A dynamic knowledge graph construction method for three-dimensional security situation analysis

Abstract

The application relates to the technical field of knowledge graph construction, in particular to a dynamic knowledge graph construction method for three-dimensional security situation analysis, which comprises the following steps: extracting entities, entity relationships, attributes and time from texts of multiple data sources, constructing multiple knowledge graphs, and determining the first intersection degree of any two entities; determining the belonging node and the same-level node of each node, and generating the inherited attributes of each node; selecting one node from any two knowledge graphs as a target node; calculating a correction factor, determining the second intersection degree and the structural approximation degree of the two target nodes; obtaining the correlation coefficient of the two target nodes, evaluating the entity corresponding to the target node, performing entity alignment, and constructing a dynamic knowledge graph. The application improves the accuracy of entity alignment between different knowledge graphs.

Description

Technical Field

[0001] This application relates to the field of knowledge graph construction technology, specifically to a method for constructing a dynamic knowledge graph for three-dimensional security situation analysis. Background Technology

[0002] In a broad sense, knowledge graph technology encompasses the entire process of knowledge formation, including extraction, fusion, evaluation, reasoning, and storage of knowledge resources. Finally, a visualized knowledge graph is constructed. Because the relationships between entities in a knowledge graph change over time, a dynamically evolving knowledge graph can dynamically perceive the security status of knowledge, thereby reducing the risk of reasoning about knowledge. Entity alignment in dynamic knowledge graphs is an important research direction in the field of knowledge graphs.

[0003] Traditional entity alignment algorithms determine whether entities belong to the same entity by matching their names with strings. This approach is prone to misjudgment, incorrectly aligning entities with different concepts. Furthermore, it ignores the broad or narrow relationships between the conceptual representations of different entities, fails to consider the inheritance characteristics of attributes between entities, and the structural characteristics of entities in different knowledge graphs. As a result, the alignment results are less accurate when aligning entities across different knowledge graphs, affecting the effectiveness of dynamic knowledge graph applications. Summary of the Invention

[0004] To address the aforementioned technical issues, a method for constructing a dynamic knowledge graph for three-dimensional security situation analysis is provided to resolve existing problems.

[0005] The solution to the technical problem presented in this application is a method for constructing a dynamic knowledge graph for three-dimensional security situation analysis, comprising the following steps:

[0006] Entities, relationships between entities, attributes, and time are extracted from text from multiple data sources to construct multiple knowledge graphs. The attributes corresponding to each node in each knowledge graph are recorded as its own attributes.

[0007] For all entities corresponding to all nodes in all knowledge graphs, analyze the differences between different self-attributes of any two entities, and determine the first degree of intersection between any two entities by combining the distribution of each self-attribute of the entity in the corresponding text.

[0008] Based on the connection relationships between nodes in the same knowledge graph and the distribution of the first degree of intersection between connected nodes, the parent node and sibling nodes of each node are determined; based on the self-attributes of the parent node, the inherited attributes of each node are generated.

[0009] Select one node from any two knowledge graphs as a target node; calculate a correction factor through the difference between the inheritance attribute of each target node and the attribute of the other target node, and determine the second intersection degree of the two target nodes in combination with the first intersection degree;

[0010] Determine the structural approximation degree of the two target nodes through the similarity of the target nodes and the similarity of the sibling nodes represented by the sibling nodes, and the difference in the number of the sibling nodes;

[0011] Based on the second intersection degree and the structural approximation degree, determine the correlation coefficient of the two target nodes, evaluate the entities corresponding to the target nodes, and perform entity alignment to construct a dynamic knowledge graph.

[0012] Preferably, the determination of the first intersection degree of the two entities comprises:

[0013] Analyze the distance relationship of each entity and its corresponding self-attribute in the text, and the frequency of the text containing each entity and its self-attribute, and calculate the semantic influence degree;

[0014] Select the entity corresponding to the minimum number of self-attributes in the two entities as the target entity;

[0015] Calculate the minimum distance between the self-attributes corresponding to the target entity and all self-attributes corresponding to the other entity in the two entities;

[0016] The first intersection degree is the sum of the ratio of the semantic influence degree of all self-attributes corresponding to the target entity in the two entities to the minimum distance.

