A heterogeneous data fusion indexing method and system based on knowledge graph

Abstract

The present invention discloses a heterogeneous data fusion indexing method and system based on knowledge graph, comprising the following steps: analyzing entity types and relationship types in heterogeneous graphs, synchronizing and defining entity type data respectively, and defining heterogeneous entity mapping relationships; using graph adaptation to obtain entity and relationship data from two types of heterogeneous graphs respectively, and caching the data such as message middleware; constructing a view domain knowledge graph and a business domain knowledge graph; fusing the constructed view domain knowledge graph and business domain knowledge graph to construct a fusion index subgraph; performing mapping relationship analysis and mining based on the constructed fusion index subgraph to achieve disambiguation and fusion of entities in the original graph; providing a standard and unified relationship query interface for heterogeneous graphs and fusion index subgraphs; this solution constructs a data fusion index subgraph through entity mapping of the two types of knowledge graphs to achieve fusion of heterogeneous data.

Description

Technical Field

[0001] The present invention relates to the field of public security, and in particular to a heterogeneous data fusion indexing method and system based on knowledge graph. Background Art

[0002] In the public security sector, there are currently two main types of data: video and image data generated primarily by video surveillance, which is mostly unstructured and semi-structured; and data generated primarily by business systems, which is mostly structured. Currently, these two types of data are developed and constructed independently from generation, aggregation, transmission, storage, analysis, and application, and their value has not been fully realized. The fusion of these two types of data urgently needs to be addressed. Current fusion methods rely on simple unified storage and interface docking, which fail to integrate the data's fundamental value. Therefore, there is an urgent need to provide a heterogeneous data fusion system and device to address these issues. Summary of the Invention

[0003] To this end, it is necessary to provide a method for constructing knowledge graphs based on two types of data separately, and through entity mapping of the two types of knowledge graphs, construct an index sub-graph for data fusion to achieve the fusion of heterogeneous data.

[0004] To achieve the above objectives, the inventors provide a heterogeneous data fusion indexing method based on knowledge graph, which includes the following steps:

[0005] S1: Analyze the entity types and relationship types in the heterogeneous graph, synchronize and define the entity type data, and define the heterogeneous entity mapping relationship;

[0006] S2 uses the graph adaptation service to obtain entity and relationship data from the two heterogeneous graphs and cache the data in the message middleware.

[0007] S3, builds view domain knowledge graph and business domain knowledge graph;

[0008] S4, fuse the constructed view domain knowledge graph and business domain knowledge graph to build a fusion index subgraph;

[0009] S5: Based on the constructed fusion index sub-graph, mapping relationship analysis and mining are performed to achieve disambiguation and fusion of entities in the original graph;

[0010] S6 provides a standard and unified relational query interface for heterogeneous graphs and fused index subgraphs.

[0011] As a preferred embodiment of the present invention, in step S1, the heterogeneous entity mapping relationship includes entity type 1 number, entity type 2 number, entity mapping confidence and mapping time.

[0012] As a preferred embodiment of the present invention, in step S3, constructing the view domain knowledge graph includes the following steps:

[0013] Based on the video image data, structured extraction is performed, and cluster analysis and data mining analysis are used to discover view target entities and view target relationships. The view entity and view relationship data are stored in a graph database to form a view domain knowledge graph.

[0014] As a preferred embodiment of the present invention, the view target entities include real-name person files, anonymous person files, vehicle files, camera equipment, non-motor vehicle files and location areas; the view target relationships include personnel companionship, person accompaniment, vehicle accompaniment, vehicle driving and equipment file capture relationships.

[0015] As a preferred method of the present invention, in step S3, constructing a business domain knowledge graph includes the following steps: constructing a business domain knowledge graph based on the business data generated by the public security industry business system, extracting entity data and relationship data from structured or text data through big data and NLP analysis, and storing the extracted entity data and relationship data through a graph database to form a business domain knowledge graph.

[0016] As a preferred embodiment of the present invention, the entity data includes people, events, places, objects, and organizations; the relationship data includes family relationships, communication relationships, transportation, transaction transfers, and property rights relationships.

