Multi-guideline knowledge alignment and difference analysis method based on causal evidence graph

Abstract

This invention discloses a method for multi-guideline knowledge alignment and difference analysis based on causal evidence graphs, belonging to the field of artificial intelligence technology. It includes: generating structured query objects corresponding to unstructured clinical questions; filtering evidence-based supporting data fragments and fact-based suggestion data fragments from an instantiated three-layer library graph; constructing standardized recommendation semantic boxes based on the fact-based suggestion data fragments and transforming their qualitative descriptions into a unified numerical space; determining the logical relationships between standardized recommendation semantic boxes within the same category; constructing a causal evidence graph based on intervention elements, mediation mechanism elements, and clinical outcome elements extracted from the evidence-based supporting data fragments; and determining the differences between different guideline recommendations corresponding to the structured query objects based on the transformed standardized recommendation semantic boxes, the logical relationships between standardized recommendation semantic boxes within the same category, and the causal evidence graph. This invention can effectively eliminate model illusions and ensure that the output results are supported by clear evidence.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence, and specifically relates to a method for multi-guideline knowledge alignment and difference analysis based on causal evidence graphs. Background Technology

[0002] The exponential growth of medical knowledge has led to information overload. Guidelines issued by different countries and institutions often differ in their recommendations and the strength of evidence for the same clinical problem. Most current clinical decision support systems are answer-oriented, outputting only a single conclusion and failing to systematically present the logical differences between multiple guidelines and the supporting evidence.

[0003] Furthermore, existing technologies suffer from fundamental flaws when dealing with unstructured guidelines. Clinical guidelines are mostly free texts, and different languages ​​and evaluation systems make it difficult to automatically align and quantitatively compare recommendations. While introducing large language models improves interactivity, it faces a serious illusion challenge: the generated clinical support recommendations may not correspond to real evidence sources and lack causal explanation capabilities, making it difficult to elucidate the causal links behind the recommended solutions. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a multi-guide knowledge alignment and difference analysis method based on causal evidence graphs.

[0005] Firstly, this invention proposes a multi-guideline knowledge alignment and difference analysis method based on causal evidence graphs, comprising the following steps:

[0006] Step S1: Utilize the collaborative action of intelligent agents to perform semantic extraction of unstructured clinical questions and generate structured query objects corresponding to the unstructured clinical questions;

[0007] Step S2: Define a three-layer library map as an external reference source. The three-layer library map includes, from bottom to top, a terminology definition layer, an evidence support layer, and a fact recommendation layer. The fact recommendation layer stores atomic recommendation items in the guidelines, which are clinical guidelines.

[0008] The data units to be retrieved from the guideline database and the medical research literature database are linked to the corresponding levels of the three-layer library map to form an instantiated three-layer library map; based on the structured query object, evidence-based supporting data fragments and factual suggestion data fragments are selected from the instantiated three-layer library map;

[0009] Step S3: Construct standardized recommendation semantic boxes based on factual suggestion data fragments; transform the qualitative descriptions corresponding to the standardized recommendation semantic boxes into a unified numerical space;

[0010] Step S4: Calculate the similarity between different standardized recommendation semantic boxes and perform clustering based on the similarity; determine the logical relationships between standardized recommendation semantic boxes within the same class based on the logical condition set and recommendation polarity corresponding to each standardized recommendation semantic box in the same class; logical relationships include equivalence relations, expansion relations, contraction relations and conflict relations;

[0011] Step S5: Construct a causal evidence map based on the intervention elements, mediation mechanism elements, and clinical outcome elements extracted from the evidence-supporting data fragments;

[0012] Step S6: Based on the transformed standardized recommendation semantic boxes, the logical relationships between standardized recommendation semantic boxes within the same category, and the causal evidence graph, determine the differences between different guideline recommendation schemes corresponding to the structured query object.

[0013] Preferably, step S1 includes:

[0014] Step S11: Utilize agent cluster collaboration to perform semantic extraction of unstructured clinical questions;

[0015] Step S12: For each extracted semantic unit including fuzzy conditions, extract the clinical field, judgment operator, and threshold of the semantic unit including fuzzy conditions; construct a disjunction paradigm including clinical field, judgment operator, and threshold using logical operators; encapsulate the disjunction paradigm as the reconstruction result of the semantic unit including fuzzy conditions; all reconstruction results form a logical condition set.

[0016] Step S13: For the extracted semantic units including clinical scoring tools, determine the scoring calculation model corresponding to the clinical scoring tools, and the clinical scoring tool context is composed of the clinical scoring tools and the scoring calculation model corresponding to the clinical scoring tools.

[0017] Step S14: Based on the extracted other semantic units, logical condition set, and clinical scoring tool context, construct a structured query object corresponding to the unstructured clinical question; wherein, other semantic units are semantic units other than those including fuzzy conditions and those including clinical scoring tools.

[0018] Structured Query Object The definition is as follows:

[0019]

[0020] in, Indicates the target audience. Indicates intervention measures, Represents a set of logical conditions. Indicates the result event, Indicates the context of the recommendation. Indicates the context of clinical scoring tools. This indicates a time window constraint.

[0021] Preferably, the terminology definition layer provides conceptual explanations based on a medical dictionary, and the evidence-based support layer stores mechanistic information extracted from medical research literature.

[0022] Preferably, in step S2, the data units to be retrieved from the guideline database and the medical research literature database are linked to the corresponding levels of the three-layer library map to form an instantiated three-layer library map, including:

[0023] Step S21: Construct [fact, evidence-based, terminology] triples based on medical dictionaries, guideline databases, and medical research literature databases;

[0024] Step S22: Determine the relationship between factual entities and empirical entities, and the relationship between empirical entities and terminological entities in the triple;

[0025] The relationship between factual entities and fact-based entities is established based on the following calculation formula:

[0026]

[0027] in, This is a set of directed links established between factual entities and evidence-based entities, used to construct the traceability link between the factual suggestion layer and the evidence-based support layer in the three-layer library graph. As a factual entity, For the sake of evidence-based practice, For comprehensive semantic information of factual entities, Embed vectors for the content of fact entities. For the comprehensive semantic information of the evidence-based entity, Embed the content vector of the evidence-based entity. It is a second normal form. The first threshold for similarity between targeted links is preset. To connect factual entities and evidence-based entities Semantic relation annotation; in and The cosine similarity calculation result is greater than or equal to the first directional link similarity threshold. In this case, Relation1 is marked as "referenced from";