[0017] Preferably, the calculation of the semantic influence degree comprises:

[0018] All entities corresponding to the nodes in each knowledge graph are extracted from the text set;

[0019] Select the text containing each entity and its self-attribute from the text set as the target text;

[0020] Calculate the character distance between the entity and its corresponding self-attribute in each target text;

[0021] Statistical frequency of the target text in the text set;

[0022] Calculate the sum of the character distance between each entity and its corresponding self-attribute in all target texts, denoted as the relative distance;

[0023] The semantic influence degree is the ratio of the frequency to the relative distance.

[0024] Preferably, the determining the belonging node and the peer node of each node comprises:

[0025] All the remaining nodes connected with each node in each knowledge graph are recorded as connection nodes;

[0026] The first intersection degrees between each node and all the connection nodes thereof are clustered to obtain two clustering clusters;

[0027] The average value of all the first intersection degrees in each clustering cluster is calculated, and the connection node in the clustering cluster with the maximum average value is recorded as the belonging node, and vice versa, as the peer node.

[0028] Preferably, the inherited attribute of each node is an attribute that does not exist in the self attribute of each node and is inherited from the self attribute of all the corresponding belonging nodes.

[0029] Preferably, the calculating the correction factor comprises:

[0030] The corresponding all self attributes and inherited attributes of each target node are recorded as target attributes;

[0031] The reciprocal of the distance between each inherited attribute of any target node in the two target nodes and each target attribute of the other target node is recorded as a matching degree;

[0032] The maximum matching degree between each inherited attribute of the any target node and all the target attributes of the other target node in the two target nodes is selected, and the sum of the maximum matching degrees of all the inherited attributes of the any target node is taken as the correlation degree of the any target node.

[0033] The correction factor is the sum of the correlation degrees between the two target nodes.

[0034] Preferably, the second intersection degree is the product of the first intersection degree and the correction factor between the two target nodes.

[0035] Preferably, the relationship representation is represented by a quadruple between nodes in a knowledge graph as, s is the head entity, p is the relationship, o is the tail entity, and t is the time when the relationship representation is established, wherein E represents an entity set, R represents a relationship set, T represents a set of time intervals, and the further measurement process of the relationship representation between the target node and the peer node thereof is:

[0036] Based on the relationship between each node and each peer node thereof in each knowledge graph, the quadruple between each node and each peer node thereof is extracted, and the quadruple is converted into a word vector by a pre-training model.

[0037] Preferably, the determining the structural approximation of the two target nodes comprises:

[0038] Selecting a target node with the least number of sibling nodes in the two target nodes as a key node;

[0039] Selecting the maximum correlation between each word vector corresponding to the key node and all word vectors corresponding to the other target node in the two target nodes as a maximum correlation;

[0040] Calculating the cumulative sum of the maximum correlation of all word vectors corresponding to the key node;

[0041] Calculating the difference between the number of all sibling nodes corresponding to the key node and the other target node in the two target nodes as a number difference;

[0042] The structural approximation is the ratio of the cumulative sum and the number difference.

[0043] Preferably, the determining the correlation coefficient of the two target nodes comprises evaluating the entities corresponding to the target nodes.

[0044] The normalized result of the product of the structural approximation and the second intersection degree is taken as the correlation coefficient of the two target nodes.

[0045] If the correlation coefficient is greater than or equal to a preset threshold, the entities corresponding to the two target nodes are the same entity, otherwise, the entities corresponding to the two target nodes do not belong to the same entity.

[0046] The present application has at least the following beneficial effects:

