A method, device and medium for constructing a medical and nursing knowledge graph database based on an information-based innovation system

Abstract

The present invention relates to the field of data processing technology. The present invention discloses a method, device and medium for constructing a medical and nursing knowledge graph database based on an information innovation system, including: calculating the sparsity of each original node in a basic medical and nursing knowledge graph, and screening out sparse nodes according to the sparsity; traversing the basic medical and nursing knowledge graph, obtaining original nodes that are not directly connected to the sparse nodes, marking them as nodes to be mined, and judging whether there is a potential relationship between the sparse nodes and the nodes to be mined; when it is determined that there is a potential relationship, inferring a specific representation of the potential relationship between the sparse nodes and the nodes to be mined; creating and adding new relationship edges to the basic medical and nursing knowledge graph according to the specific representation of the potential relationship to form an optimized medical and nursing knowledge graph; the present invention can enrich the structure of the graph, improve the overall integrity of the medical and nursing knowledge graph, and thus enhance the search accuracy.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and more specifically, to a method, device and medium for constructing a medical and nursing knowledge graph database based on an information-based innovation system. Background Art

[0002] In recent years, with the development of technologies such as big data and artificial intelligence, data-driven medical and health services have gradually become possible; however, data in the field of elderly medical care often present multi-source heterogeneous characteristics, including hospital electronic medical records, nursing records, personal health monitoring data, lifestyle data, etc.; these data vary greatly in form, source, structure, etc., and are difficult to effectively integrate and utilize through traditional database management methods.

[0003] At present, there is a lack of a systematic and intelligent method to integrate these multi-source heterogeneous data, so as to provide accurate and personalized medical and nursing services for the elderly. Although there are some relevant technical literatures, such as the patent with authorization announcement number CN113704492B, which discloses a method and system for constructing a knowledge graph of elderly care data; however, research and practical application of the above methods and existing technologies have found that the above methods and existing technologies cannot filter out sparse nodes in the knowledge graph, and further, cannot perform relationship mining and expansion on sparse nodes, thereby making it difficult to enrich the structure of the graph, reduce information isolation, and improve the integrity of the knowledge graph, which in turn leads to inaccurate search answers when users use the knowledge graph for self-medical diagnosis and understanding of personalized treatment plans. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a method, device and medium for constructing a medical and nursing knowledge graph database based on an information-based innovation system.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for constructing a medical and nursing knowledge graph database based on an information innovation system, the method comprising:

[0007] S101: Obtain a pre-built basic medical and nursing knowledge graph, calculate the sparsity of each original node in the basic medical and nursing knowledge graph, and filter out sparse nodes according to the sparsity;

[0008] S102: traverse the basic medical knowledge graph, obtain the original nodes that are not directly connected to the sparse nodes, mark them as nodes to be mined, and determine whether there is a potential relationship between the sparse nodes and the nodes to be mined;

[0009] S103: when it is determined that there is a potential relationship, a specific representation of the potential relationship between the sparse node and the node to be mined is inferred;

[0010] S104: Create and add new relationship edges for the basic medical and nursing knowledge graph according to the specific representation of the potential relationship to form an optimized medical and nursing knowledge graph, import the optimized medical and nursing knowledge graph into a preset graph database, and obtain a medical and nursing knowledge graph database.

[0011] Furthermore, the pre-construction logic of the basic medical and nursing knowledge graph is as follows:

[0012] The corpus data is segmented to remove meaningless words or symbols, and part-of-speech tagging is performed to form a subject-predicate-object structure, and all subject and object phrases are combined into a first phrase set, and all predicate phrases are combined into a second phrase set;

[0013] According to predefined rules provided by a preset knowledge base, extract the corresponding subject or object phrases in the first phrase set as target entities, and extract the corresponding predicate phrases in the second phrase set as the relationship between the entities;

[0014] According to the relationship between entities, the target entity is associated and bound to construct a triple containing 〈entity, relationship, entity〉;

[0015] The target entity in the triple is taken as the original node, and the relationship between entities is taken as the original relationship edge between the original nodes to establish a basic medical and nursing knowledge graph.

[0016] Furthermore, before associating and binding the target entity, the following steps are included:

[0017] Search multiple target entities according to a preset regular expression rule, and mark the searched multiple target entities as suspected identical entities;

[0018] Obtain the attributes and attribute values ​​of each suspected identical entity in the corpus data after word segmentation processing, and extract attribute values ​​with the same attributes;

[0019] The Word2Vec model is used to convert the attribute values ​​of the same attribute into attribute vectors, and the difference coefficient of the attribute vectors of every two suspected identical entities is measured; the specific calculation formula is as follows: ; Where: is the difference coefficient, is the attribute value of the i-th identical attribute of a suspected identical entity, is the attribute value of the i-th identical attribute of another suspected identical entity, and N is the total number of identical attributes;

[0020] Two suspected identical entities whose difference coefficient is greater than or equal to a preset difference coefficient threshold are identified as the same target entity, and the entities are merged to obtain a fused target entity;

[0021] Two suspected identical entities whose difference coefficient is less than a preset difference coefficient threshold are identified as non-identical target entities and there is no need to merge the entities.

