selector; Perform attribute verification, cardinality verification, type verification, relationship verification, and cross-node consistency verification on the applicable nodes; For applicable nodes that pass verification, generate node-by-node satisfaction evidence; for applicable nodes that fail verification, generate node-by-node failure evidence. The set of quality problem nodes is generated based on the node-by-node failure evidence and the failure output template.

5. The method according to claim 4, characterized in that, The method further includes: The quality coverage matrix is ​​obtained by aggregating the node-by-node satisfaction evidence and the node-by-node failure evidence according to the quality dimension, field path, quality shape version, and severity level. The quality coverage matrix records the verification pass status, the verification failure status, and the uncovered status, respectively. Identify the field type or relation type that is not covered by any applicable quality shape based on the uncovered state; Generate the insufficient validation coverage state node for the field type or relation type that is not covered by any applicable quality shape; The nodes with insufficient verification coverage and the quality problem nodes generated based on the node-by-node failure evidence are stored in a hierarchical manner. According to the organization's policy, the node with insufficient verification coverage triggers a manual configuration request or a downgrade processing flag, and the node with insufficient verification coverage is prohibited from triggering a hard blocking relationship by default. Write the quality coverage matrix and the verification coverage insufficiency status nodes into the quality status view.

6. The method according to claim 1, characterized in that, The process of generating an invariant specimen source snapshot based on the collected multi-departmental structured physical examination data, and parsing the invariant specimen source snapshot to obtain the subject source map skeleton, includes: Calculate an overall summary of the structured physical examination data from the multiple departments; The overall summary is bound with the interface mode version, the inspected object identifier, and the physical examination cycle identifier to form the immutable physical examination source snapshot; After verifying the overall summary of the immutable source snapshot, the case level, cycle level, department level, examination group level, summary level, and detail level are parsed. Isolate client fields that do not participate in physical examination data processing based on a preset whitelist; The hierarchical parsing results are converted into graph skeleton nodes with parent-child edges to obtain the source graph skeleton of the inspected object; When performing concept mapping, unit conversion, and reference range structuring on the original set of observation nodes, the standardized attributes are written as derived attributes into the corresponding original observation nodes, and the derived attributes are prohibited from overriding the original values ​​and original units.

7. The method according to claim 1, characterized in that, The process of performing summary detail merging and quality rule verification based on the standardized observation node set yields the source relationship between the summary and the details, as well as a set of quality problem nodes, including: Construct summary detail matching keys based on department, project joint identifier, standard concept, testing method, location, side, and results; Generate candidate correspondences for summary details based on the summary detail matching key; Perform consistency checks, source version checks, and duplicate suppression on the candidate correspondences of the summary details; If the verification passes, a summary pointing to the source relationship of the details is generated; if the verification fails, a summary detail conflict node is generated. The quality rules are used to identify missing fields, abandoned projects, unavailable sources, insufficient units, conflicting methods and results with the same name, and generate the set of quality problem nodes. A deterministic comparison operator is applied to the standardized observation nodes that are not restricted or blocked by the set of quality problem nodes to generate the deterministic discovery set; The summary discovery and detailed discovery corresponding to the same standardized observation node are merged into a deterministic discovery node, while preserving the source relationship of the summary discovery and the detailed discovery.

8. A system for compiling and quality node-based mapping of physical examination data source graphs, using the method for compiling and quality node-based mapping of physical examination data source graphs as described in any one of claims 1-7, characterized in that, include: The source snapshot parsing module is used to generate an invariant physical examination source snapshot based on the collected multi-departmental physical examination structured data, and to parse the invariant physical examination source snapshot to obtain the source map skeleton of the examined object; The observation node standardization module is used to configure field addresses, content hashes, source versions, and node identifiers for result fields based on the source graph skeleton of the inspected object, to obtain an original set of observation nodes. The original set of observation nodes is then subjected to concept mapping, unit conversion, and reference range structuring to obtain a standardized set of observation nodes in which original values ​​and standardized values ​​coexist. The merge verification module is used to perform summary detail merging based on the standardized observation node set to obtain the source relationship between the summary and the detail; and to perform node-by-node quality verification on the standardized observation node set using a versioned quality shape set, and generate a quality problem node set based on the verification results; aggregate the evidence from the node-by-node quality verification to obtain a quality coverage matrix, the quality coverage matrix records the verification pass state, verification failure state, and uncovered state, generates verification coverage insufficient state nodes for the uncovered state, and stores the verification coverage insufficient state nodes and quality problem nodes hierarchically; The source graph formation module is used to generate a deterministic discovery set that does not exceed the semantics of the original result based on the standardized set of observation nodes, and to establish restrictive, blocking, or contradictory relationships by combining the source relationship and the set of quality problem nodes, thereby forming an initial source graph and quality status view of the inspected object.

Citation Information

Patent Citations

  • Multi-source heterogeneous gene associated data index construction and retrieval method

    CN122493994A

  • Health check path evaluation indicator building system, method thereof, device therewith, and computer program product therein

    TW201426620A