[0047] This application calculates the semantic influence of each entity's attributes by analyzing the distribution of entities and their attributes in the text within each data source. This is beneficial because it considers the information representation of an entity by its attributes. Secondly, by combining the differences between the different attributes of any two entities, the application calculates the first cross-degree between them. This is beneficial because it considers the relevance of information contained in the attributes of two entities, reflecting the association between them and indicating the possibility that the two entities represent the same thing or concept. Furthermore, by using the first cross-degree between each node in the knowledge graph and its connected nodes, the application filters the connected nodes to determine the node's parent node and sibling nodes. This is beneficial because it considers the parent nodes that may belong to the same concept category as each node, reflecting the possible parent-child relationship between each node and its parent node, thus allowing each node to inherit the attributes of its parent node and acquire inherited attributes. Finally, by analyzing the differences in inherited attributes between two target nodes in any two knowledge graphs, a correction factor is calculated to determine... Determining the second cross-degree between the two target nodes further illustrates the association between the two entities, reflecting the possibility that the entities corresponding to the two target nodes represent the same thing or concept. This ensures that the association between the two target nodes considers not only the correlation of information contained in their own attributes but also the association of inherited attributes, thus more accurately reflecting the degree of association between the entities corresponding to the two target nodes. By analyzing the four-tuple relationships represented by the target node and its sibling nodes, the word vectors corresponding to the four-tuples between the two target nodes are analyzed, and the structural similarity between the two target nodes is calculated. Its beneficial effect is that it considers the structural similarity in the knowledge graph where the two target nodes are located, reflecting the possibility that the two target nodes represent the same entity. Determining the association coefficient between the two target nodes, evaluating the entities corresponding to the target nodes, and performing entity alignment, a dynamic knowledge graph is constructed. Its beneficial effect is that it uses the association coefficient to represent the possibility that the two target nodes are the same entity, while considering the inheritance of attributes between nodes and the structural differences of nodes in different knowledge graphs, improving the accuracy of entity alignment between different knowledge graphs. Attached Figure Description

[0048] The following section, in conjunction with the accompanying drawings, provides a more detailed explanation of the dynamic knowledge graph construction method for three-dimensional security situation analysis based on this application.

[0049] Figure 1 A flowchart illustrating the steps of a dynamic knowledge graph construction method for three-dimensional security situation analysis provided in this application embodiment;

[0050] Figure 2 A flowchart illustrating the steps of the method for obtaining correlation coefficients provided in this application embodiment. Detailed Implementation

[0051] In order to make the purposes, technical solutions and advantages of the present application clearer, a three-dimensional security situation analysis dynamic knowledge graph construction method is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0053] Please refer to Figure 1 which shows a step flowchart of a three-dimensional security situation analysis dynamic knowledge graph construction method according to an embodiment of the present application. The method comprises the following steps:

[0054] Step 1: Extract entities, relationships between entities, attributes and time from texts in multiple data sources, and construct multiple knowledge graphs. The attributes of each node in each knowledge graph are recorded as self-attributes.

[0055] Dynamic knowledge graph is generally composed of time attributes in the triples of static knowledge graph. Therefore, dynamic knowledge graph is usually in the form of four or five tuples.

[0056] In a static knowledge graph, entity alignment is the process of determining whether two entities in the same or different knowledge graphs point to the same object in the real world. Specifically, the same entity in different encyclopedias often has different forms of expression and concept levels, such as full name and nickname, slang and professional terms, original name and nickname, etc. Therefore, entity alignment in a dynamic knowledge graph refers to the process of finding the same entity with different descriptions or representations at different times.

[0057] Use Scrapy to obtain data sets from different data sources, pre-process data sets from different data sources, extract entities, relationships between entities, and attribute and time information of entities from texts in data sets from different data sources through GPT-3.5-turbo, and then construct knowledge graphs. Each data set from a data source corresponds to a knowledge graph, thereby obtaining multiple knowledge graphs. Each knowledge graph is composed of multiple nodes corresponding to entities. Therefore, each entity corresponds to a node, and each node corresponds to multiple attributes. All attributes corresponding to each node are recorded as self-attributes of each node.

[0058] In the embodiment, Scrapy is used to obtain data sets of two data sources of Baidu Encyclopedia and Wikipedia. Secondly, the data preprocessing process includes removing escape characters and sentences containing invalid characters; writing regular rules, and unifying the name format to standardize the data; extracting the entry name and its disambiguation name, synonyms, and storing them in MySQL for entity linking; extracting the entry attribute list from the entry information box and storing it in MySQL for importing into the graph database and constructing relationship extraction supervision data. It should be noted that the data preprocessing process and the construction of the knowledge graph are known technologies, and will not be described here.

[0059] It should be noted that for an entity of a person name, its own attributes can include "date of birth", "nationality", and "occupation"; and for an entity of a company, it can have multiple own attributes such as "founding date", "headquarters location", and "number of employees".

[0060] At this point, multiple knowledge graphs are obtained.