[0028] The relationship between empirical entities and terminological entities is established based on the following calculation formula:

[0029]

[0030] in, This is a set of directed links established between evidence-based entities and terminological entities, used to construct a tracing link from the evidence-based support layer to the terminology definition layer in a three-layer library graph. For term entities, For the comprehensive semantic information of term entities, Embed vectors for the content of term entities. The second targeted link similarity threshold is preset; To connect with empirical evidence With term entities semantic relation annotation, in and The cosine similarity calculation result is greater than or equal to the second directional link similarity threshold. In this case, Marked as "defined as";

[0031]

[0032] Where num = {1, 2, 3}, when num takes the values ​​1, 2, and 3 respectively. These represent the comprehensive semantic information of factual entities, the comprehensive semantic information of fact-based entities, and the comprehensive semantic information of terminological entities, respectively; when num takes values ​​of 1, 2, and 3, These are respectively factual entities, evidence-based entities, and terminological entities; , They are respectively The corresponding entity names and term names, , They are respectively The corresponding entity type and semantic type, , They are respectively The corresponding context content and context description text.

[0033] Preferably, in step S2, based on the structured query object, evidence-based supporting data fragments and factual suggestion data fragments are selected from the instantiated three-layer library graph, including:

[0034] Step S23: Obtain the structured query object and search the instantiated three-layer library graph using a U-shaped search strategy;

[0035] The U-shaped retrieval strategy involves first performing a top-down retrieval on the instantiated three-layer library map, and then performing a bottom-up retrieval.

[0036] The top-down search is as follows:

[0037] Generate a first set of query tags corresponding to the structured query objects using a large model. The first set of tags includes the structured query objects. middle The corresponding medical category tags; the tag summary corresponding to the first tag set is denoted as ;

[0038] Based on the first query tag set, the fact suggestion layer, evidence-based support layer, and terminology definition layer are retrieved sequentially. The first result set obtained is { When num takes the values ​​1, 2, and 3 respectively, These are the search results corresponding to the fact suggestion layer, the evidence support layer, and the terminology definition layer when searching from top to bottom.

[0039] A first evidence subgraph is constructed based on the first result set, and the nodes in the first evidence subgraph are entities in the first result set;

[0040] The edges in the first evidence subgraph include: edges established between entities with semantic relationships in the first result set; edges established between fact entities marked "referenced to" by Relation1 and evidence-based entities in the first result set; edges established between entities in the first result set, An edge is established between a empirical entity marked "defined as" and a defined entity;

[0041] in, , ;exist When the values ​​are 1 and 2 respectively, These represent the search results for the factual suggestion layer and the evidence-based support layer, respectively, when searching from top to bottom; with i taking values ​​of 1 and 2, respectively. These are the search results corresponding to the evidence-based support layer and the terminology definition layer when searching from top to bottom. for The first in A tag summary object, For calculation and The similarity between them; the sim function is a similarity function. This indicates the search for the tag summary object corresponding to the highest similarity;

[0042] Bottom-up retrieval is:

[0043] Generate a second set of query tags corresponding to the structured query objects using a large model. The second set of query tags includes the structured query objects. middle The corresponding medical category tags; the tag summary corresponding to the second tag set is denoted as ;

[0044] Tag summary based on the second tag set Retrieve the terminology definition layer of the three-layer library map to obtain the preceding... Individual and The semantically most relevant term entity is denoted as ; It is a positive integer;

[0045] Tag summary based on the second tag set Retrieve the evidence-based support layer of the three-layer library map to obtain the previous Individual and The most semantically relevant empirical evidence is denoted as ;based on Correction Semantic representation;

[0046] Tag summary based on the second tag set Retrieve the fact suggestion layer of the three-layer library graph to obtain the previous Individual and The most semantically relevant fact entity is denoted as ;based on Correction Preliminary conclusions were drawn;

[0047] The revised Corrected as well as Encapsulate as a second result set;

[0048] The first evidence subgraph and the second result set are used together as the search results;

[0049] Step S24: Filter out evidence-based supporting data fragments and factual suggestion data fragments from the search results, including:

[0050] From the search results, select evidence-based supporting data fragments and factual suggestion data fragments with a comprehensive relevance score not lower than the preset relevance threshold.

[0051] Preferably, step S3 includes:

[0052] Step S31: Construct the octet corresponding to the fact suggestion data fragment based on the fact suggestion data fragment. The octet is used as a standardized recommendation semantic box; where, Indicates the target audience. Indicates intervention measures, Represents a set of logical conditions. Indicates the original recommendation strength. Indicates the level of original evidence; Indicates the recommended polarity. This indicates a recommendation to implement. This indicates that the suggestion should not be implemented; Indicate exceptions and restrictions, and record inapplicable scenarios or contraindications mentioned in clinical guidelines; This indicates the source identification information, including the clinical guideline name, issuing organization, version, and original paragraph number;

[0053] Step S32: Using the original evidence level E as a qualitative description of the standardized recommendation semantic box, the qualitative description is converted into... The values ​​are between; where a predefined mapping standard reference table defines the mapping relationship between the original evidence level E and the numerical value.

[0054] Preferably, in step S4, calculating the similarity between different standardized recommendation semantic boxes includes:

[0055] Step S41: Two standardized recommendation semantic boxes to be used for similarity calculation and Standardize recommendation semantic boxes and In and All were normalized using a medical terminology normalization mapping function;

[0056] Step S42: Determine the normalized result and The corresponding set of logical conditions and Intersection, Recommendation Polarity and Level of original evidence and Semantic similarity between them; determine the number of logical conditions num1 in the intersection, and determine... and The corresponding set of logical conditions and Given the number of logical conditions in the union set num2, calculate num1 / num2, and denote it as the first ratio; recommend polarity. and When consistent, polarity is recommended. and The consistency score between them is recorded as 1, and the recommendation polarity is... and When consistent, polarity is recommended. and The consistency score between them is recorded as 0; based on the first ratio and recommendation polarity and Consistency score and level of original evidence and Determining the semantic similarity between and The similarity between them.

