A human body clinical information evaluation method based on a knowledge graph

Abstract

This invention discloses a knowledge graph-based method for evaluating human clinical information, comprising the following steps: collecting human clinical information of a target individual to generate a standardized clinical dataset; performing medical semantic extraction to generate a set of medical semantic elements; constructing a patient-level evaluation knowledge graph for the target individual to generate a clinical efficacy structure; identifying risk endpoint nodes, constructing an Even-Shiloach tree to form an Even-Shiloach forest; identifying a set of invalid edges, deleting the edges corresponding to the invalid edge set to obtain an updated Even-Shiloach forest; generating a hierarchical difference set and identifying the affected subgraphs, performing local evaluation updates to generate corresponding evidence chains; determining the initial evaluation result of the human clinical information, correcting the initial evaluation result to obtain a corrected evaluation result. This invention achieves dynamic updability and traceable evidence chains in the evaluation of human clinical information.

Description

Technical Field

[0001] This invention relates to the field of smart healthcare technology, and in particular to a method for evaluating human clinical information based on knowledge graphs. Background Technology

[0002] With the widespread adoption of hospital information systems and smart healthcare systems, human clinical information is continuously generated and accumulated across multiple sources, including electronic medical records, laboratory test results, imaging reports, vital sign monitoring, diagnostic records, medication records, and treatment records. Furthermore, the clinical information of the same individual often exhibits characteristics such as heterogeneous sources, inconsistent formats, long time spans, and complex semantic relationships. Current technologies, to improve the efficiency of clinical information utilization, often employ a combination of structured data standardization and unstructured text information extraction. This involves extracting information such as diagnoses, symptoms, signs, examination findings, and medications to form semantic elements, and then constructing clinical knowledge representations or knowledge graphs based on these elements for risk assessment, decision support, or early warning output.

[0003] However, existing technologies still have significant shortcomings. Clinical information is prone to problems such as inconsistencies in fields, units, time alignment difficulties, and the mixing of duplicate and unresolvable records during the collection and standardization process. Although some systems can complete data cleaning and normalization, they lack a quality record organization method that is traceable for subsequent assessment, making it difficult to guarantee the reliability and consistency of assessment input. Existing knowledge graph-based assessments usually focus on one-time graph construction and static reasoning or model calculation. When faced with dynamic changes in clinical facts, such as time window expiration, evidence invalidation due to laboratory review or imaging correction, and conflicting substitution between similar evidence, conclusions are often updated by full recalculation or re-retrieval. It is difficult to achieve efficient incremental maintenance in edge deletion scenarios, resulting in high computational overhead, slow response, and difficulty in stable operation when assessing multiple risk endpoints in parallel.

[0004] Therefore, how to provide a knowledge graph-based method for assessing human clinical information is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a knowledge graph-based method for evaluating human clinical information. This invention introduces a clinical efficacy structure and a failure edge deletion mechanism into the patient-level evaluation knowledge graph. It combines risk endpoint-anchored Even-Shiloach forest decreasing update with hierarchical differential-driven local evaluation update to achieve real-time correction of evaluation results as clinical evidence changes and output a traceable evidence chain, thereby improving the real-time performance, accuracy, and interpretability of the evaluation.

[0006] A knowledge graph-based method for evaluating human clinical information according to an embodiment of the present invention includes the following steps: Collect human clinical information of the target individual, perform standardized preprocessing, and generate standardized clinical datasets; Perform medical semantic extraction on human clinical information to generate a set of medical semantic elements; Based on standardized clinical datasets and medical semantic element sets, a patient-level assessment knowledge graph for the target individual is constructed, and a clinical efficacy structure is generated for each edge. In the patient-level assessment knowledge graph, risk endpoint nodes are identified, and an Even-Shiloach tree is constructed using each risk endpoint node as the source point to form an Even-Shiloach forest. The set of failed edges is determined according to the clinical efficacy failure judgment rules. The edges corresponding to the set of failed edges are deleted from the patient-level assessment knowledge graph, and the Even-Shiloach forest is triggered to reduce the time, resulting in the updated Even-Shiloach forest. Based on the updated Even-Shiloach forest, a hierarchical difference set is generated and the affected subgraphs are identified. Local evaluation updates are performed on the affected subgraphs to generate the corresponding evidence chains. The evidence certificate structure is generated based on the evidence chain. The initial evaluation result of human clinical information is determined based on the evidence certificate structure. When the rebuttal trigger condition is met, a rebuttal event is generated. The initial evaluation result is corrected based on the rebuttal event to obtain the corrected evaluation result.

[0007] Optionally, the generation of the standardized clinical dataset specifically includes: Collect human clinical information of target individuals within a preset time range, perform record-level normalization processing, and generate a record-level normalized clinical information set and a data quality record set. Perform medical terminology standardization processing on the record-level normalized clinical information set to generate a terminology mapping record table, and at the same time generate the terminology-normalized clinical information set, and incorporate the terminology mapping record table into the data quality record set; The terminology-standardized clinical information set is subjected to metrological and temporal standardization. After the metrological and temporal standardization is completed, a standardized clinical dataset is generated, and the data quality record set is stored together as the quality control record of the standardized clinical dataset.

[0008] Optionally, the generation of the medical semantic element set specifically includes: Semantic extraction preprocessing is performed on unstructured clinical text in human clinical information. The text is normalized, and it is segmented into sentence sequences according to preset sentence segmentation rules. Each sentence is segmented to generate a word sequence. Each word is normalized and mapped to a standard word identifier. The start and end position indices of each word are recorded to generate the text preprocessing result. Perform medical entity extraction based on the text preprocessing results to generate a set of medical entity records; Entity relation extraction is performed based on the lexical sequence and the medical entity record set. Entity pairs in the medical entity record set are candidate-paired to generate an entity pair set. For each entity pair, the corresponding context fragment is located in the lexical sequence to generate a relation determination context. For each entity pair, relation trigger words are identified and relation direction is determined based on the relation determination context. For each entity pair, relation type is determined and relation type identifier is generated. For each entity pair, relation confidence is calculated and relation confidence value is generated. For each entity pair, time information is associated and relation time identifier is generated. The entity pair identifier, relation type identifier, relation direction identifier, relation confidence value, and relation time identifier are combined to generate an entity relation record set. The medical entity record set and the entity relation record set are combined to generate a medical semantic element set.