[0061] Step 2: For all entities corresponding to nodes in all knowledge graphs, analyze the differences between different own attributes of any two entities, and determine the first intersection degree of the two entities based on the distribution of each own attribute of the entity in the corresponding text.

[0062] Since different knowledge graphs are constructed from data sets from different data sources, entity alignment can combine related information from different sources to form a more complete data set, which helps to improve the quality and comprehensiveness of the data, and also improves the effectiveness of information retrieval. Secondly, entity alignment is also called knowledge fusion, that is, merging two knowledge graphs, which is one of the key steps in the dynamic knowledge graph construction process. How to fuse the description information of the same entity or concept from multiple sources is a problem that needs attention. For example, the description information of "Tang Sanzang", "Xuanzang", and "Jin Chanzi" is actually the information of the same entity, but it is expressed differently in different knowledge graphs. Therefore, entity alignment is needed to correspond the same entity described or represented differently.

[0063] Firstly, according to which texts the entity is extracted from, the semantic influence brought by each attribute in different texts is often different. The distance relationship between each entity and its corresponding own attributes in the corresponding text, and the frequency of each own attribute appearing in different texts are analyzed to calculate the semantic influence degree, which is specifically:

[0064] The texts in which all the entities represented by the nodes in each knowledge graph are extracted are combined to form a text set;

[0065] From the text set, select texts that contain each entity and its own attributes, denoted as target texts;

[0066] counting the frequency of occurrence of the target text in the text set;

[0067] calculating the character distance between each entity in the target text and its corresponding self attribute;

[0068] calculating the sum of the character distance between each entity and its self attribute in all target texts, denoted as the relative distance;

[0069] taking the ratio of the frequency and the relative distance as the semantic influence degree of each self attribute corresponding to each entity;

[0070] It should be noted that the smaller the relative distance, the more likely the self attribute directly describes the information of the entity, and the greater the semantic influence degree, indicating that the self attribute corresponding to the entity contains more information in the text and is more likely to represent the characteristic information of the entity; otherwise, it indicates that the self attribute corresponding to the entity contains less information in the text and is less likely to represent the characteristic information of the entity.

[0071] Secondly, by matching the self attributes of different entities and combining the semantic influence degree, a first cross degree is calculated to evaluate the correlation of the description information of different entities and reflect the possibility that an entity represents the same thing or concept, specifically:

[0072] For all entities corresponding to all nodes in the knowledge graph, taking any two entities as an example, the two entities are denoted as and , selecting the entity with the minimum number of attributes between entity and entity , denoted as the target entity; assuming that the number of attributes corresponding to entity is small, entity is denoted as the target entity;

[0073] calculating the minimum distance between each self attribute corresponding to the target entity and all self attributes corresponding to entity ;

[0074] In this embodiment, the distance is calculated by calculating the edit distance between each self attribute corresponding to the target entity and each self attribute corresponding to entity , selecting the minimum value of the edit distance between each self attribute corresponding to the target entity and all self attributes corresponding to entity ; wherein the calculation of the edit distance is a known technology and will not be described here.

[0075] calculating the minimum distance between each self attribute corresponding to the target entity The ratio of the semantic influence degree of each corresponding self attribute to the minimum distance is denoted as a relative ratio;

[0076] The target entity The sum of the relative ratios of all corresponding self attributes is taken as the target entity The first intersection degree between the entity

[0077] It should be noted that the smaller the minimum distance, the more matched the corresponding two self attributes, the more the same semantic information contained, and the greater the first intersection degree, which means that the description information of the two entities is more related, and the two entities are more likely to represent the same thing or concept. Conversely, the description information of the two entities is less related, and the two entities are less likely to represent the same thing or concept.

[0078] It should be understood that the first intersection degree of any two entities is the first intersection degree between the entities corresponding to any two nodes in the same knowledge graph, and the first intersection degree between the entities corresponding to each node in each knowledge graph and the entities corresponding to each node in another knowledge graph.

[0079] Thus, the first intersection degree of any two entities is obtained.

[0080] Step 3, based on the connection relationship of each node in the same knowledge graph and the distribution of the first intersection degree between the connected nodes, the belonging node and the same level node of each node are determined; based on the self attribute of the belonging node, the inherited attribute of each node is generated; one node is selected from any two knowledge graphs, denoted as a target node; by the difference between the inherited attribute of each target node in the two target nodes and the attribute of the other target node, a correction factor is calculated, and the first intersection degree is combined to determine the second intersection degree of the two target nodes.