[0057] Preferably, in step S4, determining the logical relationship between standardized recommendation semantic boxes within the same class based on the logical condition set and recommendation polarity corresponding to each standardized recommendation semantic box within the same class includes:

[0058] When the logical condition set of two standardized recommendation semantic boxes within the same category When the two standardized recommendation semantic boxes are equivalent and have the same recommendation polarity, they are equivalent.

[0059] When two standardized recommendation semantic boxes within the same category have the same recommendation polarity and the set of logical conditions If it is an inclusion relationship, then the set of logical conditions A broader set of standardized recommendation semantic boxes and logical conditions The relationship between standardized recommendation semantic boxes with a narrower scope is a contraction relationship; logical condition set Narrower-scope standardized recommendation semantic boxes and logical condition sets The relationships between standardized recommendation semantic boxes with a broader scope are extended relationships;

[0060] When two standardized recommendation semantic boxes within the same category have inconsistent recommendation polarities and the logical condition sets of the two standardized recommendation semantic boxes are... When there is an overlap, the two standardized recommendation semantic boxes are in conflict.

[0061] Preferably, in step S5, a causal evidence map is constructed based on intervention elements, mediation mechanism elements, and clinical outcome elements extracted from evidence-based support data fragments, including:

[0062] The extracted intervention elements, mediation mechanism elements, and clinical outcome elements are all used as nodes in the causal evidence map;

[0063] The causal relationship between intervention elements or mediating mechanisms explicitly described in evidence-based data fragments and clinical outcome elements is used as the connecting edge between nodes. The direction of the connecting edge is determined according to the causal relationship, and the weight of the connecting edge is determined based on the literature research type, evidence quality grade, sample size, and effect size between the two nodes connected by the connecting edge. Edges with connecting edge weights lower than a preset weight threshold are deleted to obtain the causal evidence map.

[0064] Preferably, step S6 includes:

[0065] Step S61: Based on the transformed standardized recommendation semantic boxes, the logical relationships between standardized recommendation semantic boxes within the same category, and the causal evidence graph, determine the conclusion corresponding to the structured query object and the causal path supporting the conclusion; for each conclusion, based on the source identifier information of the conclusion... Obtain the clinical guideline name, issuing organization, version, and original paragraph number corresponding to the conclusion;

[0066] Step S62: Obtain all complete causal paths from all causal paths; where a complete causal path is a path with a complete chain of evidence, and determine the differences between different guideline-recommended solutions based on the set of logical conditions corresponding to the complete causal path.

[0067] Step S63: Store the conclusions corresponding to the complete causal path in the output file, and use the contents of the output file as the output result.

[0068] Secondly, this invention proposes a multi-guide knowledge alignment and difference analysis device based on causal evidence graphs, comprising:

[0069] Query object generation module: Configured to use intelligent agent cluster collaboration to perform semantic extraction of unstructured clinical questions and generate structured query objects corresponding to unstructured clinical questions;

[0070] Three-layer library map generation module: configured to define a three-layer library map as an external reference source. The hierarchy of the three-layer library map from bottom to top includes a terminology definition layer, an evidence support layer, and a fact recommendation layer. The fact recommendation layer stores atomic recommendation items in the guidelines, which are clinical guidelines.

[0071] The data units to be retrieved from the guideline database and the medical research literature database are linked to the corresponding levels of the three-layer library map to form an instantiated three-layer library map; based on the structured query object, evidence-based supporting data fragments and factual suggestion data fragments are selected from the instantiated three-layer library map;

[0072] The recommendation semantic box generation module is configured to construct standardized recommendation semantic boxes based on factual suggestion data fragments; and to transform the qualitative descriptions corresponding to the standardized recommendation semantic boxes into a unified numerical space.

[0073] The logical relationship determination module is configured to calculate the similarity between different standardized recommendation semantic boxes and perform clustering based on the similarity; based on the set of logical conditions and recommendation polarity corresponding to each standardized recommendation semantic box in the same class, it determines the logical relationship between standardized recommendation semantic boxes in the same class; the logical relationship includes equivalence relationship, expansion relationship, contraction relationship and conflict relationship;

[0074] Causal evidence map generation module: configured to construct a causal evidence map based on intervention elements, mediation mechanism elements, and clinical outcome elements extracted from evidence-supporting data fragments;

[0075] The difference determination module is configured to determine the differences between different guideline recommendation schemes corresponding to structured query objects based on the transformed standardized recommendation semantic boxes, the logical relationships between standardized recommendation semantic boxes within the same category, and the causal evidence graph.

[0076] The method of the present invention can achieve the following beneficial effects:

[0077] (1) The technical solution of the present invention can effectively eliminate model illusion. The present invention forcibly blocks the generation of content that cannot be traced back to the structured knowledge base or causal path, ensuring that the output results have clear evidence support.

[0078] (2) The technical solution of the present invention can achieve accurate alignment and difference visualization of multi-source heterogeneous guidelines. Clustering and aligning guide entries from different institutions, versions and languages, and explicitly determining whether they are equivalent, extended, contracted or conflicting, helps to quickly identify consensus areas.

[0079] (3) The technical solution of the present invention can provide causal explanations and enhance the credibility of decision-making.

[0080] (4) The technical solution of the present invention can break the semantic silos across languages ​​and evaluation systems. Attached Figure Description

[0081] Figure 1 This is a flowchart of the multi-guide knowledge alignment and difference analysis method based on causal evidence graph proposed in this invention. Detailed Implementation

[0082] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0083] This invention proposes a multi-guideline knowledge alignment and difference analysis method based on causal evidence graphs, such as... Figure 1 As shown, it includes the following steps:

[0084] Step S1: Utilize the collaborative action of an agent cluster to perform semantic extraction of unstructured clinical questions and generate structured query objects corresponding to the unstructured clinical questions.

[0085] Step S2: Define a three-layer library map as an external reference source. The three-layer library map includes, from bottom to top, a terminology definition layer, an evidence support layer, and a fact recommendation layer. The fact recommendation layer stores atomic recommendation items in the guidelines, which are clinical guidelines.

[0086] The data units to be retrieved from the guideline database and the medical research literature database are linked to the corresponding levels of the three-layer library map to form an instantiated three-layer library map; based on the structured query object, evidence-based supporting data fragments and factual suggestion data fragments are selected from the instantiated three-layer library map.

[0087] Step S3: Construct standardized recommendation semantic boxes based on factual suggestion data fragments; transform the qualitative descriptions corresponding to the standardized recommendation semantic boxes into a unified numerical space.