[0017] As a preferred embodiment of the present invention, in step S4, constructing the fusion index sub-graph includes the following steps:

[0018] Through AI feature comparison, rule models and graph reasoning, entity relationship mapping between heterogeneous graphs is performed, and the mapping relationship is stored in the form of sub-graphs to form a fusion index sub-graph.

[0019] As a preferred embodiment of the present invention, constructing the fusion index sub-graph further includes the following steps:

[0020] S401: Through AI feature comparison, the images of unknown entities in the view domain knowledge graph and the entity images in the business domain knowledge graph are matched for similarity. If the similarity meets the set threshold, a heterogeneous graph entity mapping relationship is established.

[0021] S402: Based on the attribute values ​​of entities and relationships in the heterogeneous graph, data matching is performed in the two types of graphs through rule configuration, and mapping relationships are established between the entities in the matched heterogeneous graphs;

[0022] S403: Combined with the spatiotemporal dimension information, first-degree relationship search and reasoning are performed on the entities in the heterogeneous graph to find a set of entities with similar relationships within a limited spatiotemporal range. These entities are then mapped to each other. The weight of the mapping relationship is set based on the number of in- and out-degrees, and the confidence level of the mapping relationship is increased through manual judgment.

[0023] S404: The heterogeneous graph entity mapping relationship formed from S401 to S403 is stored in a graph database to form a fusion index subgraph.

[0024] To achieve the above objectives, the inventors provide a heterogeneous data fusion indexing system based on knowledge graph, comprising:

[0025] The knowledge management module is used to define and configure entities and relationships in heterogeneous graphs, as well as to configure entity mapping relationships in fusion index subgraphs;

[0026] The entity mapping module provides entity mapping between heterogeneous graphs through tasks. Task types include AI construction tasks, rule model tasks, and knowledge reasoning tasks. It obtains data from heterogeneous graphs through the graph data provided by the library adaptation module, and realizes the mapping relationship between entities in heterogeneous graphs through AI feature comparison, big data stream batch computing, and graph mining calculations based on machine learning. The results of the analysis and calculation are uniformly stored through the knowledge storage module to construct a fusion index sub-graph;

[0027] The entity association analysis module is used to implement secondary association of entities within the constructed fusion index sub-graph through a path analysis algorithm, thereby achieving entity disambiguation in the original heterogeneous graph.

[0028] The graph unified query module is used to uniformly encapsulate the graph data interface, support relational query on a single graph, and relational fusion query on fusion index sub-graphs;

[0029] AI engine module, used to provide portrait comparison based on AI capabilities;

[0030] Graph mining and analysis module, used to provide graph mining and analysis, and supports graph search algorithm, shortest path algorithm, N-degree relationship tree, centrality algorithm and community discovery algorithm;

[0031] The graph database adaptation module is used to adapt the operation interface of different graph databases;

[0032] The knowledge storage module includes storing the data configured by the knowledge management module, storing the results of the entity mapping module analysis in a graph structure, and performing full-text index storage on heterogeneous entity and relationship data.

[0033] Different from the existing technology, the above technical solution has the following beneficial effects:

[0034] (1) This method and system can effectively fuse the two major types of data currently existing in the public security field through the entity similarity between heterogeneous graphs, realizing entity disambiguation of heterogeneous graphs, so that the two major types of data can fully reflect their value, which is of great significance to the public security field;

[0035] (2) This method and system provide a valuable data fusion solution based on knowledge graph, and a subgraph index-based method to achieve the fusion of heterogeneous data. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A flow chart of the method described in the specific embodiment;

[0037] Figure 2 This is a system structure diagram described in a specific implementation method. DETAILED DESCRIPTION

[0038] In order to explain the technical content, structural features, achieved objectives and effects of the technical solution in detail, the following is a detailed description in conjunction with specific embodiments and accompanying drawings.

[0039] like Figure 1 As shown, this embodiment provides a heterogeneous data fusion indexing method based on knowledge graph, including the following steps:

[0040] S1: Analyze the entity types and relationship types in the heterogeneous graph, synchronize and define the entity type data, and define the heterogeneous entity mapping relationship;

[0041] S2 uses the graph adaptation service to obtain entity and relationship data from the two heterogeneous graphs and cache the data in the message middleware.

