Government affair knowledge graph construction method based on data fusion driving

Abstract

The invention relates to a data fusion driving-based government affair knowledge graph construction method, which comprises the following steps of: firstly, proposing an original data preprocessing framework in the government affair field, researching a multi-modal data fusion model oriented to government affair convenience service, proposing an original data mapping rule and an attribute fusion rule in the government affair field, and constructing a data fusion model; the ViT model is utilized to perform feature extraction of visual modals of part of government affair data, multi-modal data alignment is realized, and standardized high-quality multi-modal government affair field data is obtained. And then, designing a government affair field ontology model based on the government affair convenience service scene and the standardized high-quality multi-mode government affair field data features, and realizing construction and storage of the government affair field ontology by adopting protege. And finally, designing a data mapping rule, and realizing construction and storage of the multi-modal government affair knowledge graph by adopting a graph database Neo4j. According to the construction of the multi-modal government affair knowledge graph, multi-modal and semi-structured original data in the government affair field are effectively integrated, and important knowledge base support is provided for government affair convenience services.

Description

Technical Field

[0001] The present invention relates to the technical field of government services for the convenience of the people, and in particular to a method for constructing a government knowledge graph driven by data fusion. Background Art

[0002] As government information disclosure channels shift from traditional official media paper documents to digital platforms, ordinary people can more conveniently access massive amounts of e-government documents through various government portals. However, faced with such a huge amount of e-government information, both government staff and ordinary people face the problem of difficulty in quickly and accurately filtering out the required information in actual operations. The fragmentation of information makes it complicated and time-consuming to find specific policy provisions or service guidelines, and work efficiency is low. Summary of the invention

[0003] In view of this, the purpose of the present invention is to provide a method for constructing a government knowledge graph driven by data fusion. First, by combining a multimodal data fusion model with structured data mapping rules and attribute fusion rules, and extracting some original data visual modal features based on the ViT model, standardized high-quality multimodal government data is obtained. Based on the needs of government service scenarios, combined with real data, a government domain ontology model is designed. After completing data preprocessing and the construction of the government domain ontology, data mapping rules are designed to achieve the mapping of the ontology to the government knowledge graph, and the graph database Neo4j is used to realize the construction and storage of multimodal knowledge graphs, providing data support for the establishment of character portraits and the design of intelligent questions and answers in the downstream.

[0004] To achieve the above purpose, the present invention adopts the following technical solution: a method for constructing a government knowledge graph based on data fusion drive, comprising the following steps: Step S1, obtaining original data in the government affairs field, combining the multimodal data fusion model with structured data mapping rules and attribute fusion rules, and extracting some visual modal features of the original data based on the ViT model, realizing the alignment and attribute fusion of multimodal data, and obtaining standardized high-quality multimodal government affairs field data; Step S2: Design a government affairs domain ontology model based on the needs of government affairs service scenarios and standardized multimodal government affairs domain data features, and adopt Realize the construction and storage of government affairs ontology; Step S3: After completing data preprocessing and building the government domain ontology, design the mapping rules between the ontology and the graph data, realize the mapping of the ontology to the government domain knowledge graph, and use the graph database Neo4j to realize the construction and storage of the multimodal knowledge graph.

[0005] In a preferred embodiment, in step S1, by collecting original data in the government field of different structures and different modes in the government website, and performing appropriate data cleaning, fusion processing and data labeling on the original data with data redundancy, complexity, dispersion and fragmentation problems, standardized high-quality multi-modal government field data is formed, which specifically includes the following steps: Obtaining the original data in the government affairs field and mapping it into tuple-structured government affairs field information data specifically includes the following steps: Step S1A1: According to service requirements, an automated script is customized based on the Scarpy framework to obtain original data in the government field on the government website, and obtain semi-structured text data and visual modality data; collect government information disclosure and public service information data; Step S1A2: Clean the semi-structured government affairs original data obtained in step S1A1 by writing an automated script and map the obtained structured government affairs original data into tuple data. ;in, For entities in the government sector, Represents entity class E i The first attribute of Represents entity class E i The second attribute, m is the entity class E i The number of attributes, , N represents the number of attributes of the entity class.