[0009] Optionally, obtaining the patient-level assessment knowledge graph and clinical efficacy structure specifically includes: Perform clinical event processing on standardized clinical datasets to generate clinical event record sets; Perform record processing on the medical semantic element set, assign a unique identifier to each semantic element and create an index, record the type information of the semantic elements, and generate a semantic element record set; Based on the clinical event record set and semantic element record set, a unique node identifier is assigned and target individual nodes, event nodes and semantic nodes are generated according to the preset node generation rules. Individual event association edges are generated between target individual nodes and event nodes according to the preset edge connection rules. Event evidence association edges are generated between event nodes and semantic nodes according to the preset edge connection rules. Semantic relationship edges are generated between semantic nodes according to the relationship indication information carried in the semantic element record set, thus obtaining a patient-level assessment knowledge graph. For each edge in the edge set of the patient-level assessment knowledge graph, the effective time window interval is calculated based on the relation time identifier according to the preset time window rules, and the conflict anchor point identifier is determined according to the relation type identifier of the edge. The revision record in the standardized clinical dataset is used to match the revision record for the edge to generate a revision token value. The preset validity state is set as the initial validity state. The effective time window interval, conflict anchor point identifier, revision token value and initial validity state are written and generated into a one-to-one bound clinical validity structure.

[0010] Optionally, the formation of the Even-Shiloach forest specifically includes: Read the preset risk label set, retrieve the corresponding risk label node in the node set, and create and write the node if it is not found. Gather the corresponding risk label nodes to form a risk endpoint node set, and assign a unique endpoint identifier to each node in the risk endpoint node set and establish an endpoint index. For each risk endpoint node in the risk endpoint node set, edges that are directly or indirectly connected to the current risk endpoint node are selected from the edge set of the patient-level assessment knowledge graph as candidate edge sets. The candidate edge sets are selected according to the edge type selection rules, and the edges used for evidence propagation are retained and the edges that do not participate in the assessment are removed. The retained edges are selected according to the time validity selection rules, and the edges that exceed the valid time range are removed, thus forming a valid edge set. The nodes associated with the valid edge set are written into the traversable node set, and a traversable subgraph is formed based on the traversable node set and the valid edge set. For each node in the traversable subgraph, mark the level distance as unreachable, set the parent pointer to null, set the level distance of the current risk endpoint node to zero and write it to the level distance record, write the current risk endpoint node to the queue to be expanded, retrieve the current node from the queue to be expanded and sequentially traverse the adjacent edges to locate the neighboring nodes, set the level distance of the neighboring node in the unreachable state to the current node's level distance plus one and write it to the level distance record, set the parent pointer of the neighboring node to the current node and record the parent edge, write the neighboring node to the queue to be expanded, and repeat the operations of retrieving the queue node, traversing the adjacent edges, writing the level distance, writing the parent pointer and recording the parent edge until the queue to be expanded is empty, and form an Even-Shiloach tree with the risk endpoint node as the root based on the level distance record and the parent pointer record; A forest index is built for the Even-Shiloach tree corresponding to each risk endpoint node in the risk endpoint node set. The endpoint identifier of each risk endpoint node is associated with the corresponding root node identifier and the hierarchical distance record is associated with the parent pointer record. All registration results in the risk endpoint node set are aggregated to form the Even-Shiloach forest.

[0011] Optionally, obtaining the updated Even-Shiloach forest specifically includes: Read the edge set of the patient-level assessment knowledge graph and read the clinical efficacy structure bound to each edge, obtain the current assessment time, and initialize the set of invalid edges to an empty set; Failure determination is performed edge by edge in the edge set. The effective time window interval bound to the edge is read and compared with whether the current evaluation time falls within the effective time window interval. If the current evaluation time does not fall within the effective time window interval, the edge is written into the failed edge set. The revision record of the standardized clinical dataset is retrieved and the revision token value is matched for the edge. At the same time, a consistency comparison is performed with the revision token value bound to the edge. If the consistency comparison result is inconsistent, the edge is written into the failed edge set. The conflict anchor point identifier bound to the edge is read and the candidate alternative edge with the same conflict anchor point identifier is retrieved in the edge set. The effective time window interval of the candidate alternative edge is read and the candidate alternative edge that does not fall within the effective time window interval at the current evaluation time is filtered out. After filtering, if the candidate alternative edge exists, the edge is written into the failed edge set. Remove the edges corresponding to the invalid edge set from the edge set of the patient-level assessment knowledge graph and generate a deletion list. Write the deletion list into the deletion record set and establish an association with the invalid edge set. Trigger the decreasing update of the Even-Shiloach forest anchored to the risk endpoint and initialize the update support structure. Read the level distance record and parent pointer record of the Even-Shiloach tree corresponding to each risk endpoint root. Read the current parent edge identifier of each node and write it into the parent edge dependency index. Map the parent edge identifier to the set of nodes that depend on the parent edge and register them in groups according to the risk endpoint root. Complete the initialization of the parent edge dependency index. The parent edge dependency index is used to perform impact localization on the edge deletion list. The deleted edge identifier is read one by one in the edge deletion list. The set of nodes that depend on the deleted edge identifier is retrieved in the parent edge dependency index. If the node set is not empty, the node is written to the corresponding queue to be updated according to the risk endpoint node identifier and the node level distance is marked as to be recalculated. If the node set is empty, it is not written to the queue to be updated. The set of nodes to be updated is limited to the set of nodes whose parent edge dependency index is hit. The queue to be updated is updated in a decremental manner. The nodes to be updated are retrieved in a loop, the adjacent edges of the nodes to be updated are traversed, the edges in the deletion record are filtered, and the candidate parent edge is searched in the remaining adjacent edges. When the candidate parent edge exists, the level distance of the starting node of the candidate parent edge is read, the level distance of the candidate parent node is incremented by one and written into the level distance record of the node to be updated, the candidate parent edge is written into the parent pointer record of the node to be updated, and the parent edge dependency index is updated. When the candidate parent edge is empty, the level distance of the node to be updated is marked as unreached, the parent pointer record of the node to be updated is set to null, and the parent edge dependency index is updated. The child nodes of the node to be updated are written into the queue to be updated, and the level distance of the child nodes is marked as to be recalculated. When the queue to be updated is empty, the level distance record and the parent pointer record are combined to form the updated Even-Shiloach forest. Optionally, the generation of the chain of evidence specifically includes: Read the updated Even-Shiloach forest, read the level distance value before and after the update for each node in the node set, and perform a level difference judgment operation. Mark the nodes whose level distance values ​​before and after the update are inconsistent as level change nodes and write them into the level difference set. The affected node set and the affected subgraph are determined based on the hierarchical difference set. The node corresponding to the hierarchical difference set is located in the node set of the patient-level assessment knowledge graph and written into the affected node set. The edge adjacent to the node in the affected node set is retrieved in the edge set of the patient-level assessment knowledge graph and written into the affected edge set. The affected subgraph is formed based on the affected node set and the affected edge set. Perform a local evaluation update on the affected subgraph, read the set of risk endpoint nodes in the updated Even-Shiloach forest, filter the affected nodes connected to the risk endpoint nodes in the affected subgraph and generate a set of evidence nodes, record the backtracking of each evidence node in the evidence node set along the parent pointer to the risk endpoint node, and record the node identifier sequence and edge identifier sequence traversed during the backtracking, and collect the node identifier sequence and edge identifier sequence corresponding to the risk endpoint node to generate an evidence chain.