[0022] Furthermore, the calculation of the sparsity of each original node in the basic medical knowledge graph includes:

[0023] Traverse the basic medical knowledge graph, record each original node and the neighbor set of each original node, and the neighbor set includes the number N(v) of neighbor nodes directly adjacent to the original node;

[0024] Obtain the number of edges E(v) directly connected between the original node and each neighboring node, substitute the number of adjacent nodes N(v) and the number of edges E(v) into the pre-built calculation model of the local clustering coefficient C(v), and calculate the local clustering coefficient C(v) of each original node;

[0025] The calculation model of the local clustering coefficient C(v) is expressed as follows: ; In the formula, represents the neighbor nodes directly connected to the original node, For each neighbor node of the original node Count them all once, in other words, sum the number of these neighbor nodes;

[0026] Substitute the local clustering coefficient C(v) of the original node into the sparsity calculation formula to obtain the sparsity of each original node ; The specific calculation formula is as follows: ; In the formula, Represents a natural constant.

[0027] Furthermore, the step of selecting sparse nodes according to sparsity includes:

[0028] The sparsity Compare with a preset sparsity threshold;

[0029] If the sparsity If it is greater than or equal to the preset sparsity threshold, the corresponding original node is determined to be a sparse node;

[0030] If the sparsity is small If the value of the node is less than the preset sparsity threshold, the corresponding original node is determined to be a non-sparse node.

[0031] Furthermore, the step of determining whether there is a potential relationship between the sparse nodes and the nodes to be mined includes:

[0032] Extract any sparse node and a node to be mined in the basic medical knowledge graph, combine them into a first node pair, obtain all original nodes directly connected to a sparse node in the first node pair, and use them as the first intermediate node, and obtain all original nodes directly connected to another node to be mined in the second node pair, and use them as the second intermediate node;

[0033] Perform a non-repetitive combination on each first intermediate node and the second intermediate node to obtain M second node pairs, and input the M second node pairs into a preconfigured relationship analysis model for identification to obtain an identification result, wherein the identification result includes the existence of a potential relationship and the non-existence of a potential relationship, and M is an integer greater than zero;

[0034] The training logic of the preconfigured relationship analysis model is as follows:

[0035] Acquire historical training data for identifying potential relationships, and divide the training data for identifying potential relationships into a potential relationship training set and a potential relationship test set, wherein the training data for identifying potential relationships includes a plurality of second node pair sets and their corresponding annotation labels;

[0036] Each of the second node pair sets includes M second node pairs, and the annotation labels include the existence of a potential relationship and the non-existence of a potential relationship;

[0037] Initialize the classifier, use the second node pair set in the potential relationship training set as the input of the classifier, and use the labeled labels in the potential relationship training set as the output of the classifier, train the classifier, and obtain a classification network that has been preliminarily trained;

[0038] The classification network that has been initially trained is validated using the potential relationship test set, and a classification network with a test accuracy greater than or equal to a preset threshold is output as a preconfigured relationship analysis model.

[0039] Furthermore, the logic for generating the annotated labels in the training data of the historical recognition potential relationship is as follows:

[0040] Obtain each first intermediate node and second intermediate node in the M second node pairs;

[0041] The cosine similarity coefficients of the first intermediate node and the second intermediate node are calculated, and the cosine similarity coefficients are cumulatively calculated to obtain the cumulative cosine similarity coefficient; the specific calculation formula is: ; Where: is the cumulative cosine similarity coefficient, is the cosine similarity coefficient, is the first intermediate node in the jth second node pair, is the first intermediate node in the jth second node pair; wherein, The calculation formula is ; Where: is the norm;

[0042] The cumulative cosine similarity coefficient is compared with the preset cumulative cosine similarity coefficient threshold. If the cumulative cosine similarity coefficient is greater than or equal to the preset cumulative cosine similarity coefficient threshold, it is determined that there is a potential relationship between the sparse nodes in the corresponding first node pair and the nodes to be mined. Conversely, if the cumulative cosine similarity coefficient is less than the preset cumulative cosine similarity coefficient threshold, it is determined that there is no potential relationship between the sparse nodes in the corresponding first node pair and the nodes to be mined.