[0006] In a preferred embodiment, entity alignment and attribute fusion are performed on tuple-structured government affairs information data to obtain standardized high-quality multimodal government affairs data. Assuming that two government affairs entity instances of the government affairs entity class are given and , m is the number of attributes of the entity; specifically includes the following steps: Step S1B1, divide the tuple structured government affairs information data into three different data structures: numerical, textual and image. According to different entity attribute categories, select different attribute feature quantification methods, and then assign different calculation weights to different entity attributes according to their entity identification capabilities and the difficulty of fiction, so as to calculate the similarity between different entities and determine whether the entities are aligned. The specific calculation process is as follows: when and When they are of the same attribute type and are numerical, their numerical values ​​are directly used as quantitative features, and the distance formula is used to calculate the similarity between attributes. The specific calculation uses the following formula: Among them, when represents the Kth attribute of all entities of this type, represents the maximum value of the Kth attribute of all entities of this type, Represents the first The minimum value of an attribute; when and When they are of the same text type, their attribute values ​​are often short or long descriptive texts. For shorter texts, the Levenshtein Distance can be used to quantify the attribute similarity. The specific calculation uses the following formula: in yes Before i characters and Before j The Levenshtein distance between characters, where i and j Respectively and A prefix length of max(i,j) and min(i,j) Respectively i and j The maximum and minimum values ​​between ; For long texts, use Jieba word segmentation tool to implement text segmentation, divide the text into independent word sets, get the union of two text word sets, calculate the frequency of each word to get the TF-IDF vector, and use the following formula for specific calculation: in express and The cosine similarity of two n-dimensional TF-IDF vectors A1 and A2, A 1K Representation attributes The size of the kth dimension of the n-dimensional TF-IDF vector, A 2K Representation attributes The size of the kth dimension of the n-dimensional TF-IDF vector; For attributes The square of the k-th dimension of the n-dimensional TF-IDF vector, Representation attributes The square of the k-th dimension of the n-dimensional TF-IDF vector; The value range is [0, 1], the larger the value, the more entities e 1 and e 2Properties and The higher the similarity; when and If they are of the same image type, the image is segmented first, and then the ViT model is used to extract the image and The feature representation and , and finally calculate the image and The cosine similarity of ; the specific calculation formula is as follows: Represents image attributes and The cosine similarity of Representing Entities e 1 About Image Attributes The eigenvector of i Dimension size, Representing Entities e 1 About Image Attributes The eigenvector of i Dimension size; Indicates that the entity is modeled through the ViT model e j Image attributes Extract visual features to get entities e j About Image Attributes The eigenvector of ; After calculating the entity e 1 and e 2 After calculating the similarity between the attributes of the entities, it is necessary to assign corresponding weights W according to the identification ability of different attributes for entities and the difficulty of attribute fiction, calculate the final entity similarity, and determine whether the entities are aligned; the specific calculation formula is as follows: in, W i Indicates i The weight of the attribute, Sim i Indicates i The similarity between attributes; Step S1B2: For two entities from different data sets pointing to the same entity, or an entity in the same data set having multiple virtual identities, that is, when the entity and entities Both refer to the same entity e 0 According to the knowledge completeness rule, when m>n, and Included e 2 All the attributes in e 0 Add attributes to , redundancy removal rule: when ,and , are correct, but if there is inconsistency in the description, e 0 Add or Attributes, but not added at the same time, attribute completion rules are: when there is ,and , When both refer to the same attribute class and the descriptions are correct but incomplete, e 0 Add a value to The property and consistency principle is: when ,and , Pointing to the same attribute class, but there are some errors, the wrong part will be deleted and only e 0 Add a value to , The attributes of the correct combination of parts are combined, and the attributes of the entities belonging to the same object discovered in step S1B1 are fused and the data is labeled to obtain standardized high-quality multimodal government affairs data.