[0088] Step S4: Calculate the similarity between different standardized recommendation semantic boxes and perform clustering based on the similarity; determine the logical relationships between standardized recommendation semantic boxes within the same class based on the logical condition set and recommendation polarity corresponding to each standardized recommendation semantic box in the same class; logical relationships include equivalence relations, expansion relations, contraction relations and conflict relations.

[0089] Step S5: Construct a causal evidence map based on the intervention elements, mediation mechanism elements, and clinical outcome elements extracted from the evidence-supporting data fragments.

[0090] Step S6: Based on the transformed standardized recommendation semantic boxes, the logical relationships between standardized recommendation semantic boxes within the same category, and the causal evidence graph, determine the differences between different guideline recommendation schemes corresponding to the structured query object.

[0091] In a preferred embodiment, the specific implementation process of step S1 includes the following sub-steps:

[0092] Step S11: Utilize agent cluster collaboration to perform semantic extraction of unstructured clinical questions.

[0093] Step S12: For each extracted semantic unit including fuzzy conditions, extract the clinical field, judgment operator and threshold of the semantic unit including fuzzy conditions; construct a disjunction paradigm including clinical field, judgment operator and threshold using logical operators; encapsulate the disjunction paradigm as the reconstruction result of the semantic unit including fuzzy conditions; all reconstruction results form a logical condition set.

[0094] Step S13: For the extracted semantic units including clinical scoring tools, determine the scoring calculation model corresponding to the clinical scoring tools, and the clinical scoring tools and the scoring calculation models corresponding to the clinical scoring tools constitute the clinical scoring tool context.

[0095] Step S14: Based on the extracted other semantic units, logical condition set and clinical scoring tool context, construct the structured query object corresponding to the unstructured clinical question; wherein, other semantic units are semantic units other than semantic units including fuzzy conditions and semantic units including clinical scoring tools.

[0096] Structured Query Object The definition is as follows:

[0097]

[0098] in, Indicates the target audience. Indicates intervention measures, Represents a set of logical conditions. Indicates the result event, Indicates the context of the recommendation. Indicates the context of clinical scoring tools. This indicates a time window constraint.

[0099] This invention utilizes heterogeneous large language models to form an intelligent agent cluster. Through collaboration, these agents parse unstructured natural language problems into computable, standardized structured query objects. .

[0100] For example, in a community hospital without a neurosurgery department, when a CT scan reveals an acute subdural hematoma but the patient's vital signs are temporarily stable, the doctor might ask unstructured clinical questions when deciding to transfer the patient. These unstructured clinical questions might be: "What non-therapeutic steps need to be completed before the transfer, such as contacting the receiving facility and preparing transport monitoring records?"

[0101] In this invention, This indicates the target population, including disease characteristics, demographic information, and comorbid background. This refers to interventions, such as medications, surgery, or management strategies. Represents a set of logical conditions. Indicates the outcome event, such as thrombosis risk, mortality, or bleeding events; This indicates the context of the recommendation, used to distinguish the specific scenario of clinical support advice. It can identify the prevention level of the current plan, such as primary prevention or secondary prevention. Indicates the context of a clinical scoring tool, such as the CHA2DS2-VASc score; This indicates a time window constraint, used to constrain the duration of a protocol or the clinical observation period, such as a follow-up within 30 days.

[0102] The terminology definition layer provides conceptual explanations based on a medical dictionary, the evidence-based support layer stores mechanistic information extracted from medical research literature, and the factual advice layer stores atomized recommendation entries from guidelines, which are clinical guidelines.

[0103] In a preferred embodiment, step S2, "directly linking the data units to be retrieved from the guide database and the medical research literature database to the corresponding level of the three-layer library map to form an instantiated three-layer library map," specifically includes the following sub-steps:

[0104] Step S21: Construct a [fact, evidence-based, terminology] triple based on medical dictionaries, guideline databases, and medical research literature databases.

[0105] Step S22: Determine the relationship between factual entities and empirical entities, and the relationship between empirical entities and terminological entities in the triple.

[0106] The relationship between factual entities and fact-based entities is established based on the following calculation formula:

[0107]

[0108] in, This is a set of directed links established between factual entities and evidence-based entities, used to construct the traceability link between the factual suggestion layer and the evidence-based support layer in the three-layer library graph. As a factual entity, For the sake of evidence-based practice, For comprehensive semantic information of factual entities, Embedding vectors for factual entities involves transforming unstructured text into computationally computable high-dimensional vectors using a pre-trained medical language model. For the comprehensive semantic information of the evidence-based entity, for The corresponding content embedding vector, It is a second normal form. The first threshold for similarity between targeted links is preset. For connection and Semantic relation annotations are used to define the edge types in the three-layer library graph; in and The cosine similarity calculation result is greater than or equal to In this case, Relation1 is marked as "referenced from" and is used to establish a source relationship.

[0109] The relationship between empirical entities and terminological entities is established based on the following calculation formula:

[0110]

[0111] in, This is a set of directed links established between evidence-based entities and terminological entities, used to construct a tracing link from the evidence-based support layer to the terminology definition layer in a three-layer library graph. For term entities, For the comprehensive semantic information of term entities, The content embedding vector for terminology entities is generated by transforming the aforementioned comprehensive semantic information into a computable high-dimensional vector through a pre-trained medical language model. The second directional link similarity threshold is preset, and logical association is only allowed when the semantic association strength between the evidence-based entity and the term entity exceeds the preset second directional link similarity threshold; For connection and The semantic relationship annotation is used to clarify the physical meaning of edges in the three-layer library graph. and The cosine similarity calculation result is greater than or equal to In this case, Marked as "defined as".

[0112]

[0113] Where num = {1, 2, 3}, when num takes the values ​​1, 2, and 3 respectively. These represent the comprehensive semantic information of factual entities, the comprehensive semantic information of fact-based entities, and the comprehensive semantic information of terminological entities, respectively; when num takes values ​​of 1, 2, and 3, These are respectively factual entities, evidence-based entities, and terminological entities; , They are respectively The corresponding entity names and term names, , They are respectively The corresponding entity type and semantic type, , They are respectively The corresponding context content and context description text.

[0114] This invention uses cosine similarity to determine the relationship between factual entities and empirical entities, and the relationship between empirical entities and terminological entities in a triple.

