A domain knowledge graph visualization construction method and device

Abstract

The application relates to the general technical field of artificial intelligence, and discloses a domain knowledge graph visualization construction method and device, which can be used for automatically and continuously constructing a knowledge graph for specific domain multi-source heterogeneous data. The construction method comprises the following steps: acquiring a data source used for constructing a domain knowledge graph; mapping the data source and a domain ontology model through a visualization mode to generate an information extraction meta-model; automatically extracting data from the data source based on the information extraction meta-model to generate a knowledge graph triple; performing knowledge fusion and storage on the generated domain knowledge graph triple to realize continuous construction of a visualized domain knowledge graph. The application reduces the technical complexity of the domain knowledge graph construction process through the visualization mode, the construction process is intuitive, the construction efficiency of the knowledge graph can be significantly improved, the application has flexibility and universality, and can be applied to various business fields.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence technology, specifically relating to a method and apparatus for constructing a domain knowledge graph visualization. Background Technology

[0002] With the rise of next-generation artificial intelligence technologies, represented by deep learning, knowledge graph technology has gradually occupied an important position in the field of artificial intelligence, achieving widespread application and remarkable results. A knowledge graph is a large-scale semantic network that depicts various physical entities, attributes, and their semantic relationships in the objective world. A knowledge graph is also a structured knowledge relationship graph formed by converting various data in the objective world into subject-predicate-object (SPO) triples.

[0003] Knowledge graph construction is a complex systems engineering project. For a specific domain knowledge graph, the construction process requires the full participation and even leadership of domain experts, such as in domain ontology model construction and data annotation. Furthermore, the knowledge graph construction process is technology-intensive, currently requiring deep support from technical personnel. Therefore, technical and business personnel are tightly coupled in the current domain knowledge graph construction process. Consequently, the quality of the constructed domain knowledge graph depends heavily on the degree of close collaboration among domain experts, technical personnel, and other stakeholders, directly leading to problems such as high complexity, low flexibility, and poor scalability in the current domain knowledge graph construction process. Summary of the Invention

[0004] To address the problems of high complexity, low flexibility, and poor scalability in the current domain knowledge graph construction process, this invention discloses a domain knowledge graph visualization construction method and apparatus. This method can fully leverage the leading role of domain experts, significantly reduce the tight coupling between domain experts, technicians, and other personnel during the knowledge graph construction process, significantly improve the efficiency of domain knowledge graph construction, shorten the construction time, and has advantages such as intuitive construction process and strong scalability.

[0005] This invention provides a method for constructing a domain knowledge graph visualization, comprising the following steps:

[0006] Step S101: Obtain multi-source heterogeneous data sources for constructing the domain knowledge graph;

[0007] Step S102: Establish the mapping between the data source and the domain ontology model through visualization, and generate an information extraction meta-model;

[0008] Step S103: Based on the information extraction meta-model, automatically extract data from the data source to generate knowledge graph triples;

[0009] Step S104: Perform knowledge fusion and storage on the generated domain knowledge graph triples to achieve continuous construction of a visualized domain knowledge graph.

[0010] Step S101 specifically includes:

[0011] Based on specific domain business scenarios, create domain-related business concepts (entity types) and relationships, and generate a domain ontology model, which includes concepts, attribute types, and relationships between concepts;

[0012] Data sources are categorized according to specific domain business scenarios, including structured data and unstructured data;

[0013] For structured data sources, data preprocessing is performed based on the domain ontology model, representing each business concept as a structure, with each structure field corresponding to a concept and its attributes in the domain ontology model; one or more related structured data tables or files in the data source are preprocessed into the aforementioned structure.

[0014] For unstructured data, it is uniformly preprocessed into text file data with UTF-8 encoding.

[0015] When creating a domain concept, the concept's own attribute types are created simultaneously.

[0016] In step S102, the method for generating a domain-specific information extraction meta-model specifically includes:

[0017] Step S201: Read the pre-created domain ontology model and visualize the domain concepts, attributes, and various relationships within it;

[0018] Step S202: Read the data source used to construct the domain knowledge graph;

[0019] Step S203: Visually map the entities and relationships in the data source to the concepts and relationships in the domain ontology model. This includes:

[0020] For preprocessed structured data in a specific domain, the table or file structure corresponding to each type of entity is read, and the fields in the structure are mapped one by one to the concepts and attributes in the domain ontology model through visualization, generating a meta-model for extracting structured data information.