[0007] In a preferred embodiment, in step S2, based on the needs of the government service scenario, combined with the multimodal government data obtained in step S1, the core concepts and important terms in the government service field are integrated and summarized, and the government ontology is designed and implemented, which specifically includes the following steps: Step S2A1, define the ontology class using a top-down approach, and build the ontology models of personnel information PersonnelInformation, policy documents PolicyDocuments, document materials DocumentaryMaterial, departmental institutions SectoralInstitutions, government matters GovernmentMatters and the relationships between them for the government service scenario; complete the government domain ontology design; Step S2A2: Based on the government domain ontology obtained in step S2A1, define its hierarchical structure, attributes and attribute constraints respectively; Step S2A3: Based on the basic framework of the government affairs domain ontology obtained in steps S2A1 and S2A2, define the association relationships and constraints between ontologies; Step S2A4: Use the government domain ontology obtained in steps S2A1, S2A2 and S2A3 to The ontology editor assists in creating ontology instances in the government affairs field.

[0008] In a preferred embodiment, in step S3, the mapping rules from the government domain ontology to the knowledge graph are designed and the data is stored to realize the generation of the government knowledge graph, which specifically includes the following steps: Step S3A1, mapping the multimodal government affairs domain data obtained in step S1 according to the government affairs domain ontology model obtained in step S2, mapping specific government affairs domain data into the corresponding subdomain, and realizing the mapping of the ontology to the government affairs knowledge graph; Step S3A2: Use the Neo4j graph database for storage, visualize the entities and their relationships in the form of a graph, and realize the generation of a knowledge graph.

[0009] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a method for constructing a government knowledge graph driven by data fusion. First, by combining a multimodal data fusion model with structured data mapping rules and attribute fusion rules, and extracting some of the original data visual modal features based on the ViT model, standardized, high-quality multimodal government data is obtained. Based on the needs of government service scenarios and combined with real data, an ontology model for the government affairs field is designed. After completing data preprocessing and the construction of the government affairs field ontology, data mapping rules are designed to realize the mapping of the ontology to the government affairs knowledge graph, and the graph database Neo4j is used to realize the construction and storage of a multimodal knowledge graph, providing data support for the establishment of character portraits and the design of intelligent questions and answers in the downstream. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a schematic diagram of the framework of the method of the present invention.

[0011] Figure 2Schematic diagram of the multimodal entity alignment framework of the method of the present invention. DETAILED DESCRIPTION

[0012] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0013] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0014] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application; as used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or their combinations.

[0015] The present invention provides a method for constructing a government affairs knowledge graph based on data fusion drive. The overall framework of the invention method is as follows: Figure 1-2 As shown. First, based on the Scarpy framework, an automated script is customized to obtain the original data in the government field on government websites, mainly the portal websites of the people's government. Through automated script writing, the semi-structured original data in the government field is cleaned to make it structured data. The obtained structured data is mapped to tuple data and divided into three different data structures: numerical, textual and image. According to different entity attribute categories, different attribute feature quantification methods are selected. Then, different entity attributes are given different calculation weights according to their entity identification capabilities and the difficulty of fiction. In this way, the similarity between different entities is calculated to determine whether the entities are aligned. For two entities pointing to the same entity from different data sets, or an entity in the same data set with multiple virtual identities, attribute fusion and data annotation are performed to obtain standardized high-quality multimodal government field data. Then, based on the needs of government service scenarios, combined with real data, a top-down approach is used to complete the design of the government field ontology, further define the hierarchical structure, attributes and attribute constraints of each field ontology, design the association relationship between government field ontologies, and use Create an ontology instance in the government domain. Finally, design the mapping rules between ontology and graph data, realize the mapping of ontology to government knowledge graph, map specific government domain data into the corresponding subdomain, and use the graph database Neo4j to realize the construction and storage of multimodal government knowledge graph.