[0012] Optionally, obtaining the corrected evaluation result specifically includes: Read the evidence chain and group it according to the risk endpoint node identifier. Write the grouping results into the evidence chain record table and assign evidence chain identifiers to form an evidence chain set. Based on the evidence chain set, edge identifiers are extracted to form a core evidence edge set. Candidate alternative edges that share the start node identifier and end node identifier with the core evidence edges are retrieved from the edge set of the patient-level assessment knowledge graph. Invalid candidate alternative edges at the current assessment time are screened out. An alternative mapping relationship is established based on the core evidence edge set and the retained candidate alternative edges to form an evidence certificate structure. Based on the evidence certificate structure and the deletion list, the existence of core evidence is determined. The core evidence in the existing state is weighted and accumulated according to the preset weight table to obtain the risk score value. The risk level identifier is then mapped according to the preset score threshold table to generate the initial assessment result. Based on the evidence certificate structure and the deletion list, a disproving evidence trigger determination is performed. The core evidence edge set is read edge by edge, and it is checked whether the core evidence edge appears in the deletion list. For the core evidence edge that appears in the deletion list, it is checked whether the alternative mapping relationship provides a valid alternative edge. When the core evidence edge appears in the deletion list and the alternative mapping relationship does not provide a valid alternative edge, a disproving evidence event is generated and written to the disproving evidence event record table. Based on the disproving evidence event, the initial evaluation result is corrected, and a corrected evaluation result is generated.

[0013] The beneficial effects of this invention are: This invention standardizes and preprocesses the clinical information of target individuals to generate a standardized clinical dataset. At the same time, it extracts medical semantics from unstructured clinical text to form a set of medical semantic elements. This enables multi-source heterogeneous clinical information to be expressed in a consistent manner in terms of fields, terms, units, and time dimensions, thereby improving the completeness and consistency of patient-level clinical information aggregation and reducing evaluation bias caused by differences in data format, duplicate records, and inconsistent time alignment.

[0014] This invention generates a clinical efficacy structure for edges in the patient-level assessment knowledge graph, determines the set of invalid edges according to the clinical efficacy invalidation judgment rules, and performs edge deletion processing, thereby triggering an Even-Shiloach forest decreasing update oriented towards risk endpoints. This enables the graph to achieve local maintenance and rapid response in edge deletion scenarios caused by evidence expiration, revision, and conflict substitution, avoiding the high computational overhead and update lag caused by full recalculation common in existing technologies, and improving dynamic update efficiency and assessment timeliness.

[0015] This invention generates a hierarchical difference set based on the updated Even-Shiloach forest and identifies the affected subgraphs. It performs local evaluation updates only on the affected subgraphs and generates a chain of evidence. Simultaneously, it forms an evidence certificate structure based on the chain of evidence and corrects the initial evaluation results when the conditions for rebuttal are met. This allows for the output of corrected evaluation results while maintaining the traceability of the evaluation basis and the verifiability of the conclusions, enhancing the interpretability and auditability of the evaluation process, and improving the credibility and usability for clinical use. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a knowledge graph-based human clinical information evaluation method proposed in this invention; Figure 2 This diagram illustrates the Even-Shiloach forest decremental update and evidence chain generation of a knowledge graph-based human clinical information assessment method proposed in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figures 1-2 A knowledge graph-based method for evaluating human clinical information includes the following steps: Collect human clinical information of the target individual, perform standardized preprocessing, and generate standardized clinical datasets; Perform medical semantic extraction on human clinical information to generate a set of medical semantic elements; Based on standardized clinical datasets and medical semantic element sets, a patient-level assessment knowledge graph for the target individual is constructed, and a clinical efficacy structure is generated for each edge. In the patient-level assessment knowledge graph, risk endpoint nodes are identified, and an Even-Shiloach tree is constructed using each risk endpoint node as the source point to form an Even-Shiloach forest. The set of failed edges is determined according to the clinical efficacy failure judgment rules. The edges corresponding to the set of failed edges are deleted from the patient-level assessment knowledge graph, and the Even-Shiloach forest is triggered to reduce the time, resulting in the updated Even-Shiloach forest. Based on the updated Even-Shiloach forest, a hierarchical difference set is generated and the affected subgraphs are identified. Local evaluation updates are performed on the affected subgraphs to generate the corresponding evidence chains. The evidence certificate structure is generated based on the evidence chain. The initial evaluation result of human clinical information is determined based on the evidence certificate structure. When the rebuttal trigger condition is met, a rebuttal event is generated. The initial evaluation result is corrected based on the rebuttal event to obtain the corrected evaluation result.