[0115] In a preferred embodiment, step S2, "based on the structured query object, selecting evidence-based supporting data fragments and factual suggestion data fragments from the instantiated three-layer library graph," specifically includes the following sub-steps:

[0116] Step S23: Obtain the structured query object and perform a search on the instantiated three-layer library graph using a U-shaped search strategy.

[0117] The U-shaped retrieval strategy involves first performing a top-down retrieval on the instantiated three-layer library map, and then performing a bottom-up retrieval.

[0118] The top-down search is as follows:

[0119] Generate a first set of query tags corresponding to the structured query objects using a large model. The first set of tags includes the structured query objects. middle The corresponding medical category tags; the tag summary corresponding to the first tag set is denoted as ;

[0120] Based on the first query tag set, the fact suggestion layer, evidence-based support layer, and terminology definition layer are retrieved sequentially. The first result set obtained is { When num takes the values ​​1, 2, and 3 respectively, These are the search results corresponding to the fact suggestion layer, the evidence support layer, and the terminology definition layer when searching from top to bottom.

[0121] A first evidence subgraph is constructed based on the first result set, and the nodes in the first evidence subgraph are entities in the first result set;

[0122] The edges in the first evidence subgraph include: edges established between entities with semantic relationships in the first result set; edges established between fact entities marked "referenced to" by Relation1 and evidence-based entities in the first result set; edges established between entities in the first result set, An edge is established between a empirical entity marked "defined as" and a defined entity;

[0123] in, Let i = {1, 2}; when i takes the values ​​1 and 2 respectively, These represent the search results for the factual suggestion layer and the evidence-based support layer, respectively, when searching from top to bottom; with i taking values ​​of 1 and 2, respectively. These are the search results corresponding to the evidence-based support layer and the terminology definition layer when searching from top to bottom. for The first in A tag summary object, For calculation and The similarity between them; the sim function is a similarity function. This indicates the search for the tag summary object with the highest similarity.

[0124] Bottom-up retrieval is:

[0125] Generate a second set of query tags corresponding to the structured query objects using a large model. The second set of query tags includes the structured query objects. middle The corresponding medical category tags; the tag summary corresponding to the second tag set is denoted as ;

[0126] Tag summary based on the second tag set Retrieve the terminology definition layer of the three-layer library map to obtain the preceding... Individual and semantically most relevant term entities ;

[0127] Tag summary based on the second tag set Retrieve the evidence-based support layer of the three-layer library map to obtain the previous Individual and The most semantically relevant empirical evidence ;based on Correction Semantic representation;

[0128] Tag summary based on the second tag set Retrieve the fact suggestion layer of the three-layer library graph to obtain the previous Individual and The most semantically relevant fact entity ;based on Correction Preliminary conclusions were drawn;

[0129] The revised Corrected and factual entities Encapsulate as a second result set;

[0130] The first evidence subgraph and the second result set are used together as the search results.

[0131] Step S24: Filter out evidence-based supporting data fragments and factual suggestion data fragments from the search results, including:

[0132] From the search results, select evidence-based supporting data fragments and factual suggestion data fragments with a comprehensive relevance score not lower than the preset relevance threshold.

[0133] In a preferred embodiment, the specific implementation process of step S3 includes the following sub-steps:

[0134] Step S31: Construct the octet corresponding to the fact suggestion data fragment based on the fact suggestion data fragment. The octet is used as a standardized recommendation semantic box; where, Indicates the target audience. Indicates intervention measures, Represents a set of logical conditions. Indicates the original recommendation strength. Indicates the level of original evidence; Indicates the recommended polarity. This indicates a recommendation to implement. This indicates that the suggestion should not be implemented; Indicate exceptions and restrictions, and record inapplicable scenarios or contraindications mentioned in clinical guidelines; This indicates the source identification information, including the clinical guideline name, issuing organization, version, and original paragraph number;

[0135] Step S32: Using the original evidence level E as a qualitative description of the standardized recommendation semantic box, the qualitative description is converted into... The values ​​are between; where a predefined mapping standard reference table defines the mapping relationship between the original evidence level E and the numerical value.

[0136] Because clinical guidelines from different sources and versions exhibit significant heterogeneity in their expression, recommendation strength labeling, and evidence grading systems, this invention constructs a standardized recommendation semantic box, converts factual suggestion data fragments into a unified octet structure, and uses a mapping function to transform qualitative descriptions into quantitative representations within a unified numerical space.

[0137] In a preferred embodiment, the step S4, "calculating the similarity between different standardized recommendation semantic boxes," specifically includes the following sub-steps:

[0138] Step S41: Two standardized recommendation semantic boxes to be used for similarity calculation and Standardize recommendation semantic boxes and In and All were normalized using a medical terminology normalization mapping function.

[0139] Step S42: Determine the normalized result and The corresponding set of logical conditions and Intersection, Recommendation Polarity and Level of original evidence and Semantic similarity between them; determine the number of logical conditions num1 in the intersection, and determine... and The corresponding set of logical conditions and Given the number of logical conditions in the union set num2, calculate num1 / num2, and denote it as the first ratio; recommend polarity. and When consistent, polarity is recommended. and The consistency score between them is recorded as 1, and the recommendation polarity is... and When consistent, polarity is recommended. and The consistency score between them is recorded as 0; based on the first ratio and recommendation polarity and Consistency score and level of original evidence and Determining the semantic similarity between and The similarity between them.

[0140] In this invention, semantic alignment is performed before calculating the similarity between standardized recommendation semantic boxes, thereby ensuring that only suggestions that perform the same intervention for the same population are aligned.

[0141] Furthermore, based on the logical condition set and recommendation polarity corresponding to each standardized recommendation semantic box within the same class, the logical relationships between standardized recommendation semantic boxes within the same class are determined; the logical relationships include equivalence relations, expansion relations, contraction relations, and conflict relations, wherein:

[0142] When the logical condition set of two standardized recommendation semantic boxes within the same category When the two standardized recommendation semantic boxes are equivalent and have the same recommendation polarity, they are equivalent.