[0016] The present invention provides a method for constructing a government knowledge graph driven by data fusion. First, by combining a multimodal data fusion model with structured data mapping rules and attribute fusion rules, and extracting some of the original data visual modal features based on the ViT model, standardized, high-quality multimodal government data is obtained. Based on the needs of government service scenarios for the convenience of the people and combined with real data, a government domain ontology model is designed. After completing data preprocessing and the construction of the government domain ontology, data mapping rules are designed to achieve the mapping of the ontology to the government knowledge graph, and the graph database Neo4j is used to realize the construction and storage of multimodal knowledge graphs, so as to provide data support for the establishment of character portraits and the design of intelligent questions and answers downstream. The specific steps are as follows: Step S1, obtaining original data in the government affairs field, combining the multimodal data fusion model with structured data mapping rules and attribute fusion rules, and extracting some visual modal features of the original data based on the ViT model, realizing the alignment and attribute fusion of multimodal data, and obtaining standardized high-quality multimodal government affairs field data; Step S2: Design a government affairs domain ontology model based on the needs of government affairs service scenarios and standardized multimodal government affairs domain data features, and adopt Realize the construction and storage of government affairs ontology; Step S3: After completing data preprocessing and building the government domain ontology, design the mapping rules between the ontology and the graph data, realize the mapping of the ontology to the government domain knowledge graph, and use the graph database Neo4j to realize the construction and storage of the multimodal knowledge graph.

[0017] In step S1, the original data of government affairs with different structures and modalities in government affairs websites are collected, and appropriate data cleaning, fusion processing and data labeling are performed on the original data with problems such as data redundancy, complexity, dispersion and fragmentation, so as to form standardized high-quality multimodal government affairs data, which specifically includes the following steps: Obtaining the original data in the government affairs field and mapping it into tuple-structured government affairs field information data specifically includes the following steps: Step S1A1: According to service requirements, an automated script is customized based on the Scarpy framework to obtain original data in the government field on government websites, mainly the government portal website, to obtain semi-structured text data and visual modality data. The data collected mainly include government information (such as laws and regulations, notices, etc.), public service information (such as education enrollment, services for the disabled, etc.), etc.

[0018] Step S1A2: Clean the semi-structured government affairs original data obtained in step S1A1 by writing an automated script and map the obtained structured government affairs original data into tuple data. .in, For entities in the government sector, Represents entity class E i The first attribute of Represents entity class E i The second attribute, m is the entity class E i The number of attributes, , N represents the number of attributes of the entity class.

[0019] The tuple-structured government information data is aligned with entities and attributes to obtain standardized high-quality multimodal government data. Assume that two government entity instances of the government entity class are given. and , m is the number of attributes of the entity. Specifically, the following steps are included: Step S1B1, divide the tuple structure government affairs information data into three different data structures: numerical type, text type and image type. According to different entity attribute categories, select different attribute feature quantification methods, and then assign different calculation weights to different entity attributes according to their entity identification ability and the difficulty of fiction, so as to calculate the similarity between different entities and determine whether the entities are aligned. The specific calculation process is as follows: when and When they are of the same attribute type and are numerical, their numerical values ​​can be directly used as quantitative features, and the distance formula can be used to calculate the similarity between attributes. The specific calculation uses the following formula: Among them, when represents the Kth attribute of all entities of this type, represents the maximum value of the Kth attribute of all entities of this type, Represents the first The minimum value of an attribute; when and When they are of the same text type, their attribute values ​​are often short or long descriptive texts. For shorter texts, the Levenshtein distance can be used to quantify the attribute similarity. The specific calculation uses the following formula: in yes Before i characters and Before j The Levenshtein distance between characters, where i and jRespectively and A prefix length of max(i,j) and min(i,j) Respectively i and j The maximum and minimum values ​​between .