[0081] Further, in each knowledge graph, each entity corresponds to a node, and there is a relationship between the entities corresponding to two nodes, which can form a quadruple. The quadruple represents the relationship between the nodes in the knowledge graph, and the quadruple includes “subject”, “predicate”, “object” and “time”. For example, (Xiaoming, lives in, Beijing, 2020-2023), each node in each knowledge graph can extract the corresponding quadruple, and the quadruple in the knowledge graph is represented as , s is the head entity, p is the relationship, o is the tail entity, and t is the time when the quadruple is established, wherein E represents the entity set, R represents the relationship set, T represents the set of time intervals, and the quadruple represents that the entity s has a relationship p with the object entity o within the time interval t.

[0082] ​Secondly, the first degree of crossover can effectively reflect whether two entities share the same attributes. However, directly using the first degree of crossover for entity alignment can easily lead to misalignment of entities with similar features, resulting in subsequent mismatches. Therefore, by analyzing the structural relationships between different entities in each knowledge graph, we obtain their parent nodes and sibling nodes, specifically:

[0083] In each knowledge graph, the first Taking the nth node as an example, each knowledge graph will be related to the nth node. All other nodes connected to a given node are denoted as connected nodes;

[0084] For the first Clustering is performed on the first cross-degree between each node and all its connected nodes to obtain two clusters;

[0085] In this embodiment, the k-means clustering algorithm is used to obtain two clusters. The k-means clustering algorithm is a well-known technology and will not be described in detail here. As other implementation methods, implementers can use other methods of existing technology, such as the DBSCAN clustering algorithm. This embodiment does not impose any special restrictions on this.

[0086] Calculate the average value of all the first cross-degrees within each cluster, and denote the connecting nodes within the cluster with the largest average value as the belonging node; otherwise, denote them as sibling nodes.

[0087] It should be noted that the first The greater the first cross-degree between a node and each of its connected nodes, the more it indicates that the... The more likely a node is to be a subclass of the connecting node or a general or narrow relationship of the same concept, for example, (mammal, is, animal), where mammal is a subclass of animal. The subclass of the entity "animal" can be inherited by its subclass "mammal", which means that the first degree of overlap of the attributes of the two entities is relatively large.

[0088] Furthermore, based on the inheritance characteristics of attributes between the subclasses and parent classes of nodes, the attributes possessed by the node can be obtained, and the first... To obtain attributes that the node itself does not possess, The inherited attributes of each node are as follows:

[0089] The first A node whose own attributes do not contain attributes corresponding to the own attributes of all its member nodes is considered as the first node. Inherited attributes of each node;

[0090] It should be noted that, for ease of understanding, we assume that the first... The attribute of the first node is inherited by the attribute of the second node , the first node has attribute A, the first node itself does not have attribute A, and the attribute A of the first node is inherited by the first node, so the attribute A is the inherited attribute of the first node; the attribute corresponding to the first node is divided into two categories, one of which is the attribute extracted from the data source, and the other is the inherited attribute between nodes.

[0091] Because the data sources of different knowledge graphs are different, the information contained in the data sources is often different, which will lead to the phenomenon that the attributes represented by the same node may be missing or missing, and further lead to the fact that the attributes corresponding to the entities in different knowledge graphs are different, that is, some may contain inherited attributes and some do not contain inherited attributes. The first intersection degree between the entities corresponding to two nodes in different knowledge graphs ignores this situation, therefore, by constructing a correction factor through the changes of the inherited attributes between the nodes in two knowledge graphs, the first intersection degree is corrected to calculate the second intersection degree, which is specific to:

[0092] Select one node from any two knowledge graphs, respectively denoted as target node and target node ;

[0093] All the attributes of each target node are denoted as target attributes, including the inherited attributes and the attributes extracted from the data source;

[0094] Calculate the reciprocal of the distance between each inherited attribute of target node and each target attribute of target node , denoted as matching degree;

[0095] In this embodiment, the reciprocal of the edit distance between each inherited attribute of target node and each target attribute of target node is calculated as the matching degree, wherein the calculation of the edit distance is a known technology and will not be described here. The target attributes of target node include the inherited attributes of the node and the attributes extracted from the data source.