[0043] Furthermore, the specific representation of the potential relationship between the inferred sparse nodes and the nodes to be mined includes:

[0044] Using the Apriori algorithm to obtain frequent item sets of sparse nodes and nodes to be mined, and obtaining part-of-speech collocation forms of the sparse nodes and nodes to be mined, the part-of-speech collocation forms including subject and subject, subject and object, and object and object;

[0045] Extract frequent item sets and entity combinations in frequent item sets that are greater than the support threshold, and obtain the part-of-speech collocation form of the entity combination;

[0046] If the entity combination has the same part-of-speech collocation form as the sparse node and the node to be mined, the relationship between the entities in the entity combination is used as a specific representation of the potential relationship between the sparse node and the node to be mined;

[0047] If the part-of-speech collocation forms of the entity combination and the sparse nodes and the nodes to be mined are inconsistent, the relationship between the entities possessed by the entity combination will not be used as a specific representation of the potential relationship between the sparse nodes and the nodes to be mined.

[0048] An electronic device comprises a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, any one of the above-mentioned methods for constructing a medical and nursing knowledge graph database based on an information and communication system is implemented.

[0049] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed, implements any of the above-mentioned methods for constructing a medical and nursing knowledge graph database based on an information and communication system.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] The present application discloses a method, device and medium for constructing a medical and nursing knowledge graph database based on an information-based innovation system, including: calculating the sparsity of each original node in a basic medical and nursing knowledge graph, and screening out sparse nodes according to the sparsity; traversing the basic medical and nursing knowledge graph, obtaining original nodes that are not directly connected to the sparse nodes, and marking them as nodes to be mined, and judging whether there is a potential relationship between the sparse nodes and the nodes to be mined; when it is determined that there is a potential relationship, inferring a specific representation of the potential relationship between the sparse nodes and the nodes to be mined; creating and adding new relationship edges for the basic medical and nursing knowledge graph according to the specific representation of the potential relationship to form an optimized medical and nursing knowledge graph; based on the above features, the present invention can screen out sparse nodes in the knowledge graph, and further, by performing relationship mining and expansion on sparse nodes, it is beneficial to enrich the structure of the graph, reduce the situation of information isolation, and improve the integrity of the knowledge graph, thereby allowing users to enhance the accuracy of search answers when using the knowledge graph for self-medical and nursing diagnosis and understanding personalized treatment plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A flowchart of a method for constructing a medical and nursing knowledge graph database based on an information innovation system provided by the present invention;

[0053] Figure 2 A schematic diagram of the structure of an electronic device provided by the present invention;

[0054] Figure 3 A schematic diagram of the structure of a computer-readable storage medium provided by the present invention. DETAILED DESCRIPTION

[0055] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] Example 1

[0057] See also Figure 1 As shown, this embodiment discloses a method for constructing a medical and nursing knowledge graph database based on an information innovation system, the method comprising:

[0058] S101: Obtain a pre-built basic medical and nursing knowledge graph, calculate the sparsity of each original node in the basic medical and nursing knowledge graph, and filter out sparse nodes according to the sparsity;

[0059] In practice, the pre-construction logic of the basic medical and nursing knowledge graph is as follows:

[0060] The corpus data is segmented to remove meaningless words or symbols, and part-of-speech tagging is performed to form a subject-predicate-object structure, and all subject and object phrases are combined into a first phrase set, and all predicate phrases are combined into a second phrase set;

[0061] It should be noted that: the word segmentation process can be implemented by applying natural language processing tools, and the natural language processing tools include but are not limited to Jieba word segmenter, NLTK, etc.; the meaningless words and symbols include but are not limited to repeated words, stop words, onomatopeia, onomatopoeia, "," "!" and ".", etc.;

[0062] According to predefined rules provided by a preset knowledge base, extract the corresponding subject or object phrases in the first phrase set as target entities, and extract the corresponding predicate phrases in the second phrase set as the relationship between the entities;

[0063] It should be noted that: the preset knowledge base defines the predefined rules required for establishing the basic medical knowledge graph, and the predefined rules specifically define and define the target entity and the relationship between entities, and the target entity includes but is not limited to diseases, symptoms, drugs, treatment plans, clinical manifestations and management measures, etc.; for example, there is a phrase "type 1 diabetes" in the preset knowledge base, and the corresponding subject or object phrase in the first phrase set is "type 1 diabetes", then "type 1 diabetes" is taken as the target entity. Similarly, the logic for determining the relationship between entities is the same, and it will not be repeated here; wherein, the relationship between entities is specifically determined according to the qualifiers in the preset knowledge base, such as the qualifiers such as "cause", "treat" and "exist";

[0064] The preset knowledge base may be established by the technicians themselves in advance, or the existing knowledge base may be directly called, and the existing knowledge base includes standardized medical and elderly care knowledge bases; wherein the standardized medical knowledge base includes but is not limited to UMLS (Unified Medical Language System), SNOMED CT (Systematized Nomenclature of Medicine-Clinical Terms) or CD (International Classification of Diseases), and the standardized elderly care knowledge base includes but is not limited to MDS (Minimum Data Set) or OMAHA system;