[0021] For preprocessed text data in a specific domain, a certain number of text data are selected for visual sequence labeling to generate a text data information extraction meta-model. Based on the domain ontology model, the entity sequence in the text data is labeled as the corresponding concept entity in the domain ontology model, and the entity attribute sequence in the text data is labeled as the corresponding concept entity attribute in the domain ontology model. The relationship between entities is automatically generated through the corresponding concept relationship in the domain ontology model.

[0022] Step S204: Extract the meta-model from the stored information.

[0023] Each generated information extraction meta-model is applied to a data source with the same structure.

[0024] In step S103, the method for automatically extracting data from the data source and generating knowledge graph triples based on the information extraction meta-model specifically includes:

[0025] For structured data sources, by parsing the aforementioned generated information extraction meta-model, the information extraction model obtains the mapping relationship between the data source fields and the domain ontology model, and extracts data from the data source according to the mapping relationship to generate domain knowledge graph triples (entity, relation, attribute) or (entity, relation, entity).

[0026] For unstructured data sources, the information extraction model automatically identifies entities, attributes, and relationships in a specific domain from the input text data, and combines the extracted information to generate domain knowledge graph triples (entity, relation, attribute) or (entity, relation, entity) based on the information extraction meta-model.

[0027] In step S103, a visualization training method for an information extraction model for text data in a specific domain is described below. Figure 3 Specifically, it includes:

[0028] Step S301: Create an information extraction model for text data.

[0029] The information extraction model can be either a sequential extraction model that extracts entities first and then relations, or a joint extraction model that combines entity extraction and relation extraction.

[0030] Step S302: Train the information extraction model based on the labeled domain-specific information extraction dataset.

[0031] Based on the text data information extraction meta-model generated through visual annotation, the selected information extraction model is trained to obtain a domain-specific text data information extraction model.

[0032] Step S303: Publish the domain-specific text data information extraction model.

[0033] Select an information extraction model for domain-specific text data from a pre-created list of information extraction models using a visual approach.

[0034] Step S104 specifically includes:

[0035] Based on a pre-created domain ontology model, the fusion rules for entities corresponding to each domain concept are set in a visual manner according to the concepts and their attributes. For each concept in the domain ontology model, attributes for knowledge fusion are selected, and a similarity measurement function and a similarity measurement threshold are set for each attribute value. The relationship between multiple knowledge fusion measurement attributes under the same concept entity is a logical AND relationship.

[0036] The knowledge fusion module reads the knowledge fusion rules, calculates the entity or attribute similarity based on the fusion rules, and performs knowledge fusion processing on the knowledge graph triples.

[0037] For the knowledge graph triples obtained after knowledge fusion, the knowledge storage module stores them in the corresponding database;

[0038] For knowledge graph data sources that are continuously generated in a specific domain, the information extraction module continuously extracts information from data sources with the same structure based on the information extraction meta-model and generates domain knowledge graph triples.

[0039] The present invention also provides a domain knowledge graph visualization construction device, comprising:

[0040] The data source acquisition module is used to continuously acquire structured and unstructured data from a specific domain for building a domain knowledge graph, and to perform preprocessing operations on the data source.

[0041] The data mapping module provides a visual data mapping mechanism that maps entities and attributes in the domain knowledge graph of the data source to concepts and attributes in the pre-created domain ontology model.

[0042] The information extraction module is used to continuously and automatically extract entities, attributes, and relationships from data sources in a specific domain, and generate domain knowledge graph triples (entity, relationship, attribute) or (entity, relationship, entity).

[0043] The knowledge fusion module provides a visual configuration mechanism for fusion of similar entities in a domain knowledge graph. It calculates the similarity between entities and between attributes according to pre-created fusion rules, thereby realizing knowledge fusion operations.

[0044] The knowledge storage module provides a mechanism for the continuous persistence of domain knowledge graph triples.

[0045] Optionally, the data source acquisition module includes:

[0046] The domain ontology model creation unit is used to provide a visualization mechanism for building models of concepts, attributes, and their relationships in a specific domain.

[0047] The structured / semi-structured data preprocessing unit provides data preprocessing operations for original structured or semi-structured data in a specific domain, enabling the mapping of data structure semantics to domain-related concepts, attributes, and relationships.

[0048] The unstructured data preprocessing unit provides an extraction and transformation program to unify the file storage format of text files into UTF-8 encoded text data.

[0049] Optionally, the data mapping module includes:

[0050] The structured data mapping unit is used to provide a visualization mechanism to map data columns in structures such as tables or files to concepts and attributes in the domain ontology model.

[0051] The text data mapping unit is used to provide a visualization mechanism to map specific domain entities and attributes in text data to concepts and attributes in the domain ontology model.