[0019] In this embodiment, the generation of standardized clinical datasets specifically includes: The system collects human clinical information of target individuals within a preset time range and performs record-level normalization processing. Record-level normalization processing includes mapping field names from various data sources to uniform field names to form a uniform field set, unifying field value formats to a preset format, assigning missing values ​​to missing values ​​in numerical fields, performing a full match on records based on a preset primary key field set to determine duplicate records and retain only one unique record, cleaning up abnormal characters and blank characters, removing unparseable records and generating a list of removed records, and generating a record-level normalized clinical information set and a data quality record set. The data quality record set includes a deduplication record list and a list of removed records. The clinical information set after record-level standardization is subjected to medical terminology standardization processing. The medical terminology standardization processing includes segmenting and mapping diagnosis-related content, symptom and sign-related content, examination findings-related content, and drug-related content into unified terms; performing unified terminology replacement on synonyms, abbreviations, and alternative names; performing standardization processing on different spellings of the same term; writing the replaced unified terms back to the corresponding fields; generating a terminology mapping record table; generating a clinical information set after terminology standardization; and incorporating the terminology mapping record table into the data quality record set. The standardized clinical information set is subjected to metrological and temporal standardization. This process includes converting numerical index fields to uniform units to form a unified unit of measurement, retaining the original and converted values ​​and their corresponding units, converting all time fields to a uniform time format and generating timestamps, sorting records with timestamps in ascending order by timestamp, and sorting records with the same timestamp according to a preset priority rule. After completing the metrological and temporal standardization, a standardized clinical dataset is generated, and the data quality record set is stored together as the quality control record of the standardized clinical dataset.

[0020] In this embodiment, the generation of the medical semantic element set specifically includes: Semantic extraction preprocessing is performed on unstructured clinical text in human clinical information. The text is normalized, and it is segmented into sentence sequences according to preset sentence segmentation rules. Each sentence is segmented to generate a word sequence. Each word is normalized and mapped to a standard word identifier. The start and end position indices of each word are recorded to generate the text preprocessing result. Based on the text preprocessing results, a medical entity extraction operation is performed. The medical entity extraction operation identifies entity candidate segments and determines entity boundaries in the word sequence. For each entity candidate segment, an entity type determination is performed and an entity type identifier is generated. For each entity candidate segment, entity value standardization is performed and a standardized entity value is generated. For each entity candidate segment, entity confidence is calculated and an entity confidence value is generated. For each entity candidate segment, its corresponding start position index and end position index are recorded. The standardized entity value, entity type identifier, entity confidence value, start position index, and end position index are combined to generate a medical entity record set. The medical entity record set includes different medical entity records. Each medical entity record contains an entity identifier, entity type identifier, entity value, entity position index, and entity confidence. Entity relation extraction is performed based on the lexical sequence and the medical entity record set. Entity pairs in the medical entity record set are candidate-paired to generate an entity pair set. For each entity pair, the corresponding context fragment is located in the lexical sequence to generate a relation determination context. For each entity pair, relation trigger words are identified and relation direction is determined based on the relation determination context. For each entity pair, relation type is determined and relation type identifier is generated. For each entity pair, relation confidence is calculated and relation confidence value is generated. For each entity pair, time information is associated and relation time identifier is generated. The entity pair identifier, relation type identifier, relation direction identifier, relation confidence value, and relation time identifier are combined to generate an entity relation record set. The medical entity record set and the entity relation record set are combined to generate a medical semantic element set.

[0021] In this embodiment, the acquisition of the patient-level assessment knowledge graph and clinical efficacy structure specifically includes: Clinical event processing is performed on the standardized clinical dataset. The clinical event processing group each standardized record into a clinical event according to the preset event type rules, associates each clinical event with a target individual identifier and generates an event object identifier, writes the event status or event value into the event status field, converts the event occurrence time into an event timestamp and writes it into the event timestamp field, writes the event source into the event source field, and generates a set of clinical event records. Perform record processing on the medical semantic element set, assign a unique identifier to each semantic element and create an index, record the type information of the semantic elements, and generate a semantic element record set; Based on the clinical event record set and semantic element record set, a unique node identifier is assigned and target individual nodes, event nodes and semantic nodes are generated according to the preset node generation rules. Individual event association edges are generated between target individual nodes and event nodes according to the preset edge connection rules. Event evidence association edges are generated between event nodes and semantic nodes according to the preset edge connection rules. Semantic relationship edges are generated between semantic nodes according to the relationship indication information carried in the semantic element record set, thus obtaining a patient-level assessment knowledge graph. For each edge in the edge set of the patient-level assessment knowledge graph, the effective time window interval is calculated based on the relation time identifier according to the preset time window rules, and the conflict anchor point identifier is determined according to the relation type identifier of the edge. The revision record in the standardized clinical dataset is used to match the revision record for the edge to generate a revision token value. The preset validity state is set as the initial validity state. The effective time window interval, conflict anchor point identifier, revision token value and initial validity state are written and generated into a one-to-one bound clinical validity structure.

[0022] In this embodiment, the formation of the Even-Shiloach forest specifically includes: Read the preset risk label set, retrieve the corresponding risk label node in the node set, and create and write the node if it is not found. Gather the corresponding risk label nodes to form a risk endpoint node set, and assign a unique endpoint identifier to each node in the risk endpoint node set and establish an endpoint index. For each risk endpoint node in the risk endpoint node set, edges that are directly or indirectly connected to the current risk endpoint node are selected from the edge set of the patient-level assessment knowledge graph as candidate edge sets. The candidate edge sets are selected according to the edge type selection rules, and the edges used for evidence propagation are retained and the edges that do not participate in the assessment are removed. The retained edges are selected according to the time validity selection rules, and the edges that exceed the valid time range are removed, thus forming a valid edge set. The nodes associated with the valid edge set are written into the traversable node set, and a traversable subgraph is formed based on the traversable node set and the valid edge set. For each node in the traversable subgraph, mark the level distance as unreachable, set the parent pointer to null, set the level distance of the current risk endpoint node to zero and write it to the level distance record, write the current risk endpoint node to the queue to be expanded, retrieve the current node from the queue to be expanded and sequentially traverse the adjacent edges to locate the neighboring nodes, set the level distance of the neighboring node in the unreachable state to the current node's level distance plus one and write it to the level distance record, set the parent pointer of the neighboring node to the current node and record the parent edge, write the neighboring node to the queue to be expanded, and repeat the operations of retrieving the queue node, traversing the adjacent edges, writing the level distance, writing the parent pointer and recording the parent edge until the queue to be expanded is empty, and form an Even-Shiloach tree with the risk endpoint node as the root based on the level distance record and the parent pointer record; A forest index is built for the Even-Shiloach tree corresponding to each risk endpoint node in the risk endpoint node set. The endpoint identifier of each risk endpoint node is associated with the corresponding root node identifier and the hierarchical distance record is associated with the parent pointer record. All registration results in the risk endpoint node set are aggregated to form the Even-Shiloach forest.