[0065] According to the relationship between entities, the target entity is associated and bound to construct a triple containing 〈entity, relationship, entity〉;

[0066] In a specific implementation, before associating and binding the target entity, the following steps are included:

[0067] Search multiple target entities according to a preset regular expression rule, and mark the searched multiple target entities as suspected identical entities;

[0068] It should be noted that the preset regular expression rules are set in advance by the technical personnel. For example, according to the preset regular expression rule of "\d+type diabetes", "type 1 diabetes" and "type 2 diabetes" can be directly searched out;

[0069] Obtain the attributes and attribute values ​​of each suspected identical entity in the corpus data after word segmentation processing, and extract attribute values ​​with the same attributes;

[0070] For example, suppose that two suspected identical entities are searched, namely "type 1 diabetes" and "type 2 diabetes", and the same attributes of "type 1 diabetes" and "type 2 diabetes" include the patient population and symptom description, wherein the patient population of "type 1 diabetes" is adolescents (i.e., attribute value), and the patient population of "type 2 diabetes" is the elderly (i.e., attribute value), and the symptom description of "type 1 diabetes" is insulin dependence, while the symptom description of "type 2 diabetes" is obesity;

[0071] It should be understood that: in structured corpus data, the attributes and attribute values ​​of each suspected identical entity exist in the relational data table (library), which can be directly extracted, while for unstructured corpus data, they can be extracted through pre-trained models, pre-defined regular expressions or rules, and the pre-trained models, pre-defined attribute extraction regular expressions or rules are manually set by technicians;

[0072] The Word2Vec model is used to convert the attribute values ​​of the same attribute into attribute vectors, and the difference coefficient of the attribute vectors of every two suspected identical entities is measured; the specific calculation formula is as follows: ; Where: is the difference coefficient, is the attribute value of the i-th identical attribute of a suspected identical entity, is the attribute value of the i-th identical attribute of another suspected identical entity, and N is the total number of identical attributes;

[0073] Two suspected identical entities whose difference coefficient is greater than or equal to a preset difference coefficient threshold are identified as the same target entity, and the entities are merged to obtain a fused target entity;

[0074] Two suspected identical entities whose difference coefficient is less than a preset difference coefficient threshold are identified as different target entities, and there is no need to merge the entities;

[0075] It is understandable that most existing knowledge graph disambiguation methods are implemented by direct comparison, that is, directly calculating the similarity of two target entities and then judging whether they are the same target entity. This method has various drawbacks and cannot capture the semantic features of each target entity, which easily leads to the phenomenon of no judgment. For example, for "type 1 diabetes" and "type 2 diabetes", directly calculating the similarity between the two often determines that the two are the same. However, according to medical theory, the two are not the same. Therefore, the constructed medical knowledge graph is prone to search errors or inaccurate searches in actual use.

[0076] In another specific embodiment, before associating and binding the target entity, the method further includes: aligning the relationship between the entities;

[0077] It should be understood that alignment can help identify and merge semantically identical relationships, reduce duplicate relationship entries in the graph, make the subsequent basic medical knowledge graph structure more concise, and help improve query and reasoning efficiency;

[0078] It should be noted that the alignment of the relationship between entities can be achieved based on the similarity comparison of word vectors or based on the comparison of string similarity, which is not limited in this embodiment, and any existing method for achieving the alignment of the relationship between entities can be used as an application object of the present invention;

[0079] The target entity in the triple is taken as the original node, and the relationship between entities is taken as the original relationship edge between the original nodes to establish a basic medical and nursing knowledge graph.

[0080] In implementation, the calculation of the sparsity of each original node in the basic medical knowledge graph includes:

[0081] Traverse the basic medical knowledge graph, record each original node and the neighbor set of each original node, and the neighbor set includes the number N(v) of neighbor nodes directly adjacent to the original node;

[0082] It should be understood that: the neighbor set refers to the set of all (original) nodes directly connected to a specific (original) node;

[0083] Obtain the number of edges E(v) directly connected between the original node and each neighboring node, substitute the number of adjacent nodes N(v) and the number of edges E(v) into the pre-built calculation model of the local clustering coefficient C(v), and calculate the local clustering coefficient C(v) of each original node;

[0084] The calculation model of the local clustering coefficient C(v) is expressed as follows: ; In the formula, represents the neighbor nodes directly connected to the original node, For each neighbor node of the original node Count them all once, in other words, sum the number of these neighbor nodes;

[0085] Substitute the local clustering coefficient C(v) of the original node into the sparsity calculation formula to obtain the sparsity of each original node ; The specific calculation formula is as follows: ; In the formula, represents a natural constant;

[0086] In implementation, the step of selecting sparse nodes according to sparsity includes:

[0087] The sparsity Compare with a preset sparsity threshold;