[0052] Optionally, the information extraction module includes:

[0053] The structured data extraction unit is used to convert structured data into knowledge graph triples based on the information extraction meta-model.

[0054] The text data information extraction model training unit is used to train an information extraction model that automatically extracts domain knowledge graph triples from text data based on pre-labeled sequential text data.

[0055] The text data information extraction unit is used to automatically extract domain knowledge graph triples from continuously input text data using a pre-trained information extraction model.

[0056] Optionally, the knowledge fusion module includes:

[0057] The entity fusion unit is used to calculate the similarity between entity names and related attribute values, and provides entity fusion functionality according to entity fusion rules.

[0058] The attribute fusion unit is used to provide attribute fusion functionality for the same entity based on entity similarity and attribute relationships.

[0059] The present invention also provides a domain knowledge graph visualization construction terminal device, which includes at least a processor and a memory, as well as multiple computer instructions stored in the memory and capable of running on the processor. When the processor executes the computer instructions in the memory, it implements the above-described domain knowledge graph visualization construction method.

[0060] The present invention also provides a computer-readable storage medium for storing computer-readable instructions, wherein when the readable computer instructions stored in the storage medium are loaded and executed by the processor, the above-mentioned domain knowledge graph visualization construction method is implemented.

[0061] The beneficial effects of this invention are:

[0062] The present invention has the following advantages:

[0063] (1) It can automatically extract information from structured, semi-structured and unstructured data in a specific domain, generate domain knowledge graph triples, realize the continuous construction of domain knowledge graphs, give full play to the leading role of domain experts, significantly improve the construction efficiency of domain knowledge graphs, and has the characteristics of intuitive, flexible and highly scalable construction process.

[0064] (2) The process of constructing domain knowledge graphs has been highly abstracted and standardized, and the efficient construction of domain knowledge graphs has been achieved by using a visual human-computer interaction method;

[0065] (3) This invention is not limited to knowledge graph construction in a certain field. It has universality and domain adaptability and can be applied to various fields with similar knowledge graph construction needs. Attached Figure Description

[0066] Figure 1 This is a flowchart of a domain knowledge graph visualization construction method according to Embodiment 1 of the present invention;

[0067] Figure 2 This is a flowchart of the method for generating a meta-model for information extraction in a specific domain according to the present invention;

[0068] Figure 3 This is a flowchart of a visualization training method for an information extraction model for text data in a specific domain, as described in this invention.

[0069] Figure 4 This is a flowchart of the method for knowledge fusion based on domain knowledge graph triples according to the present invention;

[0070] Figure 5 This is a schematic diagram of the structure of a domain knowledge graph visualization construction device in Embodiment 2 of the present invention. Detailed Implementation

[0071] This invention provides a method and apparatus for visually constructing knowledge graphs for a specific domain. It enables customized settings for the visualization of the domain knowledge graph construction process, and can automatically and continuously acquire domain data sources to continuously generate, merge, update, and store domain knowledge graph triples. The domain knowledge graph constructed using this invention can be used for domain-specific intelligent question answering, precise retrieval, intelligent reasoning, and other applications. The technical solution of this invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention.

[0072] Example 1, such as Figure 1 As shown, in this embodiment, the present invention provides a method for constructing a domain knowledge graph visualization, specifically including:

[0073] Step S101: Obtain multi-source heterogeneous data sources for constructing the domain knowledge graph.

[0074] The field described in this invention is not limited to any specific field. In this embodiment, the invention is explained in the context of a specific field.

[0075] Based on specific domain business scenarios, domain-related business concepts and relationships are created to generate a domain ontology model, which includes domain concepts, their own attributes, and the relationships between concepts. Building a domain ontology is a systematic project that requires the participation of domain experts with a deep understanding of the specific domain. During the construction process, existing ontology building tools (such as Protégé) can be used, or a custom-developed, visual, domain-specific ontology building tool can be created.

[0076] For data sources used to build knowledge graphs in specific domains, they are classified according to the domain ontology model and the specific domain business scenarios, including structured data and unstructured data.

[0077] Optionally, for semi-structured data in the domain, a conversion program can be used to transform it into the corresponding structured data.

[0078] For structured data sources, data preprocessing is performed based on the domain ontology model. Data sources belonging to a certain business concept are represented as a type of structure. The fields of this structure correspond one-to-one with a concept and its attributes in the domain ontology model. One or more related structured data tables or files in the data source are preprocessed into the aforementioned structure.

[0079] Unstructured data in a specific domain may be stored in various file formats (such as Word, PDF, text files, etc.), and the file storage format can be unified into UTF-8 encoded text file data through appropriate extraction and conversion programs.