[0096] Select the maximum value of the matching degree between each inherited attribute of target node and all target attributes of target node ; and the sum of the maximum values of all inherited attributes of target node is taken as the association degree of target node ;

[0097] It should be noted that the greater the matching degree, the more the inheritance attributes of the target node match the target attributes in the target node , the more the semantic information is the same, the greater the correlation degree, and the higher the information correlation between the inheritance attributes of the target node and the target attributes in the target node .

[0098] Similarly, the reciprocal of the distance between each inheritance attribute of the target node and each target attribute of the target node is calculated, denoted as the matching degree.

[0099] In this embodiment, the reciprocal of the edit distance between each inheritance attribute of the target node and each target attribute of the target node is calculated, denoted as the matching degree, wherein the calculation of the edit distance is a known technology and will not be described here. The target attributes of the target node include the attributes extracted from the data source and the inheritance attributes of the node.

[0100] The maximum matching degree between each inheritance attribute of the target node and all target attributes of the target node is selected; and the sum of the maximum matching degrees of all inheritance attributes of the target node is taken as the correlation degree of the target node .

[0101] The sum of the correlation degree of the target node and the correlation degree of the target node is taken as the correction factor between the target node and the target node .

[0102] The product of the first intersection degree and the correction factor between the target node and the target node is taken as the second intersection degree between the target node and the target node .

[0103] It should be noted that the greater the correction factor, the closer the nodes to which the two target nodes belong, the closer the actual attributes between the target nodes, and the greater the actual first intersection degree should be; otherwise, the greater the difference between the nodes to which the two target nodes belong, the greater the difference between the actual attributes between the target nodes, and the smaller the actual first intersection degree should be, and the greater the obtained second intersection degree, indicating that the description information of the two entities corresponding to the two target nodes in different knowledge graphs is more related, and the two entities are more likely to represent the same thing or concept.

[0104] At this point, the second intersection degree between the two target nodes in any two knowledge graphs is obtained.

[0105] Step 4, determining the structural approximation degree of the two target nodes through the similarity of the target nodes and their sibling nodes represented by the relationship between the two target nodes, and the difference in the number of their sibling nodes; determining the correlation coefficient of the two target nodes based on the second intersection degree and the structural approximation degree, evaluating the entities corresponding to the target nodes, and performing entity alignment to construct a dynamic knowledge graph.

[0106] Further, through the structural relationship of each target node and its sibling nodes in the knowledge graph, the structural correlation of the nodes in different knowledge graphs can be reflected, and the structural approximation degree is calculated, specifically:

[0107] Based on the relationship between each node and its sibling nodes in each knowledge graph, the relationship is represented by a quadruple, and the quadruple between each node and its sibling nodes is obtained. The quadruple is converted into a word vector by a pre-training model; wherein the number of all word vectors corresponding to each node is consistent with the number of its corresponding sibling nodes;

[0108] In the present embodiment, the pre-training model uses a Word2Vec model to obtain a word vector, wherein Word2Vec is a known technology and will not be described here. As other embodiments, the implementer can use other methods of prior art, for example, GloVe model, etc., and the present embodiment does not make special restrictions on this.

[0109] Select one node from any two knowledge graphs respectively, and mark them as target node and target node ;

[0110] Taking target node and target node as an example, the node with the least number of sibling nodes in target node and target node is marked as a key node, and it is assumed that the number of sibling nodes of target node is less than the number of sibling nodes of target node .

[0111] selecting the target node corresponding to each word vector and the target node the maximum value of the correlation degree between the corresponding all word vectors, denoted as the maximum correlation degree;

[0112] In this embodiment, the correlation degree is measured by the target node corresponding to each word vector and the target node The cosine similarity between the corresponding word vectors is measured, wherein the calculation of the cosine similarity is a known technology, and will not be described here. As other embodiments, the implementer can use other methods of prior art, for example, Jaccard similarity, and the present embodiment does not make special restrictions.

[0113] It should be noted that the greater the maximum correlation degree, the more similar the two word vectors are in semantics, and the higher the correlation.