[0042] S3, builds view domain knowledge graph and business domain knowledge graph;

[0043] S4, fuse the constructed view domain knowledge graph and business domain knowledge graph to build a fusion index subgraph;

[0044] S5: Based on the constructed fusion index sub-graph, mapping relationship analysis and mining are performed to achieve disambiguation and fusion of entities in the original graph;

[0045] S6 provides a standard and unified relational query interface for heterogeneous graphs and fused index subgraphs.

[0046] In the specific implementation process of the above embodiment, step S1 is the entity relationship definition step, which analyzes the entity types and relationship types in the heterogeneous graph, synchronizes and defines the entity type data in the system, and defines a new relationship type: heterogeneous entity mapping relationship, for example: entity type 1 number, entity type 2 number, entity mapping confidence, mapping time and other attributes;

[0047] Step S2 is the data extraction step, which uses the graph library adaptation service capability to obtain entity and relationship data from the two types of heterogeneous graphs and cache the data into the message middleware;

[0048] Step S3 is the step of constructing the view domain knowledge graph and the business domain knowledge graph, which is used to construct the index subgraph in the subsequent process. It specifically includes:

[0049] The view domain knowledge graph is constructed based on structured extraction of video image data, using cluster analysis and data mining analysis techniques to discover view target entities and view target relationships. Common view target entities include real-name person files, anonymous person files, vehicle files, camera equipment, non-motor vehicle files, and venue areas; common view target relationships include personnel companions, personnel companions, vehicle companions, vehicle driving, and equipment file snapshot relationships; view entity data and view relationship data are stored in a graph database to form a view domain knowledge graph. In this embodiment, based on the already constructed graph, the data extraction module extracts entity and relationship data from the view graph for use in the index sub-graph construction of the subsequent process.

[0050] The business domain knowledge graph is constructed based on the business data generated by the business system of the public security industry. It mainly uses big data technology and NLP analysis technology to extract entities and relationships from structured or text data. The main entities can be divided into major categories such as people, things, places, objects, and organizations, such as permanent residents, cases, hotels, banks, airplanes, trains, cars, units, houses, communities, etc.; common relationships include family relationships, communication relationships, transportation, transaction transfers, property rights relationships, etc. Such entity data and relationship data are stored in a graph database to form a business domain knowledge graph; in this embodiment, based on the already constructed graph, entity and relationship data are extracted through the data extraction module for the construction of index sub-graphs for subsequent processes.

[0051] In step S4 of the above embodiment, the constructed view domain knowledge graph and business domain knowledge graph are fused, which includes three sub-processes. AI feature comparison technology, rule model and graph reasoning technology are mainly used to realize the entity mapping relationship between heterogeneous graphs. The mapping relationship is stored in the form of a sub-graph, thereby completing the construction of the fusion index sub-graph. The specific construction process is as follows:

[0052] S401, using AI technology to build: Through AI feature comparison, the images of unknown entities in the view domain knowledge graph and the entity images in the business domain knowledge graph are matched for similarity. If the threshold is met, a heterogeneous graph entity mapping relationship is established;

[0053] S402, using rule model construction: Based on the attribute values ​​of entities and relationships in the heterogeneous graph, through rule configuration, such as ID number, mobile phone number, license plate information, etc., data matching is performed in the two types of graphs respectively, and mapping relationships are established between the matched entities in the heterogeneous graph;

[0054] S403, using graph reasoning to build: Combined with the spatial and temporal dimension information, first-degree relationship search and reasoning are performed on the entities in the heterogeneous graph to find a set of entities with similar relationships within a limited spatial and temporal range. Similar entities are then mapped and weighted based on the number of in- and out-degrees. The confidence level of the mapping relationship is then increased through manual judgment and other means.

[0055] In this embodiment, constructing a fused index subgraph includes the following steps: S404, storing the heterogeneous graph entity mapping relationships completed by the three sub-processes in steps S401 to S403 through a graph database to form a fused index subgraph.

[0056] Step S5 is to perform mapping relationship analysis and mining based on the constructed fusion index sub-graph using the graph's shortest path algorithm, N-degree relationship tree and other algorithms to achieve disambiguation and fusion of entities in the original graph;

[0057] Step S6 provides a standard and unified relational query interface for heterogeneous graphs and fused index sub-graphs, shields the compatibility and adaptation of upper-layer applications to the graph database, and improves the overall relational query and analysis efficiency.