[0023] This invention identifies risk endpoints using risk labels and constructs an Even-Shiloach forest anchored to these risk endpoints within a patient-level assessment knowledge graph. By combining edge type and time validity filtering to form traversable subgraphs and maintaining hierarchical distance and parent pointer records, it enables unified accessibility characterization and evidence path organization for multiple risk endpoint assessments within the same graph. This improves the structure and consistency of risk assessment calculations, enhances the stability and traceability of evidence propagation and evidence chain generation, and reduces the computational overhead caused by repeated traversals.

[0024] In this embodiment, obtaining the updated Even-Shiloach forest specifically includes: Read the edge set of the patient-level assessment knowledge graph and read the clinical efficacy structure bound to each edge, obtain the current assessment time, and initialize the set of invalid edges to an empty set; Failure determination is performed edge by edge in the edge set. The effective time window interval bound to the edge is read and compared with whether the current evaluation time falls within the effective time window interval. If the current evaluation time does not fall within the effective time window interval, the edge is written into the failed edge set. The revision record of the standardized clinical dataset is retrieved and the revision token value is matched for the edge. At the same time, a consistency comparison is performed with the revision token value bound to the edge. If the consistency comparison result is inconsistent, the edge is written into the failed edge set. The conflict anchor point identifier bound to the edge is read and the candidate alternative edge with the same conflict anchor point identifier is retrieved in the edge set. The effective time window interval of the candidate alternative edge is read and the candidate alternative edge that does not fall within the effective time window interval at the current evaluation time is filtered out. After filtering, if the candidate alternative edge exists, the edge is written into the failed edge set. Remove the edges corresponding to the invalid edge set from the edge set of the patient-level assessment knowledge graph and generate a deletion list. Write the deletion list into the deletion record set and establish an association with the invalid edge set. Trigger the decreasing update of the Even-Shiloach forest anchored to the risk endpoint and initialize the update support structure. Read the level distance record and parent pointer record of the Even-Shiloach tree corresponding to each risk endpoint root. Read the current parent edge identifier of each node and write it into the parent edge dependency index. Map the parent edge identifier to the set of nodes that depend on the parent edge and register them in groups according to the risk endpoint root. Complete the initialization of the parent edge dependency index. The parent edge dependency index is used to perform impact localization on the edge deletion list. The deleted edge identifier is read one by one in the edge deletion list. The set of nodes that depend on the deleted edge identifier is retrieved in the parent edge dependency index. If the node set is not empty, the node is written to the corresponding queue to be updated according to the risk endpoint node identifier and the node level distance is marked as to be recalculated. If the node set is empty, it is not written to the queue to be updated. The set of nodes to be updated is limited to the set of nodes whose parent edge dependency index is hit. The queue to be updated is updated in a decremental manner. The nodes to be updated are retrieved in a loop, the adjacent edges of the nodes to be updated are traversed, the edges in the deletion record are filtered, and the candidate parent edge is searched in the remaining adjacent edges. When the candidate parent edge exists, the level distance of the starting node of the candidate parent edge is read, the level distance of the candidate parent node is incremented by one and written into the level distance record of the node to be updated, the candidate parent edge is written into the parent pointer record of the node to be updated, and the parent edge dependency index is updated. When the candidate parent edge is empty, the level distance of the node to be updated is marked as unreached, the parent pointer record of the node to be updated is set to null, and the parent edge dependency index is updated. The child nodes of the node to be updated are written into the queue to be updated, and the level distance of the child nodes is marked as to be recalculated. When the queue to be updated is empty, the level distance record and the parent pointer record are combined to form the updated Even-Shiloach forest. This invention identifies and removes evidence edges based on the clinical efficacy structure across time windows, revisions, and conflict substitutions. It combines this with an Even-Shiloach forest decreasing update mechanism anchored to risk endpoints, and utilizes parent edge dependency indexes to quickly locate the impact of edge removal. It only maintains local hierarchical distances and parent pointers for the affected node queue, thereby avoiding full recalculation and reducing update overhead. This improves the real-time performance, stability, and scalability of the clinical assessment atlas in scenarios with dynamic changes in evidence.

[0025] In this embodiment, the generation of the chain of evidence specifically includes: Read the updated Even-Shiloach forest, read the level distance value before and after the update for each node in the node set, and perform a level difference judgment operation. Mark the nodes whose level distance values ​​before and after the update are inconsistent as level change nodes and write them into the level difference set. The affected node set and the affected subgraph are determined based on the hierarchical difference set. The node corresponding to the hierarchical difference set is located in the node set of the patient-level assessment knowledge graph and written into the affected node set. The edge adjacent to the node in the affected node set is retrieved in the edge set of the patient-level assessment knowledge graph and written into the affected edge set. The affected subgraph is formed based on the affected node set and the affected edge set. Perform a local evaluation update on the affected subgraph, read the set of risk endpoint nodes in the updated Even-Shiloach forest, filter the affected nodes connected to the risk endpoint nodes in the affected subgraph and generate a set of evidence nodes, record the backtracking of each evidence node in the evidence node set along the parent pointer to the risk endpoint node, and record the node identifier sequence and edge identifier sequence traversed during the backtracking, and collect the node identifier sequence and edge identifier sequence corresponding to the risk endpoint node to generate an evidence chain.

[0026] In this embodiment, obtaining the corrected evaluation results specifically includes: Read the evidence chain and group it according to the risk endpoint node identifier. Write the grouping results into the evidence chain record table and assign evidence chain identifiers to form an evidence chain set. Based on the evidence chain set, edge identifiers are extracted to form a core evidence edge set. Candidate alternative edges that share the start node identifier and end node identifier with the core evidence edges are retrieved from the edge set of the patient-level assessment knowledge graph. Invalid candidate alternative edges at the current assessment time are screened out. An alternative mapping relationship is established based on the core evidence edge set and the retained candidate alternative edges to form an evidence certificate structure. Based on the evidence certificate structure and the deletion list, the existence of core evidence is determined. The core evidence in the existing state is weighted and accumulated according to the preset weight table to obtain the risk score value. The risk level identifier is then mapped according to the preset score threshold table to generate the initial assessment result. Based on the evidence certificate structure and the deletion list, a disproving evidence trigger determination is performed. The core evidence edge set is read edge by edge, and it is checked whether the core evidence edge appears in the deletion list. For the core evidence edge that appears in the deletion list, it is checked whether the alternative mapping relationship provides a valid alternative edge. When the core evidence edge appears in the deletion list and the alternative mapping relationship does not provide a valid alternative edge, a disproving evidence event is generated and written to the disproving evidence event record table. Based on the disproving evidence event, the initial assessment result is corrected. The correction operation includes setting the evidence existence mark of the core evidence edge involved in the disproving evidence event to a missing state, re-performing the weighted accumulation to obtain the corrected risk score value, mapping the corrected risk score value to obtain the corrected risk level identifier according to the preset scoring threshold table, and generating the corrected assessment result.