[0143] When two standardized recommendation semantic boxes within the same category have the same recommendation polarity and the set of logical conditions If it is an inclusion relationship, then the set of logical conditions A broader set of standardized recommendation semantic boxes and logical conditions The relationship between standardized recommendation semantic boxes with a narrower scope is a contraction relationship; logical condition set Narrower-scope standardized recommendation semantic boxes and logical condition sets The relationships between standardized recommendation semantic boxes with a broader scope are extended relationships;

[0144] When two standardized recommendation semantic boxes within the same category have inconsistent recommendation polarities and the logical condition sets of the two standardized recommendation semantic boxes are... When there is an overlap, the two standardized recommendation semantic boxes are in conflict.

[0145] The four relationships described in this invention represent the relationships between the contents of different clinical guidelines. Factual suggestion data fragments corresponding to two standardized recommendation semantic boxes with conflicting relationships are highlighted.

[0146] In a preferred embodiment, step S5, "constructing a causal evidence map based on intervention elements, mediating mechanism elements, and clinical outcome elements extracted from evidence-based support data fragments," specifically includes the following sub-steps:

[0147] The extracted intervention elements, mediation mechanism elements, and clinical outcome elements are all used as nodes in the causal evidence map;

[0148] The causal relationship between intervention elements or mediating mechanisms explicitly described in evidence-based data fragments and clinical outcome elements is used as the connecting edge between nodes. The direction of the connecting edge is determined according to the causal relationship, and the weight of the connecting edge is determined based on the literature research type, evidence quality grade, sample size, and effect size between the two nodes connected by the connecting edge. Edges with connecting edge weights lower than a preset weight threshold are deleted to obtain the causal evidence map.

[0149] In a preferred embodiment, step S6 is implemented by the following steps:

[0150] Step S61: Based on the transformed standardized recommendation semantic boxes, the logical relationships between standardized recommendation semantic boxes within the same category, and the causal evidence graph, determine the conclusion corresponding to the structured query object and the causal path supporting the conclusion; for each conclusion, based on the source identifier information of the conclusion... Obtain the clinical guideline name, issuing organization, version, and original paragraph number corresponding to this conclusion;

[0151] Step S62: Obtain all complete causal paths from all causal paths; where a complete causal path is a path with a complete chain of evidence, and determine the differences between different guideline-recommended solutions based on the set of logical conditions corresponding to the complete causal path.

[0152] Step S63: Store the conclusions corresponding to the complete causal path in the output file, and use the contents of the output file as the output result.

[0153] Furthermore, the incomplete causal path is identified as being caused by model illusion, thus blocking the output of this conclusion.

[0154] This invention proposes a multi-guide knowledge alignment and difference analysis device based on causal evidence graphs, comprising:

[0155] Query object generation module: Configured to use intelligent agent cluster collaboration to perform semantic extraction of unstructured clinical questions and generate structured query objects corresponding to unstructured clinical questions;

[0156] Three-layer library map generation module: configured to define a three-layer library map as an external reference source. The hierarchy of the three-layer library map from bottom to top includes a terminology definition layer, an evidence support layer, and a fact recommendation layer. The fact recommendation layer stores atomic recommendation items in the guidelines, which are clinical guidelines.

[0157] The data units to be retrieved from the guideline database and the medical research literature database are linked to the corresponding levels of the three-layer library map to form an instantiated three-layer library map; based on the structured query object, evidence-based supporting data fragments and factual suggestion data fragments are selected from the instantiated three-layer library map;

[0158] The recommendation semantic box generation module is configured to construct standardized recommendation semantic boxes based on factual suggestion data fragments; and to transform the qualitative descriptions corresponding to the standardized recommendation semantic boxes into a unified numerical space.

[0159] The logical relationship determination module is configured to calculate the similarity between different standardized recommendation semantic boxes and perform clustering based on the similarity; based on the set of logical conditions and recommendation polarity corresponding to each standardized recommendation semantic box in the same class, it determines the logical relationship between standardized recommendation semantic boxes in the same class; the logical relationship includes equivalence relationship, expansion relationship, contraction relationship and conflict relationship;

[0160] Causal evidence map generation module: configured to construct a causal evidence map based on intervention elements, mediation mechanism elements, and clinical outcome elements extracted from evidence-supporting data fragments;

[0161] The difference determination module is configured to determine the differences between different guideline recommendation schemes corresponding to structured query objects based on the transformed standardized recommendation semantic boxes, the logical relationships between standardized recommendation semantic boxes within the same category, and the causal evidence graph.

[0162] It should be understood that the processor in the embodiments of the present invention may be an integrated circuit chip with signal processing capabilities. In implementation, the steps of the above method embodiments can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor described above may be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules may reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0163] It is understood that the memory in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DR RAM). It should be noted that the memory used in the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0164] It should be understood that the above-described memory is exemplary and not a limiting description. For example, the memory in the embodiments of this application may also be static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus RAM (DR RAM), etc. That is to say, the memory in the embodiments of this application is intended to include, but is not limited to, these and any other suitable types of memory.

[0165] This application also provides a computer-readable storage medium for storing computer programs.

[0166] Optionally, the computer-readable storage medium can be applied to the terminal device in the embodiments of this application, and the computer program causes the computer to execute the corresponding processes implemented by the mobile terminal / terminal device in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0167] This application also provides a computer program product, including computer program instructions.

[0168] Optionally, the computer program product can be applied to the terminal device in the embodiments of this application, and the computer program instructions cause the computer to execute the corresponding processes implemented by the mobile terminal / terminal device in the various methods of the embodiments of this application. For the sake of brevity, they will not be described in detail here.

[0169] This application also provides a computer program.