[0058] like Figure 2 As shown, this embodiment provides a heterogeneous data fusion indexing system based on knowledge graph, including:

[0059] Knowledge management module: provides the definition and configuration functions of heterogeneous graph entities and relationships, and provides the configuration function of entity mapping relationships in fusion index sub-graphs;

[0060] Entity Mapping Module: This module provides entity mapping capabilities between heterogeneous graphs through tasks. Task types are divided into AI construction tasks, rule model tasks, and knowledge reasoning tasks. The module uses the graph data acquisition capabilities provided by the graph library adaptation module to acquire data from heterogeneous graphs. It leverages AI comparison capabilities, big data stream batch computing capabilities, and machine learning-based graph mining computing capabilities to achieve mapping relationships between entities in heterogeneous graphs. The results of the analysis and calculations are uniformly stored through the knowledge storage module to construct a fusion index sub-graph.

[0061] Entity association analysis module: Based on the constructed fusion index sub-graph, it uses algorithms such as path analysis to achieve secondary association of entities within the sub-graph, thereby achieving entity disambiguation in the original heterogeneous graph;

[0062] Graph unified query module: This module provides a unified package for common graph data interfaces, supports relational queries on a single graph, and supports relational fusion queries on fusion index sub-graphs;

[0063] AI engine module: provides AI-based portrait matching capabilities. In this embodiment, the focus is on the 1:1 interface capability.

[0064] Graph mining and analysis module: provides basic graph mining and analysis capabilities, supporting graph search algorithms, shortest path algorithms, N-degree relationship trees, centrality algorithms, community discovery algorithms, etc.

[0065] Graph library adaptation module: This module provides the ability to adapt basic operation interfaces to different graph databases. In this embodiment, there are many possible types of heterogeneous graph underlying libraries. Therefore, this module is used to adapt to mainstream graph databases and provide basic graph database operation capabilities for the unified graph query module.

[0066] Knowledge storage module: It consists of three parts. One part is to store the data configured by the knowledge management module. The second part is to store the graph structure of the results analyzed by the entity mapping module. The third part is to perform full-text index storage on heterogeneous entities and relationship data for easy retrieval.

[0067] During the operation of this system, the steps involved in the above method embodiment are also adopted.

[0068] During the implementation of the above embodiments, the present invention can effectively fuse the two major types of data currently existing in the public security field through the entity similarity between heterogeneous graphs, thereby realizing entity disambiguation of heterogeneous graphs, so that the two major types of data can fully reflect their value, which is of great significance to the public security field; in addition, the present invention proposes that heterogeneous data fusion should classify the original data to construct a knowledge graph, and then solve the entity association mapping between graphs; it also proposes that the entity mapping relationship between heterogeneous graphs be stored in a sub-graph index manner, and the structural characteristics of the graph are used to perform secondary analysis and calculation to realize entity disambiguation between heterogeneous graphs.

[0069] It should be noted that although the above embodiments have been described herein, this does not limit the scope of patent protection of the present invention. Therefore, based on the innovative concept of the present invention, changes and modifications to the embodiments described herein, or equivalent structural or equivalent process transformations made using the contents of the present invention's specification and drawings, and direct or indirect application of the above technical solutions to other related technical fields, are all included in the scope of patent protection of the present invention.