[0027] This invention organizes the evidence chain into an evidence chain set according to the risk endpoint node identification and extracts the core evidence edge set. Combined with candidate alternative edge screening and alternative mapping relationship to form an evidence certificate structure, the existence of evidence is determined for the core evidence edge and a risk score and risk level are generated accordingly. At the same time, when the core evidence edge fails and there is no effective alternative edge, a rebuttal event is triggered and the initial assessment result is corrected. This improves the robustness and accuracy of human clinical information assessment to evidence failure and conflict changes, enhances the traceability and interpretability of the assessment basis, reduces the risk of misjudgment due to the lack of key evidence, and improves the reliability and auditability of clinical use.

[0028] Example 1: To verify the feasibility of this invention in practice, it was applied to a risk assessment and early warning scenario for inpatients in a comprehensive medical institution. The core requirement of this scenario is to continuously assess the clinical information of inpatients and output multiple risk conclusions, such as infection deterioration risk, bleeding risk, and medication safety risk. When evidence changes due to laboratory review, image correction, fluctuations in vital signs, or updates to medical records, the system should be able to quickly correct the assessment conclusions and provide a traceable chain of evidence. This avoids the high computational cost and update lag caused by repeated full recalculation when evidence changes frequently in traditional methods. In this scenario, it is common for the same patient's laboratory results to be revised after review, image reports to be corrected, and medical orders to be adjusted, leading to the invalidation or replacement of previous evidence. Existing systems often update conclusions by re-extracting and fully reconstructing the assessment graph and recalculating the risk score, resulting in increased update delays and an unstable interpretation chain. Clinical users find it difficult to understand the reasons for the changes in the assessment conclusions, and auditors also find it difficult to trace when key evidence became invalid and how the conclusions were corrected after invalidation.

[0029] In this scenario, human clinical information comes from electronic medical records, medical records, structured laboratory test records, imaging reports, vital sign data collection, and medication orders and treatment records. The system performs standardized preprocessing on information from each source to generate a standardized clinical dataset. It then performs medical semantic extraction on unstructured text to generate a medical semantic element set. Subsequently, based on the standardized clinical dataset and the medical semantic element set, a patient-level assessment knowledge graph is constructed. Patient nodes, event nodes, and semantic nodes are written into a node set and an edge set is generated. A clinical validity structure is generated on the edge set and bound to each edge. The clinical validity structure carries elements used for invalidation determination, such as the effective time window interval, conflict anchor point identifier, revision token value, and validity status value. This ensures that evidence invalidation can be consistently expressed as a set of invalid edges, and further edge deletion is performed. At the risk assessment level, the system reads a preset risk label set and retrieves the corresponding risk label nodes from the node set. The risk label nodes are aggregated to form a risk endpoint node set, and an endpoint index is established. An Ev is constructed for each risk endpoint. The en-Shiloach forest maintains hierarchical distances and parent pointer records, forming a unified evidence propagation structure for multiple risk endpoints. When an evidence edge expires within a time window, the revision token is inconsistent, or a conflict substitution is established, it is determined to be an invalid edge and deleted. The system triggers a decremental update of the Even-Shiloach forest, locates the set of affected nodes through the parent edge dependency index, and performs queue-style local maintenance only on the hit set, avoiding repeated traversal of the entire graph from risk endpoints, thus achieving rapid updates in edge deletion scenarios. Subsequently, the system generates a hierarchical difference set based on the updated Even-Shiloach forest and determines the affected subgraphs. It performs local evaluation updates only within the affected subgraphs and generates an evidence chain. An evidence certificate structure is generated based on the evidence chain, and an initial evaluation result is generated based on the evidence certificate. When the core evidence is invalid and there is no valid substitute edge, a rebuttal event is triggered, and the initial evaluation result is corrected to obtain a corrected evaluation result. Finally, the corrected evaluation result is output to clinical users, while the corresponding evidence chain and rebuttal event are retained for review and auditing.

[0030] To demonstrate the problems solved and beneficial effects of this invention, a representative sample of hospitalized patients was selected to form an evaluation set. During the evaluation, the traditional full-scale update method was used as the control group. The control group's process involved re-extracting semantic elements, reconstructing the evaluation graph, and calculating risk scores and interpretation paths from each risk endpoint every time evidence changed. The method of this invention served as the experimental group. It used a clinical efficacy structure to determine the set of invalid edges and delete edges, triggered an Even-Shiloach forest decreasing update, and limited the affected subgraph with hierarchical difference, performing a local update and generating evidence chains and evidence certificates. Specific comparative data are shown in Table 1. Table 1 Comparison of incremental update performance and interpretability of patient-level risk assessment systems

[0031] As shown in Table 1, compared with the control group's full recalculation update, the strategy of the present invention based on Even-Shiloach forest decreasing update and local evaluation of affected subgraphs shows significant advantages in update efficiency and interpretability. The median single update latency decreased from 4.80 seconds to 0.72 seconds and P95 decreased from 12.60 seconds to 1.90 seconds. At the same time, the recalculation range of nodes and edges decreased from 100% to 6.5% and 8.2%, respectively. The average CPU usage decreased from 186% to 74% and the peak memory usage decreased from 6.3GB to 3.1GB. The interpretability and traceability rate increased from 61% to 93%, and the audit review pass rate increased from 68% to 95%. This shows that the present invention can achieve more stable incremental updates and more reliable result traceability with lower computational overhead in scenarios of dynamic evidence invalidation and edge deletion.

[0032] From a business closed-loop perspective, this invention can reliably map evidence failure to edge deletion operations in clinical environments where evidence changes frequently. It achieves local maintenance only for affected nodes through Even-Shiloach forest decremental updates, and then determines the affected subgraph and performs local evaluation and updates through hierarchical difference, thereby reducing redundant calculations and latency. At the same time, the evidence chain and evidence certificate structure provide clear evidence for changes in evaluation results. The disproving event strictly binds the failure of key evidence with the revision of conclusions, enabling clinical users to quickly understand the reasons for changes in conclusions, and auditors to review the source of evidence failure and confirm the rationality of the revision process. Ultimately, this results in improved update efficiency, reduced resource consumption, and improved evaluation quality and interpretation credibility.