[0114] calculating the target node the cumulative sum of the maximum correlation degree of the corresponding all word vectors;

[0115] calculating the target node the number of all sibling nodes corresponding to the target node the difference between the number of all sibling nodes corresponding to the target node

[0116] In this embodiment, the target node the number of all sibling nodes corresponding to the target node the absolute value of the difference between the number of all sibling nodes corresponding to the target node

[0117] the ratio of the cumulative sum to the number difference as the structural approximation degree between the target node and the target node .

[0118] In this embodiment, the structural approximation degree between the target node and the target node is calculated as:

[0119]

[0120] wherein, the structural approximation degree between the target node and the target node , the maximum correlation degree of each word vector corresponding to the target node , the maximum correlation degree of each word vector corresponding to the target node the number of all sibling nodes corresponding to the target node, the number of all sibling nodes corresponding to the target node the number of all sibling nodes corresponding to the target node, less than , is a preset value greater than 0, to avoid a denominator of 0, the value range of is in the embodiment, is 1, and as another implementation, the implementer can set it according to actual conditions.

[0121] It should be noted that the greater the cumulative sum, the stronger the relevance between the target node and the target node , the smaller the number difference, the closer the number of sibling nodes connected between the target node and the target node , the greater the structural approximation, the higher the similarity of the structure in the knowledge graph in which the target node is located and the structure in the knowledge graph in which the node is located, and the greater the possibility that the target node and the target node represent the same entity.

[0122] Further, based on the structural approximation and the second intersection degree, a correlation coefficient is calculated, specifically:

[0123] The normalized result of the product of the structural approximation and the second intersection degree is taken as the correlation coefficient between the target node and the target node .

[0124] In the embodiment, a sigmoid function is used for normalization processing, wherein the sigmoid function is a known technology and will not be described here. As another implementation, the implementer can use other methods of prior art, for example, a tanh function, and the embodiment does not make special restrictions thereon. The step flowchart of the method for obtaining the correlation coefficient provided in the embodiment is shown in Figure 2 .

[0125] If the correlation coefficient is greater than or equal to a preset threshold, the corresponding entities of the target node and the target node are the same entity, otherwise, the corresponding entities of the target node and the target node do not belong to the same entity.

[0126] In the embodiment, the preset threshold value is 0.8, and as another embodiment, the implementer can set it according to the actual situation.

[0127] It should be noted that the greater the correlation coefficient, the more likely the entities corresponding to the two target nodes represent the same entity; otherwise, the entities corresponding to the two target nodes represent different entities.

[0128] The entities corresponding to the nodes in different knowledge graphs are aligned, and the information in different knowledge graphs is fused to construct a dynamic knowledge graph, thereby improving the efficiency and accuracy of information sharing and integration between different knowledge graphs, and further improving the construction efficiency and accuracy of the dynamic knowledge graph.

[0129] It should be understood that, although Figure 1 The steps in the flowchart of the method are displayed in sequence according to the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps in the method can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be executed in rotation or alternation with other steps or sub-steps or stages of other steps.

[0130] The technical features of the above embodiments can be combined in any way. To make the description concise, not all possible combinations of the technical features in the above embodiments are described, but as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0131] The above-described embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it should not be understood as a limitation of the present application. It should be noted that for ordinary skilled in the art, without departing from the concept of the present application, several modifications and improvements can be made, therefore, any simple modification, equivalent change and modification of the above embodiments according to the technical essence of the present application, all within the scope of the technical solution of the present application.