Claims

1. A heterogeneous data fusion indexing method based on knowledge graph, characterized by: The following steps are involved: S1: Analyze the entity types and relationship types in the heterogeneous graph, synchronize and define the entity type data, and define the heterogeneous entity mapping relationship; S2 uses the graph adaptation service to obtain entity and relationship data from the two heterogeneous graphs and cache the data in the message middleware. S3, builds view domain knowledge graph and business domain knowledge graph; S4, fuse the constructed view domain knowledge graph and business domain knowledge graph to build a fusion index subgraph; S5: Based on the constructed fusion index sub-graph, mapping relationship analysis and mining are performed to achieve disambiguation and fusion of entities in the original graph; S6 provides a standard and unified relational query interface for heterogeneous graphs and fused index subgraphs; In step S4, constructing the fusion index subgraph includes the following steps: Through AI feature comparison, rule models, and graph reasoning, entity relationship mapping between heterogeneous graphs is performed, and the mapping relationship is stored in the form of sub-graphs to form a fusion index sub-graph; Constructing the fusion index subgraph also includes the following steps: S401: Through AI feature comparison, the images of unknown entities in the view domain knowledge graph and the entity images in the business domain knowledge graph are matched for similarity. If the similarity meets the set threshold, a heterogeneous graph entity mapping relationship is established. S402: Based on the attribute values ​​of entities and relationships in the heterogeneous graph, data matching is performed in the two types of graphs through rule configuration, and mapping relationships are established between the entities in the matched heterogeneous graphs; S403: Combined with the spatiotemporal dimension information, first-degree relationship search and reasoning are performed on the entities in the heterogeneous graph to find a set of entities with similar relationships within a limited spatiotemporal range. These entities are then mapped to each other. The weight of the mapping relationship is set based on the number of in- and out-degrees, and the confidence level of the mapping relationship is increased through manual judgment. S404: The heterogeneous graph entity mapping relationship formed from S401 to S403 is stored in a graph database to form a fusion index subgraph.

2. The heterogeneous data fusion indexing method based on knowledge graph according to claim 1 is characterized in that: In step S1, the heterogeneous entity mapping relationship includes entity type 1 number, entity type 2 number, entity mapping confidence and mapping time.

3. The heterogeneous data fusion indexing method based on knowledge graph according to claim 1 is characterized in that: In step S3, constructing the view domain knowledge graph includes the following steps: Based on the video image data, structured extraction is performed, and cluster analysis and data mining analysis are used to discover view target entities and view target relationships. The view entity and view relationship data are stored in a graph database to form a view domain knowledge graph.

4. The heterogeneous data fusion indexing method based on knowledge graph according to claim 3 is characterized by: The view target entities include real-name person files, anonymous person files, vehicle files, camera equipment, non-motor vehicle files and location areas; the view target relationships include personnel companionship, person accompaniment, vehicle accompaniment, vehicle driving and equipment file capture relationships.

5. The heterogeneous data fusion indexing method based on knowledge graph according to claim 1 is characterized by: In step S3, building a business domain knowledge graph includes the following steps: Based on the business data generated by the public security industry business system, a business domain knowledge graph is constructed. Through big data and NLP analysis, entity data and relationship data are extracted from structured or text data. The extracted entity data and relationship data are stored in a graph database to form a business domain knowledge graph.

6. The heterogeneous data fusion indexing method based on knowledge graph according to claim 5 is characterized by: The entity data includes people, things, places, objects, and organizations; the relationship data includes family relationships, communication relationships, transportation, transaction transfers, and property rights relationships.

7. A heterogeneous data fusion indexing system based on knowledge graph, characterized by: include: The knowledge management module is used to define and configure entities and relationships in heterogeneous graphs, as well as to configure entity mapping relationships in fusion index subgraphs; The entity mapping module provides entity mapping between heterogeneous graphs through tasks. Task types include AI construction tasks, rule model tasks, and knowledge reasoning tasks. It obtains data from heterogeneous graphs through the graph data provided by the library adaptation module, and realizes the mapping relationship between entities in heterogeneous graphs through AI feature comparison, big data stream batch computing, and graph mining calculations based on machine learning. The results of the analysis and calculation are uniformly stored through the knowledge storage module to construct a fusion index sub-graph; The entity association analysis module is used to implement secondary association of entities within the constructed fusion index sub-graph through a path analysis algorithm, thereby achieving entity disambiguation in the original heterogeneous graph. The graph unified query module is used to uniformly encapsulate the graph data interface, support relational query on a single graph, and relational fusion query on fusion index sub-graphs; AI engine module, used to provide portrait comparison based on AI capabilities; Graph mining and analysis module, used to provide graph mining and analysis, and supports graph search algorithm, shortest path algorithm, N-degree relationship tree, centrality algorithm and community discovery algorithm; The graph database adaptation module is used to adapt the operation interface of different graph databases; The knowledge storage module stores the data configured by the knowledge management module, stores the results of the entity mapping module analysis in a graph structure, and performs full-text index storage on heterogeneous entity and relationship data. It also includes: a method for executing or completing the steps in any one of the methods described in claims 1-6.

Images