[0033] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for evaluating human clinical information based on knowledge graphs, characterized in that, Includes the following steps: Collect human clinical information of the target individual, perform standardized preprocessing, and generate standardized clinical datasets; Perform medical semantic extraction on human clinical information to generate a set of medical semantic elements; Based on standardized clinical datasets and medical semantic element sets, a patient-level assessment knowledge graph for the target individual is constructed, and a clinical efficacy structure is generated for each edge. In the patient-level assessment knowledge graph, risk endpoint nodes are identified, and an Even-Shiloach tree is constructed using each risk endpoint node as the source point to form an Even-Shiloach forest. The set of failed edges is determined according to the clinical efficacy failure judgment rules. The edges corresponding to the set of failed edges are deleted from the patient-level assessment knowledge graph, and the Even-Shiloach forest is triggered to reduce the time, resulting in the updated Even-Shiloach forest. Based on the updated Even-Shiloach forest, a hierarchical difference set is generated and the affected subgraphs are identified. Local evaluation updates are performed on the affected subgraphs to generate the corresponding evidence chains. The evidence certificate structure is generated based on the evidence chain. The initial evaluation result of human clinical information is determined based on the evidence certificate structure. When the rebuttal trigger condition is met, a rebuttal event is generated. The initial evaluation result is corrected based on the rebuttal event to obtain the corrected evaluation result.

2. The method for evaluating human clinical information based on knowledge graphs according to claim 1, characterized in that, The generation of the standardized clinical dataset specifically includes: Collect human clinical information of target individuals within a preset time range, perform record-level normalization processing, and generate a record-level normalized clinical information set and a data quality record set. Perform medical terminology standardization processing on the record-level normalized clinical information set to generate a terminology mapping record table, and at the same time generate the terminology-normalized clinical information set, and incorporate the terminology mapping record table into the data quality record set; The terminology-standardized clinical information set is subjected to metrological and temporal standardization. After the metrological and temporal standardization is completed, a standardized clinical dataset is generated, and the data quality record set is stored together as the quality control record of the standardized clinical dataset.

3. The method for evaluating human clinical information based on knowledge graphs according to claim 1, characterized in that, The generation of the medical semantic element set specifically includes: Semantic extraction preprocessing is performed on unstructured clinical text in human clinical information. The text is normalized, and it is segmented into sentence sequences according to preset sentence segmentation rules. Each sentence is segmented to generate a word sequence. Each word is normalized and mapped to a standard word identifier. The start and end position indices of each word are recorded to generate the text preprocessing result. Perform medical entity extraction based on the text preprocessing results to generate a set of medical entity records; Entity relation extraction is performed based on the lexical sequence and the medical entity record set. Entity pairs in the medical entity record set are candidate-paired to generate an entity pair set. For each entity pair, the corresponding context fragment is located in the lexical sequence to generate a relation determination context. For each entity pair, relation trigger words are identified and relation direction is determined based on the relation determination context. For each entity pair, relation type is determined and relation type identifier is generated. For each entity pair, relation confidence is calculated and relation confidence value is generated. For each entity pair, time information is associated and relation time identifier is generated. The entity pair identifier, relation type identifier, relation direction identifier, relation confidence value, and relation time identifier are combined to generate an entity relation record set. The medical entity record set and the entity relation record set are combined to generate a medical semantic element set.

4. The method for evaluating human clinical information based on knowledge graphs according to claim 1, characterized in that, The acquisition of the patient-level assessment knowledge graph and clinical efficacy structure specifically includes: Perform clinical event processing on standardized clinical datasets to generate clinical event record sets; Perform record processing on the medical semantic element set, assign a unique identifier to each semantic element and create an index, record the type information of the semantic elements, and generate a semantic element record set; Based on the clinical event record set and semantic element record set, a unique node identifier is assigned and target individual nodes, event nodes and semantic nodes are generated according to the preset node generation rules. Individual event association edges are generated between target individual nodes and event nodes according to the preset edge connection rules. Event evidence association edges are generated between event nodes and semantic nodes according to the preset edge connection rules. Semantic relationship edges are generated between semantic nodes according to the relationship indication information carried in the semantic element record set, thus obtaining a patient-level assessment knowledge graph. For each edge in the edge set of the patient-level assessment knowledge graph, the effective time window interval is calculated based on the relation time identifier according to the preset time window rules, and the conflict anchor point identifier is determined according to the relation type identifier of the edge. The revision record in the standardized clinical dataset is used to match the revision record for the edge to generate a revision token value. The preset validity state is set as the initial validity state. The effective time window interval, conflict anchor point identifier, revision token value and initial validity state are written and generated into a one-to-one bound clinical validity structure.

5. The method for evaluating human clinical information based on knowledge graphs according to claim 1, characterized in that, The formation of the Even-Shiloach forest specifically includes: Read the preset risk label set, retrieve the corresponding risk label node in the node set, and create and write the node if it is not found. Gather the corresponding risk label nodes to form a risk endpoint node set, and assign a unique endpoint identifier to each node in the risk endpoint node set and establish an endpoint index. For each risk endpoint node in the risk endpoint node set, edges that are directly or indirectly connected to the current risk endpoint node are selected from the edge set of the patient-level assessment knowledge graph as candidate edge sets. The candidate edge sets are selected according to the edge type selection rules, and the edges used for evidence propagation are retained and the edges that do not participate in the assessment are removed. The retained edges are selected according to the time validity selection rules, and the edges that exceed the valid time range are removed, thus forming a valid edge set. The nodes associated with the valid edge set are written into the traversable node set, and a traversable subgraph is formed based on the traversable node set and the valid edge set. For each node in the traversable subgraph, mark the level distance as unreachable, set the parent pointer to null, set the level distance of the current risk endpoint node to zero and write it to the level distance record, write the current risk endpoint node to the queue to be expanded, retrieve the current node from the queue to be expanded and sequentially traverse the adjacent edges to locate the neighboring nodes, set the level distance of the neighboring node in the unreachable state to the current node's level distance plus one and write it to the level distance record, set the parent pointer of the neighboring node to the current node and record the parent edge, write the neighboring node to the queue to be expanded, and repeat the operations of retrieving the queue node, traversing the adjacent edges, writing the level distance, writing the parent pointer and recording the parent edge until the queue to be expanded is empty, and form an Even-Shiloach tree with the risk endpoint node as the root based on the level distance record and the parent pointer record; A forest index is built for the Even-Shiloach tree corresponding to each risk endpoint node in the risk endpoint node set. The endpoint identifier of each risk endpoint node is associated with the corresponding root node identifier and the hierarchical distance record is associated with the parent pointer record. All registration results in the risk endpoint node set are aggregated to form the Even-Shiloach forest.