Claims

1. A dynamic knowledge graph construction method for three-dimensional integrated security situation analysis, characterized in that, The method comprises the following steps: extracting entities, inter-entity relationships, attributes and time from texts of multiple data sources, constructing multiple knowledge graphs, and recording attributes corresponding to each node in each knowledge graph as self attributes; for entities corresponding to all nodes in all knowledge graphs, analyzing differences between different self attributes corresponding to any two entities, combining distribution of each self attribute of an entity in corresponding texts, and determining a first intersection degree of the any two entities; based on connection relationships of nodes in a same knowledge graph and distribution of the first intersection degree between connected nodes, determining a belonging node and a peer node of each node; and generating inherited attributes of each node based on self attributes of the belonging node; selecting one node from any two knowledge graphs as a target node; calculating a correction factor based on differences between inherited attributes of each target node and attributes of the other target node, combining the first intersection degree, and determining a second intersection degree of the two target nodes; determining a structural approximation degree of the two target nodes based on the similarity between the target nodes and their peer nodes and a difference in the number of the peer nodes; based on the second intersection degree and the structural approximation degree, determining a correlation coefficient of the two target nodes, evaluating entities corresponding to the target nodes, performing entity alignment, and constructing a dynamic knowledge graph; the determination of the first intersection degree of the any two entities comprises: analyzing distance relationships of each entity and its corresponding self attributes in texts, and frequencies of texts containing each entity and its self attributes, and calculating a semantic influence degree; selecting an entity corresponding to the minimum number of self attributes in the any two entities as a target entity; calculating a minimum distance between each self attribute corresponding to the target entity and all self attributes corresponding to the other entity in the any two entities; the first intersection degree is a sum of a ratio of the semantic influence degree of all self attributes corresponding to the target entity in the any two entities to the minimum distance; the second intersection degree is a product of the first intersection degree and the correction factor between the two target nodes; the calculation of the semantic influence degree comprises: grouping texts containing each entity and its self attributes in each knowledge graph into a text set; selecting texts containing each entity and its self attributes from the text set as target texts; calculating a character distance between an entity and its corresponding self attribute in each target text; counting frequencies of the target texts in the text set; calculating a sum of the character distance between each entity and its corresponding self attribute in all target texts as a relative distance; the semantic influence degree is a ratio of the frequency to the relative distance.

2. The method of claim 1, wherein the method further comprises: the determination of the belonging node and the peer node of each node comprises: recording all other nodes connected to each node in each knowledge graph as connection nodes; clustering the first intersection degree between each node and all connection nodes of the node to obtain two clustering clusters; An average value of all the first cross degrees in each cluster is calculated, and a connection node in the cluster with the maximum average value is recorded as a subordinate node.

3. The method of claim 1, wherein the method further comprises: The inherited attribute of each node is an attribute that does not exist in the self attribute of each node and is inherited from the self attribute of all corresponding subordinate nodes.

4. The method of claim 1, wherein the method further comprises: The calculation of the correction factor comprises: All the self attributes and inherited attributes of each target node are recorded as target attributes. The reciprocal of the distance between each inherited attribute of any target node in the two target nodes and each target attribute of the other target node is recorded as a matching degree. The maximum matching degree between each inherited attribute of the any target node and all target attributes of the other target node in the two target nodes is selected, and the sum of the maximum matching degrees of all inherited attributes of the any target node is taken as a correlation degree of the any target node. The correction factor is the sum of the correlation degrees between the two target nodes.

5. The method of claim 1, wherein, The relationship representation is represented by a quadruple between nodes in the knowledge graph as, s is a head entity, p is a relationship, o is a tail entity, and t is a time at which the relationship is established, wherein E represents a set of entities, R represents a set of relationships, and T represents a set of time intervals, and a further measurement process of the relationship representation of the target node and its sibling nodes is: Based on the relationship between each node and its peer nodes in each knowledge graph, a quadruple between each node and its peer nodes is extracted, and the quadruple is converted into a word vector by a pre-training model.

6. The method of claim 5, wherein the method further comprises: The determination of the structural approximation degree of the two target nodes comprises: The target node with the minimum number of peer nodes in the two target nodes is selected as a key node. The maximum correlation degree between each word vector corresponding to the key node in the two target nodes and all word vectors corresponding to the other target node is selected as a maximum correlation degree. The cumulative sum of the maximum correlation degrees of all word vectors corresponding to the key node is calculated. The difference between the number of all peer nodes corresponding to the key node and the number of all peer nodes corresponding to the other target node in the two target nodes is recorded as a number difference. The structural approximation degree is the ratio of the cumulative sum to the number difference.

7. The method of claim 1, wherein the method further comprises: The determination of the correlation coefficient of the two target nodes and the evaluation of the entity corresponding to the target node comprise: The normalized result of the product of the structural approximation degree and the second cross degree is taken as the correlation coefficient of the two target nodes. If the correlation coefficient is greater than or equal to a preset threshold, the entities corresponding to the two target nodes are the same entity, otherwise, the entities corresponding to the two target nodes do not belong to the same entity.

Images