[0170] Optionally, the computer program can be applied to the vehicle autonomous driving device in the embodiments of this application. When the computer program is run on a computer, it causes the computer to execute the corresponding processes implemented by the terminal device in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0171] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0172] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0173] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0174] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

Claims

1. A method for multi-guideline knowledge alignment and difference analysis based on causal evidence graphs, characterized in that, The methods include: Step S1: Utilize the collaborative action of intelligent agents to perform semantic extraction of unstructured clinical questions and generate structured query objects corresponding to the unstructured clinical questions; Step S2: Define a three-layer library map as an external reference source. The three-layer library map includes, from bottom to top, a terminology definition layer, an evidence support layer, and a fact recommendation layer. The fact recommendation layer stores atomic recommendation items in the guidelines, which are clinical guidelines. The data units to be retrieved from the guideline database and the medical research literature database are linked to the corresponding levels of the three-layer library map to form an instantiated three-layer library map; based on the structured query object, evidence-based supporting data fragments and factual suggestion data fragments are selected from the instantiated three-layer library map; Step S3: Construct standardized recommendation semantic boxes based on factual suggestion data fragments; transform the qualitative descriptions corresponding to the standardized recommendation semantic boxes into a unified numerical space; Step S4: Calculate the similarity between different standardized recommendation semantic boxes and perform clustering based on the similarity; determine the logical relationships between standardized recommendation semantic boxes within the same class based on the logical condition set and recommendation polarity corresponding to each standardized recommendation semantic box in the same class; logical relationships include equivalence relations, expansion relations, contraction relations and conflict relations; Step S5: Construct a causal evidence map based on the intervention elements, mediation mechanism elements, and clinical outcome elements extracted from the evidence-supporting data fragments; Step S6: Based on the transformed standardized recommendation semantic boxes, the logical relationships between standardized recommendation semantic boxes within the same category, and the causal evidence graph, determine the differences between different guideline recommendation schemes corresponding to the structured query object.

2. The method for multi-guideline knowledge alignment and difference analysis based on causal evidence graphs according to claim 1, characterized in that, Step S1 includes: Step S11: Utilize agent cluster collaboration to perform semantic extraction of unstructured clinical questions; Step S12: For each extracted semantic unit including fuzzy conditions, extract the clinical field, judgment operator, and threshold of the semantic unit including fuzzy conditions; construct a disjunction paradigm including clinical field, judgment operator, and threshold using logical operators; encapsulate the disjunction paradigm as the reconstruction result of the semantic unit including fuzzy conditions; all reconstruction results form a logical condition set. Step S13: For the extracted semantic units including clinical scoring tools, determine the scoring calculation model corresponding to the clinical scoring tools, and the clinical scoring tool context is composed of the clinical scoring tools and the scoring calculation model corresponding to the clinical scoring tools. Step S14: Based on the extracted other semantic units, logical condition set, and clinical scoring tool context, construct a structured query object corresponding to the unstructured clinical question; wherein, other semantic units are semantic units other than those including fuzzy conditions and those including clinical scoring tools. Structured Query Object The definition is as follows: ； in, Indicates the target audience. Indicates intervention measures, Represents a set of logical conditions. Indicates the result event, Indicates the context of the recommendation. Indicates the context of clinical scoring tools. This indicates a time window constraint.

3. The method for multi-guideline knowledge alignment and difference analysis based on causal evidence graphs according to claim 2, characterized in that, The terminology definition layer provides conceptual explanations based on a medical dictionary, while the evidence-based support layer stores mechanistic information extracted from medical research literature.

4. The method for multi-guideline knowledge alignment and difference analysis based on causal evidence graphs according to claim 3, characterized in that, In step S2, the data units to be retrieved from the guideline database and the medical research literature database are linked to the corresponding levels of the three-layer library map to form an instantiated three-layer library map, including: Step S21: Construct [fact, evidence-based, terminology] triplets based on medical dictionaries, guideline databases, and medical research literature databases; Step S22: Determine the relationship between factual entities and empirical entities, and the relationship between empirical entities and terminological entities in the triple; The relationship between factual entities and fact-based entities is established based on the following calculation formula: ； in, This is a set of directed links established between factual entities and evidence-based entities, used to construct the traceability link between the factual suggestion layer and the evidence-based support layer in the three-layer library graph. As a factual entity, For the sake of evidence-based practice, For comprehensive semantic information of factual entities, Embed vectors for the content of fact entities. For the comprehensive semantic information of the evidence-based entity, Embed the content vector of the evidence-based entity. It is a second normal form. The first threshold for similarity between targeted links is preset. To connect factual entities and evidence-based entities Semantic relation annotation; in and The cosine similarity calculation result is greater than or equal to the first directional link similarity threshold. In this case, Relation1 is marked as "referenced from"; The relationship between empirical entities and terminological entities is established based on the following calculation formula: ； in, This is a set of directed links established between evidence-based entities and terminological entities, used to construct a tracing link from the evidence-based support layer to the terminology definition layer in a three-layer library graph. For term entities, For the comprehensive semantic information of term entities, Embed vectors for the content of term entities. The second targeted link similarity threshold is preset; To connect with empirical evidence With term entities semantic relation annotation, in and The cosine similarity calculation result is greater than or equal to the second directional link similarity threshold. In this case, Marked as "defined as"; ； Where num = {1, 2, 3}, when num takes the values ​​1, 2, and 3 respectively, These represent the comprehensive semantic information of factual entities, the comprehensive semantic information of fact-based entities, and the comprehensive semantic information of terminological entities, respectively; when num takes values ​​of 1, 2, and 3, These are respectively factual entities, evidence-based entities, and terminological entities; , They are respectively The corresponding entity names and term names, , They are respectively The corresponding entity type and semantic type, , They are respectively The corresponding context content and context description text.