[0080] Step S102: Establish the mapping between the data source and the domain ontology model through visualization, and generate an information extraction meta-model.

[0081] Step S103: Based on the information extraction meta-model, automatically extract data from the data source to generate knowledge graph triples.

[0082] By visually selecting an information extraction model for a specific domain's knowledge graph data source, and combining it with an information extraction meta-model, the information extraction model automatically extracts data from the data source to generate knowledge graph triples, specifically including:

[0083] For structured data sources, by parsing the aforementioned generated information extraction meta-model, the information extraction model obtains the mapping relationship between the data source fields and the domain ontology model, and extracts the corresponding column data from the data source according to the mapping relationship to generate domain knowledge graph triples (entity, relation, attribute) or (entity, relation, entity).

[0084] For unstructured data sources, the information extraction model automatically identifies entities, attributes, and relationships in a specific domain from the input text data, and combines the extracted information to generate domain knowledge graph triples (entity, relation, attribute) or (entity, relation, entity) based on the information extraction meta-model.

[0085] After the domain-specific text data information extraction model is trained, it is published as an application service and provided with application service description information so that domain knowledge graph builders can select an information extraction model for domain-specific text data from a list of pre-created and trained information extraction models in a visual manner.

[0086] Step S104: Perform knowledge fusion and storage on the generated domain knowledge graph triples to achieve continuous construction of a visualized domain knowledge graph.

[0087] Data used to build knowledge graphs in a specific domain may come from multiple sources, and there may be duplication or conflict between data from different sources. Therefore, after the domain knowledge graph triples are generated from the data sources, knowledge fusion operations need to be performed before knowledge storage.

[0088] Optionally, for knowledge graph data sources that are continuously generated in a specific domain, the information extraction module continuously extracts information from data sources with the same structure based on the information extraction meta-model, generates domain knowledge graph triples, and after the knowledge fusion operation, the knowledge storage module stores them in the corresponding database (e.g., a relational database or a graph database).

[0089] In step S102, the method for generating a domain-specific information extraction meta-model is described below. Figure 2 The specific steps are as follows:

[0090] Step S201: Read the pre-created domain ontology model and visualize the domain concepts, attributes and various relationships within it.

[0091] Step S202: Read the data source used to construct the domain knowledge graph and visualize the data; for structured data, visualize the data structure columns and fields and a small amount of preview data; for text data, visualize the specific text content.

[0092] Step S203: Visually map the entities and relationships in the data source to the concepts and relationships in the domain ontology model. This includes:

[0093] For preprocessed structured data in a specific domain, a meta-model for extracting structured data information is generated by visually mapping the fields in the table or file structure corresponding to each type of entity to the concepts and attributes in the domain ontology model.

[0094] For preprocessed text data in a specific domain, a certain number of text data are selected for visual sequence labeling to generate a text data information extraction meta-model. Based on the domain ontology model, the entity sequence in the text data is labeled as the corresponding concept entity in the domain ontology model, and the entity attribute sequence in the text data is labeled as the corresponding concept entity attribute in the domain ontology model. The relationship between entities is automatically generated through the corresponding concept relationship in the domain ontology model.

[0095] Step S204: Extract the meta-model from the stored information.

[0096] Information extraction metamodels can be stored in a database or a file (such as a JSON file). For structured data, each stored information extraction metamodel operates on a data source with the same structure. For text data, each information extraction metamodel corresponds to a specific domain business scenario.

[0097] In step S103, a visualization training method for an information extraction model for text data in a specific domain is described below. Figure 3 Specifically, it includes:

[0098] Step S301: Create an information extraction model for text data.

[0099] The information extraction model can be a sequential extraction model that first extracts entities and then extracts relations, or a joint extraction model that combines entity extraction and relation extraction. Specifically, it can be:

[0100] Pre-trained language models based on Transformers, such as BERT (Bidirectional Encoder Representation from Transformers), GCN (Graph Convolutional Network), Bidirectional Long Short-Term Memory (BiLSTM), and CRF (Conditional Random Field), as well as combined models (such as BERT-BiLSTM-CRF), are all required. Each information extraction model must provide entity (attribute) extraction and relation extraction capabilities.

[0101] Step S302: Train the information extraction model based on the labeled domain-specific information extraction dataset.

[0102] Based on the text data information extraction meta-model generated through visual annotation, namely the domain-specific information extraction annotation dataset, the annotation dataset is divided into a training set, a validation set, and a test set. The created information extraction model is then trained to obtain the domain-specific text data information extraction model.