[0088] If the sparsity If it is greater than or equal to the preset sparsity threshold, the corresponding original node is determined to be a sparse node;

[0089] If the sparsity is small If the value of the node is less than the preset sparsity threshold, the corresponding original node is determined to be a non-sparse node;

[0090] It should be noted that sparsity The larger the value is, the lower the local clustering coefficient C(v) of the corresponding original node is, which means that the number of connections between the corresponding original node and its directly connected neighbor nodes is small, and the original node is relatively isolated in the graph. On the contrary, the sparsity The smaller it is, the larger the local clustering coefficient C(v) of the corresponding original node is, which means that the number of connections between the corresponding original node and its directly connected neighbor nodes is large, and the information of the original node in the graph is relatively rich;

[0091] It can be further understood that original nodes with relatively rich information can release more and more accurate effective information when being searched, which is conducive to helping users make better medical and nursing diagnoses and provide medical and nursing solutions.

[0092] S102: traverse the basic medical knowledge graph, obtain the original nodes that are not directly connected to the sparse nodes, mark them as nodes to be mined, and determine whether there is a potential relationship between the sparse nodes and the nodes to be mined;

[0093] For example, assuming that there are two triples, namely ⟨weight loss, will exist, type 1 diabetes> and ⟨type 1 diabetes, management measures, high-fiber diet>, if "weight loss" is a sparse node, then "high-fiber diet" is marked as a raw node that is not directly connected to the sparse node; similarly, "weight loss" and "high-fiber diet" are marked as raw nodes directly connected to "type 1 diabetes";

[0094] In implementation, the step of determining whether there is a potential relationship between the sparse nodes and the nodes to be mined includes:

[0095] Extract any sparse node and a node to be mined in the basic medical knowledge graph, combine them into a first node pair, obtain all original nodes directly connected to a sparse node in the first node pair, and use them as the first intermediate node, and obtain all original nodes directly connected to another node to be mined in the second node pair, and use them as the second intermediate node;

[0096] Perform a non-repetitive combination on each first intermediate node and the second intermediate node to obtain M second node pairs, and input the M second node pairs into a preconfigured relationship analysis model for identification to obtain an identification result, wherein the identification result includes the existence of a potential relationship and the non-existence of a potential relationship, and M is an integer greater than zero;

[0097] For example, suppose there is a sparse node and a node to be mined, namely "type 1 diabetes" and "type 2 diabetes", and "type 1 diabetes" and "type 2 diabetes" each have only two triples, among which the two triples of "type 1 diabetes" are ⟨type 1 diabetes, leading to, weight loss> and ⟨type 1 diabetes, High-fiber diet, kidney disease>, while the two triples of “type 2 diabetes” are ⟨type 2 diabetes, complications, kidney disease> and ⟨type 2 diabetes, treatment methods, hypoglycemic drugs>. According to the explanation of the original nodes that are not directly connected or directly connected above, the original nodes directly connected to “type 1 diabetes” include “weight loss” and “high-fiber diet”, while the original nodes directly connected to “type 2 diabetes” include “kidney disease” and “hypoglycemic drugs”. Therefore, after non-repeated combinations of the first intermediate node and the second intermediate node, four second node pairs can be obtained, namely, the second node pair including “weight loss” and “kidney disease”, the second node pair including “weight loss” and “hypoglycemic drugs”, the second node pair including “kidney disease” and “kidney disease”, and the second node pair including “kidney disease” and “hypoglycemic drugs”;

[0098] The training logic of the preconfigured relationship analysis model is as follows:

[0099] Acquire historical training data for identifying potential relationships, and divide the training data for identifying potential relationships into a potential relationship training set and a potential relationship test set, wherein the training data for identifying potential relationships includes a plurality of second node pair sets and their corresponding annotation labels;

[0100] Each of the second node pair sets includes M second node pairs, and the annotation labels include the existence of a potential relationship and the non-existence of a potential relationship;

[0101] Specifically, the logic for generating the annotated labels in the training data of the historical recognition potential relationship is as follows:

[0102] Obtain each first intermediate node and second intermediate node in the M second node pairs;

[0103] The cosine similarity coefficients of the first intermediate node and the second intermediate node are calculated, and the cosine similarity coefficients are cumulatively calculated to obtain the cumulative cosine similarity coefficient; the specific calculation formula is: ; Where: is the cumulative cosine similarity coefficient, is the cosine similarity coefficient, is the first intermediate node in the jth second node pair, is the first intermediate node in the jth second node pair; wherein, The calculation formula is ; Where: is the norm;

[0104] Compare the cumulative cosine similarity coefficient with the preset cumulative cosine similarity coefficient threshold value. If the cumulative cosine similarity coefficient is greater than or equal to the preset cumulative cosine similarity coefficient threshold value, it is determined that there is a potential relationship between the sparse node in the corresponding first node pair and the node to be mined. Conversely, if the cumulative cosine similarity coefficient is less than the preset cumulative cosine similarity coefficient threshold value, it is determined that there is no potential relationship between the sparse node in the corresponding first node pair and the node to be mined.