[0020] For longer texts, use Jieba word segmentation tool to implement text segmentation, divide the text into independent word sets, get the union of two text word sets, calculate the frequency of each word to get the TF-IDF vector, and then calculate the cosine similarity of the two TF-IDF vectors. The specific calculation uses the following formula: in express and The cosine similarity of two n-dimensional TF-IDF vectors A1 and A2, A 1K Representation attributes The size of the kth dimension of the n-dimensional TF-IDF vector, A 2K Representation attributes The size of the kth dimension of the n-dimensional TF-IDF vector; For attributes The square of the k-th dimension of the n-dimensional TF-IDF vector, Representation attributes The square of the k-th dimension of the n-dimensional TF-IDF vector; The value range is [0, 1], the larger the value, the more entities e 1 and e 2 Properties and The higher the similarity.

[0021] when and If they are of the same image type, the image is segmented first, and then the ViT model is used to extract the image and The feature representation and , and finally calculate the image and The cosine similarity of ; the specific calculation formula is as follows: Represents image attributes and The cosine similarity of Representing Entities e 1 About Image Attributes The eigenvector of i Dimension size, Representing Entities e 1 About Image Attributes The eigenvector of i Dimension size; Indicates that the entity is modeled through the ViT model e j Image attributes Extract visual features to get entities e j About Image Attributes The eigenvector of .

[0022] After calculating the entity e 1 and e 2 After calculating the similarity between the attributes of the entities, it is necessary to assign corresponding weights W according to the identification ability of different attributes for entities and the difficulty of attribute fiction, calculate the final entity similarity, and determine whether the entities are aligned; the specific calculation formula is as follows: in, W i Indicates i The weight of the attribute, Sim i Indicates i The similarity between attributes; Step S1B2: For two entities from different data sets pointing to the same entity, or an entity in the same data set having multiple virtual identities, that is, when the entity and entities Both refer to the same entity e 0 According to the knowledge completeness rule, when m>n, and Included e 2 All the attributes in e 0 Add attributes to , redundancy removal rule: when ,and , are correct, but if there is inconsistency in the description, e 0 Add or Attributes, but not added at the same time, attribute completion rules are: when there is ,and , When both refer to the same attribute class and the descriptions are correct but incomplete, e 0 Add a value to The property and consistency principle is: when ,and , Pointing to the same attribute class, but there are some errors, the wrong part will be deleted and only e 0 Add a value to , The attributes of the correct combination of parts are combined, and the attributes of the entities belonging to the same object discovered in step S1B1 are fused and the data is labeled to obtain standardized high-quality multimodal government affairs data.

[0023] In step S2, based on the needs of government service scenarios, combined with the multimodal government data obtained in step S1, the core concepts and important terms in the government service field are integrated and summarized, and the government ontology is designed and implemented, which specifically includes the following steps: Step S2A1. Define the ontology class using a top-down approach. For the government service scenario, construct the ontology models of Personnel Information, Policy Documents, Documentary Material, Sectoral Institutions, and Government Matters, as well as the relationships between them. Complete the ontology design in the government domain.

[0024] Step S2A2: Based on the government domain ontology obtained in step S2A1, define its hierarchical structure, attributes and attribute constraints respectively.

[0025] Step S2A3: Based on the basic framework of the government affairs domain ontology obtained in steps S2A1 and S2A2, define the association relationships and constraints between ontologies.

[0026] Step S2A4: Use the government domain ontology obtained in steps S2A1, S2A2 and S2A3 to The ontology editor assists in creating ontology instances in the government affairs field.

[0027] In step S3, the mapping rules from the government domain ontology to the knowledge graph are designed and the data is stored to realize the generation of the government knowledge graph, which specifically includes the following steps: Step S3A1: Map the multimodal government affairs domain data obtained in step S1 according to the government affairs domain ontology model obtained in step S2, and map the specific government affairs domain data into the corresponding subdomain to realize the mapping of the ontology to the government affairs knowledge graph.