5. The method for multi-guideline knowledge alignment and difference analysis based on causal evidence graphs according to claim 4, characterized in that, In step S2, based on the structured query object, evidence-based supporting data fragments and factual suggestion data fragments are selected from the instantiated three-layer library graph, including: Step S23: Obtain the structured query object and search the instantiated three-layer library graph using a U-shaped search strategy; The U-shaped retrieval strategy involves first performing a top-down retrieval on the instantiated three-layer library map, and then performing a bottom-up retrieval. The top-down search is as follows: Generate a first set of query tags corresponding to the structured query objects using a large model. The first set of tags includes the structured query objects. middle The corresponding medical category tags; the tag summary corresponding to the first tag set is denoted as ; Based on the first query tag set, the fact suggestion layer, evidence-based support layer, and terminology definition layer are retrieved sequentially. The first result set obtained is { When num takes the values ​​1, 2, and 3 respectively, These are the search results corresponding to the fact suggestion layer, the evidence support layer, and the terminology definition layer when searching from top to bottom. A first evidence subgraph is constructed based on the first result set, and the nodes in the first evidence subgraph are entities in the first result set; The edges in the first evidence subgraph include: edges established between entities with semantic relationships in the first result set; edges established between fact entities marked "referenced to" in Relation1 and evidence-based entities in the first result set; edges established between entities in the first result set, An edge is established between a empirical entity marked "defined as" and a defined entity; in, , ;exist When the values ​​are 1 and 2 respectively, These represent the search results for the factual suggestion layer and the evidence-based support layer, respectively, when searching from top to bottom; with i taking values ​​of 1 and 2, respectively. These are the search results corresponding to the evidence-based support layer and the terminology definition layer when searching from top to bottom. for The first in A tag summary object, For calculation and The similarity between them; the sim function is a similarity function. This indicates the search for the tag summary object corresponding to the highest similarity; Bottom-up retrieval is: Generate a second set of query tags corresponding to the structured query objects using a large model. The second set of query tags includes the structured query objects. middle The corresponding medical category tags; the tag summary corresponding to the second tag set is denoted as ; Tag summary based on the second tag set Retrieve the terminology definition layer of the three-layer library map to obtain the preceding... Individual and The semantically most relevant term entity is denoted as ; It is a positive integer; Tag summary based on the second tag set Retrieve the evidence-based support layer of the three-layer library map to obtain the previous Individual and The most semantically relevant empirical evidence is denoted as ;based on Correction Semantic representation; Tag summary based on the second tag set Retrieve the fact suggestion layer of the three-layer library graph to obtain the previous Individual and The most semantically relevant fact entity is denoted as ;based on Correction Preliminary conclusions were drawn; The revised Corrected as well as Encapsulate as a second result set; The first evidence subgraph and the second result set are used together as the search results; Step S24: Filter out evidence-based supporting data fragments and factual suggestion data fragments from the search results, including: From the search results, select evidence-based supporting data fragments and factual suggestion data fragments with a comprehensive relevance score not lower than the preset relevance threshold.

6. The method for multi-guideline knowledge alignment and difference analysis based on causal evidence graphs according to claim 5, characterized in that, Step S3 includes: Step S31: Construct the octet corresponding to the fact suggestion data fragment based on the fact suggestion data fragment. The octet is used as a standardized recommendation semantic box; where, Indicates the target audience. Indicates intervention measures, Represents a set of logical conditions. Indicates the original recommendation strength. Indicates the level of original evidence; Indicates the recommended polarity. This indicates a recommendation to implement. This indicates that the suggestion should not be implemented; Indicate exceptions and restrictions, and record inapplicable scenarios or contraindications mentioned in clinical guidelines; This indicates the source identification information, including the clinical guideline name, issuing organization, version, and original paragraph number; Step S32: Using the original evidence level E as a qualitative description of the standardized recommendation semantic box, the qualitative description is converted into... The values ​​are between; where a predefined mapping standard reference table defines the mapping relationship between the original evidence level E and the numerical value.

7. The method for multi-guideline knowledge alignment and difference analysis based on causal evidence graphs according to claim 6, characterized in that, In step S4, the similarity between different standardized recommendation semantic boxes is calculated, including: Step S41: Two standardized recommendation semantic boxes to be used for similarity calculation and Standardize recommendation semantic boxes and In and All were normalized using a medical terminology normalization mapping function; Step S42: Determine the normalized result and The corresponding set of logical conditions and Intersection, Recommendation Polarity and Level of original evidence and Semantic similarity between them; determine the number of logical conditions num1 in the intersection, and determine... and The corresponding set of logical conditions and Given the number of logical conditions in the union set num2, calculate num1 / num2, and denote it as the first ratio; recommend polarity. and When consistent, polarity is recommended. and The consistency score between them is recorded as 1, and the recommendation polarity is... and When consistent, polarity is recommended. and The consistency score between them is recorded as 0; based on the first ratio and recommendation polarity and Consistency score and level of original evidence and Determining the semantic similarity between and The similarity between them.

8. The method for multi-guideline knowledge alignment and difference analysis based on causal evidence graphs according to claim 7, characterized in that, In step S4, based on the logical condition set and recommendation polarity corresponding to each standardized recommendation semantic box within the same class, the logical relationship between standardized recommendation semantic boxes within the same class is determined, including: When the logical condition set of two standardized recommendation semantic boxes within the same category When the two standardized recommendation semantic boxes are equivalent and have the same recommendation polarity, they are equivalent. When two standardized recommendation semantic boxes within the same category have the same recommendation polarity and the set of logical conditions If it is an inclusion relationship, then the set of logical conditions A broader set of standardized recommendation semantic boxes and logical conditions The relationship between standardized recommendation semantic boxes with a narrower scope is a contraction relationship; logical condition set Narrower-scope standardized recommendation semantic boxes and logical condition sets The relationships between standardized recommendation semantic boxes with a broader scope are extended relationships; When two standardized recommendation semantic boxes within the same category have inconsistent recommendation polarities and the logical condition sets of the two standardized recommendation semantic boxes are... When there is an overlap, the two standardized recommendation semantic boxes are in conflict.

9. The method for multi-guideline knowledge alignment and difference analysis based on causal evidence graphs according to claim 7, characterized in that, In step S5, a causal evidence map is constructed based on the intervention elements, mediation mechanism elements, and clinical outcome elements extracted from the evidence-based support data fragments, including: The extracted intervention elements, mediation mechanism elements, and clinical outcome elements are all used as nodes in the causal evidence map; The causal relationship between intervention elements or mediating mechanisms explicitly described in evidence-based data fragments and clinical outcome elements is used as the connecting edge between nodes. The direction of the connecting edge is determined according to the causal relationship, and the weight of the connecting edge is determined based on the literature research type, evidence quality grade, sample size, and effect size between the two nodes connected by the connecting edge. Edges with connecting edge weights lower than a preset weight threshold are deleted to obtain the causal evidence map.

10. The method for multi-guide knowledge alignment and difference analysis based on causal evidence graphs according to claim 9, characterized in that, Step S6 includes: Step S61: Based on the transformed standardized recommendation semantic boxes, the logical relationships between standardized recommendation semantic boxes within the same category, and the causal evidence graph, determine the conclusion corresponding to the structured query object and the causal path supporting the conclusion; for each conclusion, based on the source identifier information of the conclusion... Obtain the clinical guideline name, issuing organization, version, and original paragraph number corresponding to the conclusion; Step S62: Obtain all complete causal paths from all causal paths; where a complete causal path is a path with a complete chain of evidence, and determine the differences between different guideline-recommended solutions based on the set of logical conditions corresponding to the complete causal path. Step S63: Store the conclusions corresponding to the complete causal path in the output file, and use the contents of the output file as the output result.

Images