[0105] Exemplarily, following the above example, it can be seen from the above example that there is a first node pair including a sparse node and a node to be mined, wherein the sparse node and the node to be mined are "type 1 diabetes" and "type 2 diabetes" respectively, and four second node pairs are mined based on "type 1 diabetes" and "type 2 diabetes", namely, a second node pair including "weight loss" and "nephropathy", a second node pair including "weight loss" and "hypoglycemic drugs", a second node pair including "nephropathy" and "nephropathy", and a second node pair including "nephropathy" and "hypoglycemic drugs"; assuming that after calculating the cumulative cosine similarity coefficient of these four second node pairs, it is found that the cumulative cosine similarity coefficient is greater than or equal to the preset cumulative cosine similarity coefficient threshold, then it is determined that there is a potential relationship between the sparse node in the corresponding first node pair and the node to be mined, that is, it is determined that there is a potential relationship between "type 1 diabetes" and "type 2 diabetes", on the contrary, it is determined that there is no potential relationship between "type 1 diabetes" and "type 2 diabetes";

[0106] Initialize the classifier, use the second node pair set in the potential relationship training set as the input of the classifier, and use the labeled labels in the potential relationship training set as the output of the classifier, train the classifier, and obtain a classification network that has been preliminarily trained;

[0107] The classification network that has been initially trained is validated using the potential relationship test set, and a classification network with a test accuracy greater than or equal to a preset test accuracy threshold is output as a preconfigured relationship analysis model;

[0108] Wherein, the classifier is one of a convolutional neural network or a recurrent neural network.

[0109] S103: when it is determined that there is a potential relationship, a specific representation of the potential relationship between the sparse node and the node to be mined is inferred;

[0110] In implementation, the specific representation of the potential relationship between the inferred sparse nodes and the nodes to be mined includes:

[0111] Using the Apriori algorithm to obtain frequent item sets of sparse nodes and nodes to be mined, and obtaining part-of-speech collocation forms of the sparse nodes and nodes to be mined, the part-of-speech collocation forms including subject and subject, subject and object, and object and object;

[0112] Extract frequent item sets and entity combinations in frequent item sets that are greater than the support threshold, and obtain the part-of-speech collocation form of the entity combination;

[0113] If the entity combination has the same part-of-speech collocation form as the sparse node and the node to be mined, the relationship between the entities in the entity combination is used as a specific representation of the potential relationship between the sparse node and the node to be mined;

[0114] If the part-of-speech collocation forms of the entity combination are inconsistent with the sparse nodes and the nodes to be mined, the relationship between the entities possessed by the entity combination will not be used as a specific representation of the potential relationship between the sparse nodes and the nodes to be mined;

[0115] It should be noted that: when the part-of-speech collocation form of the entity combination is inconsistent with the part-of-speech collocation form of the sparse nodes and the nodes to be mined, it means that the relationship between the entities of the entity combination does not conform to the relationship between the entities of the sparse nodes and the nodes to be mined. At this time, the relationship between the entities can be manually generated and assigned by technical personnel;

[0116] It should also be noted that: the frequent item set is obtained by screening from M second node pairs;

[0117] Exemplarily, it is assumed that there are four second node pairs, namely, a second node pair containing "diabetes" and "increased blood sugar", a second node pair containing "weight loss" and "hypoglycemic drugs", a second node pair containing "diabetes" and "nephropathy", and a second node pair containing "weight loss" and "loss of appetite"; it is also assumed that the part-of-speech collocation form of the sparse node and the node to be mined is "subject and object" and is specifically "diabetes" and "weight loss" respectively. If the support threshold is set to 2, the frequent item set greater than the support threshold is obtained as: {diabetes, increased blood sugar} (support is 3), and the part-of-speech collocation form of the entity combination (i.e., "diabetes" and "increased blood sugar") in the frequent item set is also "subject and object", and the relationship between entities owned by the entity combination is "lead to", then the relationship between the entities is assigned to the sparse node and the node to be mined as a specific representation of the potential relationship between the sparse node and the node to be mined, and a new triplet of ⟨diabetes, lead to, weight loss> is obtained;

[0118] S104: creating and adding new relationship edges to the basic medical and nursing knowledge graph according to the specific representation of the potential relationship to form an optimized medical and nursing knowledge graph, and importing the optimized medical and nursing knowledge graph into a preset graph database to obtain a medical and nursing knowledge graph database;

[0119] Specifically, the optimized medical and nursing knowledge graph is imported into a preset graph database by using the graph database API or batch import tool; the graph database can be selected in advance by the technician according to the data scale, complexity, and performance requirements, wherein the graph database is not limited to a specific one of Neo4j, Amazon Neptune, or ArangoDB, etc.;

[0120] It should be noted that: when the potential relationship between the sparse nodes and the nodes to be mined is inferred, the sparse nodes and the nodes to be mined can be combined into new triples based on the potential relationship. By adding the new triples into the basic medical and nursing knowledge graph, the medical and nursing knowledge graph can be optimized.