[0028] Step S3A2: Use the Neo4j graph database for storage, visualize the entities and their relationships in the form of a graph, and realize the generation of a knowledge graph.

[0029] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A method for constructing a government affairs knowledge graph based on data fusion, characterized in that: The following steps are involved: Step S1, obtaining original data in the government affairs field, combining the multimodal data fusion model with structured data mapping rules and attribute fusion rules, and extracting some visual modal features of the original data based on the ViT model, realizing the alignment and attribute fusion of multimodal data, and obtaining standardized high-quality multimodal government affairs field data; Step S2: Design a government affairs domain ontology model based on the needs of government affairs service scenarios and standardized multimodal government affairs domain data features, and adopt Realize the construction and storage of government affairs ontology; Step S3: After completing data preprocessing and building the government domain ontology, design the mapping rules between the ontology and the graph data, realize the mapping of the ontology to the government domain knowledge graph, and use the graph database Neo4j to realize the construction and storage of the multimodal knowledge graph.

2. According to the method for constructing a government affairs knowledge graph based on data fusion drive according to claim 1, it is characterized in that: In step S1, the original data of government affairs with different structures and modalities in government affairs websites are collected, and appropriate data cleaning, fusion processing and data labeling are performed on the original data with data redundancy, complexity, dispersion and fragmentation problems to form standardized high-quality multimodal government affairs data, which specifically includes the following steps: Obtaining the original data in the government affairs field and mapping it into tuple-structured government affairs field information data specifically includes the following steps: Step S1A1: According to service requirements, an automated script is customized based on the Scarpy framework to obtain original data in the government field on the government website, and obtain semi-structured text data and visual modality data; collect government information disclosure and public service information data; Step S1A2: Clean the semi-structured government affairs original data obtained in step S1A1 by writing an automated script and map the obtained structured government affairs original data into tuple data. ;in, For entities in the government sector, Represents entity class E i The first attribute of Represents entity class E i The second attribute, m is the entity class E i The number of attributes, , N represents the number of attributes of the entity class.