6. The method for evaluating human clinical information based on knowledge graphs according to claim 1, characterized in that, The updated Even-Shiloach forest was obtained specifically including: Read the edge set of the patient-level assessment knowledge graph and read the clinical efficacy structure bound to each edge, obtain the current assessment time, and initialize the set of failed edges to an empty set; Failure determination is performed edge by edge in the edge set. The effective time window interval bound to the edge is read and compared with whether the current evaluation time falls within the effective time window interval. If the current evaluation time does not fall within the effective time window interval, the edge is written into the failed edge set. The revision record of the standardized clinical dataset is retrieved and the revision token value is matched for the edge. At the same time, a consistency comparison is performed with the revision token value bound to the edge. If the consistency comparison result is inconsistent, the edge is written into the failed edge set. The conflict anchor point identifier bound to the edge is read and the candidate alternative edge with the same conflict anchor point identifier is retrieved in the edge set. The effective time window interval of the candidate alternative edge is read and the candidate alternative edge that does not fall within the effective time window interval at the current evaluation time is filtered out. After filtering, if the candidate alternative edge exists, the edge is written into the failed edge set. Remove the edges corresponding to the invalid edge set from the edge set of the patient-level assessment knowledge graph and generate a deletion list. Write the deletion list into the deletion record set and establish an association with the invalid edge set. Trigger the decreasing update of the Even-Shiloach forest anchored to the risk endpoint and initialize the update support structure. Read the level distance record and parent pointer record of the Even-Shiloach tree corresponding to each risk endpoint root. Read the current parent edge identifier of each node and write it into the parent edge dependency index. Map the parent edge identifier to the set of nodes that depend on the parent edge and register them in groups according to the risk endpoint root. Complete the initialization of the parent edge dependency index. The parent edge dependency index is used to perform impact localization on the edge deletion list. The deleted edge identifier is read one by one in the edge deletion list. The set of nodes that depend on the deleted edge identifier is retrieved in the parent edge dependency index. If the node set is not empty, the node is written to the corresponding queue to be updated according to the risk endpoint node identifier and the node level distance is marked as to be recalculated. If the node set is empty, it is not written to the queue to be updated. The set of nodes to be updated from the queue is limited to the set of nodes whose parent edge dependency index is hit. A decremental update is performed on the queue. The process is repeated to retrieve nodes to be updated, traverse the adjacent edges of the nodes to be updated, filter the edges in the deletion record set, search for candidate parent edges in the remaining adjacent edges, when a candidate parent edge exists, read the level distance of the starting node of the candidate parent edge, increment the level distance of the candidate parent node by one and write it into the level distance record of the node to be updated, write the candidate parent edge into the parent pointer record of the node to be updated, and update the parent edge dependency index. When the candidate parent edge is empty, the level distance of the node to be updated is marked as unreached, the parent pointer record of the node to be updated is set to null, and the parent edge dependency index is updated. The child nodes of the node to be updated are written into the queue, and the level distances written to the child nodes are marked as to be recalculated. When the queue to be updated is empty, the level distance records and the parent pointer records are combined to form the updated Even-Shiloach forest.

7. The method for evaluating human clinical information based on knowledge graphs according to claim 1, characterized in that, The generation of the chain of evidence specifically includes: Read the updated Even-Shiloach forest, read the level distance value before and after the update for each node in the node set, and perform a level difference judgment operation. Mark the nodes whose level distance values ​​before and after the update are inconsistent as level change nodes and write them into the level difference set. The affected node set and the affected subgraph are determined based on the hierarchical difference set. The node corresponding to the hierarchical difference set is located in the node set of the patient-level assessment knowledge graph and written into the affected node set. The edge adjacent to the node in the affected node set is retrieved in the edge set of the patient-level assessment knowledge graph and written into the affected edge set. The affected subgraph is formed based on the affected node set and the affected edge set. Perform a local evaluation update on the affected subgraph, read the set of risk endpoint nodes in the updated Even-Shiloach forest, filter the affected nodes connected to the risk endpoint nodes in the affected subgraph and generate a set of evidence nodes, record the backtracking of each evidence node in the evidence node set along the parent pointer to the risk endpoint node, and record the node identifier sequence and edge identifier sequence traversed during the backtracking, and collect the node identifier sequence and edge identifier sequence corresponding to the risk endpoint node to generate an evidence chain.

8. The method for evaluating human clinical information based on knowledge graphs according to claim 1, characterized in that, The specific steps to obtain the revised evaluation results include: Read the evidence chain and group it according to the risk endpoint node identifier. Write the grouping results into the evidence chain record table and assign evidence chain identifiers to form an evidence chain set. Based on the evidence chain set, edge identifiers are extracted to form a core evidence edge set. Candidate alternative edges that share the start node identifier and end node identifier with the core evidence edges are retrieved from the edge set of the patient-level assessment knowledge graph. Invalid candidate alternative edges at the current assessment time are screened out. An alternative mapping relationship is established based on the core evidence edge set and the retained candidate alternative edges to form an evidence certificate structure. Based on the evidence certificate structure and the deletion list, the existence of core evidence is determined. The core evidence in the existing state is weighted and accumulated according to the preset weight table to obtain the risk score value. The risk level identifier is then mapped according to the preset score threshold table to generate the initial assessment result. Based on the evidence certificate structure and the deletion list, a disproving evidence trigger determination is performed. The core evidence edge set is read edge by edge, and it is checked whether the core evidence edge appears in the deletion list. For the core evidence edge that appears in the deletion list, it is checked whether the alternative mapping relationship provides a valid alternative edge. When the core evidence edge appears in the deletion list and the alternative mapping relationship does not provide a valid alternative edge, a disproving evidence event is generated and written to the disproving evidence event record table. Based on the disproving evidence event, the initial evaluation result is corrected, and a corrected evaluation result is generated.

Images