[0121] Example 2

[0122] See also Figure 2 As shown, this embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements any one of the above methods provided by a method for constructing a medical and nursing knowledge graph database based on an information technology innovation system.

[0123] Since the electronic device introduced in this embodiment is an electronic device used to implement a method for constructing a medical and nursing knowledge graph database based on a trusted innovation system in the embodiment of this application, based on the method for constructing a medical and nursing knowledge graph database based on a trusted innovation system introduced in the embodiment of this application, the technical personnel in this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application is not described in detail here. As long as the technical personnel in this field implement the electronic device used in the method for constructing a medical and nursing knowledge graph database based on a trusted innovation system in the embodiment of this application, it belongs to the scope of protection of this application.

[0124] Example 3

[0125] See also Figure 3 As shown, this embodiment discloses a computer-readable storage medium, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements any one of the above methods provided by a method for constructing a medical and nursing knowledge graph database based on an information technology innovation system.

[0126] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters, weights and thresholds in the formula are set by technicians in this field according to actual conditions.

[0127] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired network or a wireless network. The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media sets. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD) or a semiconductor medium. The semiconductor medium may be a solid-state hard disk.

[0128] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0129] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0130] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only one, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0131] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0132] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0133] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0134] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for constructing a medical and nursing knowledge graph database based on an information innovation system, characterized in that: The method comprises: S101: Obtain a pre-built basic medical and nursing knowledge graph, calculate the sparsity of each original node in the basic medical and nursing knowledge graph, and filter out sparse nodes according to the sparsity; Among them, the pre-construction logic of the basic medical and nursing knowledge graph is as follows: The corpus data is segmented to remove meaningless words or symbols, and part-of-speech tagging is performed to form a subject-predicate-object structure, and all subject and object phrases are combined into a first phrase set, and all predicate phrases are combined into a second phrase set; According to predefined rules provided by a preset knowledge base, extract the corresponding subject or object phrases in the first phrase set as target entities, and extract the corresponding predicate phrases in the second phrase set as the relationship between the entities; According to the relationship between entities, the target entity is associated and bound to construct a triple containing 〈entity, relationship, entity〉; The target entity in the triple is used as the original node, and the relationship between the entities is used as the original relationship edge between the original nodes to establish a basic medical and nursing knowledge graph; The step of calculating the sparsity of each original node in the basic medical knowledge graph includes: Traverse the basic medical knowledge graph, record each original node and the neighbor set of each original node, and the neighbor set includes the number N(v) of neighbor nodes directly adjacent to the original node; Obtain the number of edges E(v) directly connected between the original node and each neighboring node, substitute the number of adjacent nodes N(v) and the number of edges E(v) into the pre-built calculation model of the local clustering coefficient C(v), and calculate the local clustering coefficient C(v) of each original node; The calculation model of the local clustering coefficient C(v) is expressed as follows: ; In the formula, represents the neighbor nodes directly connected to the original node, For each neighbor node of the original node Count them all once, in other words, sum the number of these neighbor nodes; Substitute the local clustering coefficient C(v) of the original node into the sparsity calculation formula to obtain the sparsity of each original node ; The specific calculation formula is as follows: ; In the formula, represents a natural constant; S102: traverse the basic medical knowledge graph, obtain the original nodes that are not directly connected to the sparse nodes, mark them as nodes to be mined, and determine whether there is a potential relationship between the sparse nodes and the nodes to be mined; The step of determining whether there is a potential relationship between the sparse nodes and the nodes to be mined includes: Extract any sparse node and a node to be mined in the basic medical knowledge graph, combine them into a first node pair, obtain all original nodes directly connected to a sparse node in the first node pair, and use them as the first intermediate node, and obtain all original nodes directly connected to another node to be mined in the second node pair, and use them as the second intermediate node; Perform a non-repetitive combination on each first intermediate node and the second intermediate node to obtain M second node pairs, and input the M second node pairs into a preconfigured relationship analysis model for identification to obtain an identification result, wherein the identification result includes the existence of a potential relationship and the non-existence of a potential relationship, and M is an integer greater than zero; The training logic of the preconfigured relationship analysis model is as follows: Acquire historical training data for identifying potential relationships, and divide the training data for identifying potential relationships into a potential relationship training set and a potential relationship test set, wherein the training data for identifying potential relationships includes a plurality of second node pair sets and their corresponding annotation labels; Each of the second node pair sets includes M second node pairs, and the annotation labels include the existence of a potential relationship and the non-existence of a potential relationship; Initialize the classifier, use the second node pair set in the potential relationship training set as the input of the classifier, and use the labeled labels in the potential relationship training set as the output of the classifier, train the classifier, and obtain a classification network that has been preliminarily trained; The classification network that has been initially trained is validated using the potential relationship test set, and a classification network with a test accuracy greater than or equal to a preset test accuracy threshold is output as a preconfigured relationship analysis model; S103: when it is determined that there is a potential relationship, a specific representation of the potential relationship between the sparse node and the node to be mined is inferred; The specific representation of the potential relationship between the inferred sparse nodes and the nodes to be mined includes: Using the Apriori algorithm to obtain frequent item sets of sparse nodes and nodes to be mined, and obtaining part-of-speech collocation forms of the sparse nodes and nodes to be mined, the part-of-speech collocation forms including subject and subject, subject and object, and object and object; Extract frequent item sets and entity combinations in frequent item sets that are greater than the support threshold, and obtain the part-of-speech collocation form of the entity combination; If the entity combination has the same part-of-speech collocation form as the sparse node and the node to be mined, the relationship between the entities in the entity combination is used as a specific representation of the potential relationship between the sparse node and the node to be mined; If the part-of-speech collocation forms of the entity combination are inconsistent with the sparse nodes and the nodes to be mined, the relationship between the entities possessed by the entity combination will not be used as a specific representation of the potential relationship between the sparse nodes and the nodes to be mined; S104: Create and add new relationship edges for the basic medical and nursing knowledge graph according to the specific representation of the potential relationship to form an optimized medical and nursing knowledge graph, import the optimized medical and nursing knowledge graph into a preset graph database, and obtain a medical and nursing knowledge graph database.