3. According to the data fusion driven government affairs knowledge graph construction method of claim 2, it is characterized in that: Perform entity alignment and attribute fusion on tuple-structured government affairs information data to obtain standardized, high-quality, multimodal government affairs data; Assume that two government entity instances of the government domain entity class are given and , m is the number of attributes of the entity; specifically includes the following steps: Step S1B1, divide the tuple structured government affairs information data into three different data structures: numerical, textual and image. According to different entity attribute categories, select different attribute feature quantification methods, and then assign different calculation weights to different entity attributes according to their entity identification capabilities and the difficulty of fiction, so as to calculate the similarity between different entities and determine whether the entities are aligned. The specific calculation process is as follows: when and When they are of the same attribute type and are numerical, their numerical values ​​are directly used as quantitative features, and the distance formula is used to calculate the similarity between the attributes. The specific calculation uses the following formula: Among them, when represents the Kth attribute of all entities of this type, represents the maximum value of the Kth attribute of all entities of this type, Represents the first The minimum value of an attribute; when and When they are of the same text type, their attribute values ​​are often short or long descriptive texts. For shorter texts, the Levenshtein Distance can be used to quantify the attribute similarity. The specific calculation uses the following formula: in yes Before i characters and Before j The Levenshtein distance between characters, where i and j Respectively and A prefix length of max(i,j) and min(i,j) Respectively i and j The maximum and minimum values ​​between ; For long texts, use Jieba word segmentation tool to implement text segmentation, divide the text into independent word sets, get the union of two text word sets, calculate the frequency of each word to get the TF-IDF vector, and use the following formula for specific calculation: in express and The cosine similarity of two n-dimensional TF-IDF vectors A1 and A2, A 1K Representation attributes The size of the kth dimension of the n-dimensional TF-IDF vector, A 2K Representation attributes The size of the kth dimension of the n-dimensional TF-IDF vector; For attributes The square of the k-th dimension of the n-dimensional TF-IDF vector, Representation attributes The square of the k-th dimension of the n-dimensional TF-IDF vector; The value range is [0, 1], the larger the value, the more entities e 1 and e 2 Properties and The higher the similarity; when and If they are of the same image type, the image is segmented first, and then the ViT model is used to extract the image and The feature representation and , and finally calculate the image and The cosine similarity of ; the specific calculation formula is as follows: Represents image attributes and The cosine similarity of Representing Entities e 1 About Image Attributes The eigenvector of i Dimension size, Representing Entities e 1 About Image Attributes The eigenvector of i Dimension size; Indicates that the entity is modeled through the ViT model e j Image attributes Extract visual features to get entities e j About Image Attributes The eigenvector of ; After calculating the entity e 1 and e 2 After calculating the similarity between the attributes of the entities, it is necessary to assign corresponding weights W according to the identification ability of different attributes for entities and the difficulty of attribute fiction, calculate the final entity similarity, and determine whether the entities are aligned; the specific calculation formula is as follows: in, W i Indicates i The weight of the attribute, Sim i Indicates i The similarity between attributes; Step S1B2: For two entities from different data sets pointing to the same entity, or an entity in the same data set having multiple virtual identities, that is, when the entity and entities Both refer to the same entity e 0 According to the knowledge completeness rule, when m>n, and Included e 2 All the attributes in e 0 Add attributes to , redundancy removal rule: when ,and , are correct, but if there is inconsistency in the description, e 0 Add or Attributes, but not added at the same time, attribute completion rules are: when there is ,and , When both refer to the same attribute class and the descriptions are correct but incomplete, e 0 Add a value to The property and consistency principle is: when ,and , Pointing to the same attribute class, but there are some errors, the wrong part will be deleted and only e 0 Add a value to , The attributes of the correct combination of parts are combined, and the attributes of the entities belonging to the same object discovered in step S1B1 are fused and the data is labeled to obtain standardized high-quality multimodal government affairs data.

4. According to the method for constructing a government affairs knowledge graph based on data fusion drive according to claim 1, it is characterized in that: In step S2, based on the needs of government service scenarios, combined with the multimodal government data obtained in step S1, the core concepts and important terms in the government service field are integrated and summarized, and the government ontology is designed and implemented, which specifically includes the following steps: Step S2A1, define the ontology class using a top-down approach, and build the ontology models of personnel information PersonnelInformation, policy documents PolicyDocuments, document materials DocumentaryMaterial, departmental institutions SectoralInstitutions, government matters GovernmentMatters and the relationships between them for the government service scenario; complete the government domain ontology design; Step S2A2: Based on the government domain ontology obtained in step S2A1, define its hierarchical structure, attributes and attribute constraints respectively; Step S2A3: Based on the basic framework of the government affairs domain ontology obtained in steps S2A1 and S2A2, define the association relationships and constraints between ontologies; Step S2A4: Use the government domain ontology obtained in steps S2A1, S2A2 and S2A3 to The ontology editor assists in creating ontology instances in the government affairs field.

5. According to the data fusion driven government affairs knowledge graph construction method of claim 1, it is characterized in that: In step S3, the mapping rules from the government domain ontology to the knowledge graph are designed and the data is stored to realize the generation of the government knowledge graph, which specifically includes the following steps: Step S3A1, mapping the multimodal government affairs domain data obtained in step S1 according to the government affairs domain ontology model obtained in step S2, mapping specific government affairs domain data into the corresponding subdomain, and realizing the mapping of the ontology to the government affairs knowledge graph; Step S3A2: Use the Neo4j graph database for storage, visualize the entities and their relationships in the form of a graph, and realize the generation of a knowledge graph.

Images