2. According to the method for constructing a medical and nursing knowledge graph database based on the information innovation system according to claim 1, it is characterized in that: Before associating the target entity, include: Search multiple target entities according to a preset regular expression rule, and mark the searched multiple target entities as suspected identical entities; Obtain the attributes and attribute values ​​of each suspected identical entity in the corpus data after word segmentation processing, and extract attribute values ​​with the same attributes; The Word2Vec model is used to convert the attribute values ​​of the same attribute into attribute vectors, and the difference coefficient of the attribute vectors of every two suspected identical entities is measured; the specific calculation formula is as follows: ; Where: is the difference coefficient, is the attribute value of the i-th identical attribute of a suspected identical entity, is the attribute value of the i-th identical attribute of another suspected identical entity, and N is the total number of identical attributes; Two suspected identical entities whose difference coefficient is greater than or equal to a preset difference coefficient threshold are identified as the same target entity, and the entities are merged to obtain a fused target entity; Two suspected identical entities whose difference coefficient is less than a preset difference coefficient threshold are identified as non-identical target entities and there is no need to merge the entities.

3. According to the method for constructing a medical and nursing knowledge graph database based on the information innovation system according to claim 2, it is characterized in that: The step of selecting sparse nodes according to sparsity includes: The sparsity Compare with a preset sparsity threshold; If the sparsity If it is greater than or equal to the preset sparsity threshold, the corresponding original node is determined to be a sparse node; If the sparsity is small If the value of the node is less than the preset sparsity threshold, the corresponding original node is determined to be a non-sparse node.

4. According to claim 3, a method for constructing a medical and nursing knowledge graph database based on an information innovation system is characterized in that: The logic for generating the annotated labels in the training data of the historical recognition potential relationship is as follows: Obtain each first intermediate node and second intermediate node in the M second node pairs; Calculating the cosine similarity coefficients of the first intermediate node and the second intermediate node, and accumulating the cosine similarity coefficients to obtain the accumulated cosine similarity coefficients; The specific calculation formula is: ; Where: is the cumulative cosine similarity coefficient, is the cosine similarity coefficient, is the first intermediate node in the jth second node pair, is the first intermediate node in the jth second node pair; wherein, The calculation formula is ; Where: is the norm; The cumulative cosine similarity coefficient is compared with the preset cumulative cosine similarity coefficient threshold. If the cumulative cosine similarity coefficient is greater than or equal to the preset cumulative cosine similarity coefficient threshold, it is determined that there is a potential relationship between the sparse nodes in the corresponding first node pair and the nodes to be mined. Conversely, if the cumulative cosine similarity coefficient is less than the preset cumulative cosine similarity coefficient threshold, it is determined that there is no potential relationship between the sparse nodes in the corresponding first node pair and the nodes to be mined.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, it implements the method for constructing a medical and nursing knowledge graph database based on the information and innovation system as described in any one of claims 1 to 4.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed, implements the method for constructing a medical and nursing knowledge graph database based on an information and communication system as described in any one of claims 1 to 4.

Images