[0103] Step S303: Publish the domain-specific text data information extraction model.

[0104] After the domain-specific text data information extraction model is trained, it is published as an application service and provided with application service description information so that domain knowledge graph builders can select an information extraction model for domain-specific text data from a list of pre-created and trained information extraction models in a visual manner.

[0105] In step S104, the method for knowledge fusion based on domain knowledge graph triples is described in [link to article]. Figure 4 The specific steps include:

[0106] Step S401: Create knowledge fusion rules in the domain knowledge graph.

[0107] Based on a pre-created domain ontology model, and according to concepts and their attributes, the fusion rules for entities corresponding to each domain concept are set in a visual manner. For each concept in the domain ontology model, attributes for knowledge fusion are selected, and a similarity measurement function (such as cosine distance) and a similarity measurement threshold are set for each attribute value. The combination relationship between multiple knowledge fusion measurement attributes under the same concept entity is a logical AND relationship.

[0108] Step S402: Store the knowledge fusion rules.

[0109] The knowledge fusion rules, which are set up visually, are stored as part of the domain ontology model.

[0110] Step S403: Implement knowledge fusion.

[0111] The knowledge fusion module reads the knowledge fusion rules, calculates the entity or attribute similarity based on the fusion rules, and performs knowledge fusion processing on the knowledge graph triples.

[0112] Optionally, for knowledge graph data sources that are continuously generated in a specific domain, the information extraction module continuously extracts information from data sources with the same structure based on the information extraction meta-model, generates domain knowledge graph triples, and after the knowledge fusion operation, the knowledge storage module stores them in the corresponding database (e.g., a relational database or a graph database).

[0113] Example 2, and Figure 1 Corresponding to the method shown, embodiments of the present invention provide a domain knowledge graph visualization construction device, such as... Figure 5 As shown, it includes:

[0114] The data source acquisition module 501 is used to continuously acquire structured and unstructured data from a specific domain for constructing a domain knowledge graph, and to perform preprocessing operations on the data source.

[0115] The data mapping module 502 is used to provide a visual data mapping mechanism to map entities and attributes in the domain knowledge graph in the data source to concepts and attributes in the pre-created domain ontology model;

[0116] The information extraction module 503 is used to continuously and automatically extract entities, attributes and relationships from data sources in a specific domain, and generate domain knowledge graph triples (entity, relationship, attribute) or (entity, relationship, entity);

[0117] The knowledge fusion module 504 is used to provide a fusion configuration mechanism for similar entities in a visualized domain knowledge graph, and to calculate the similarity between entities and entities, and between attributes, according to pre-created fusion rules, so as to realize the knowledge fusion operation.

[0118] The knowledge storage module 505 provides a mechanism for the continuous persistence of domain knowledge graph triples.

[0119] Optionally, the data source acquisition module 501 includes:

[0120] Domain ontology model creation unit 5011 is used to provide a visualization mechanism to establish a model of concepts, attributes and their relationships in a specific domain.

[0121] The structured / semi-structured data preprocessing unit 5012 provides data preprocessing operations for original structured or semi-structured data in a specific domain, enabling the correspondence between the semantics of the data structure and the domain-related concepts, attributes, and relationships.

[0122] The unstructured data preprocessing unit 5013 provides an extraction and conversion program to unify the file storage format into a text file encoded in UTF-8.

[0123] Optionally, the data mapping module 502 includes:

[0124] The structured data mapping unit 5021 is used to provide a visualization mechanism to map data columns in structures such as tables or files to concepts and attributes in the domain ontology model;

[0125] The text data mapping unit 5022 is used to provide a visualization mechanism to map specific domain entities and attributes in text data to concepts and attributes in the domain ontology model.

[0126] Optionally, the information extraction module 503 includes:

[0127] The structured data extraction unit 5031 is used to convert structured data into knowledge graph triples based on the information extraction meta-model.

[0128] The text data information extraction model training unit 5032 is used to train an information extraction model for automatically extracting domain knowledge graph triples from text data based on pre-labeled sequential text data.

[0129] The text data information extraction unit 5033 is used to automatically extract and generate domain knowledge graph triples from continuously input text data using a pre-trained information extraction model.

[0130] Optionally, the knowledge fusion module 504 includes:

[0131] Entity fusion unit 5041 calculates the similarity between entity names and related attribute values, and provides entity fusion functionality according to entity fusion rules;

[0132] The attribute fusion unit 5042 provides attribute fusion functionality for the same entity based on entity similarity and attribute relationships.

[0133] Those skilled in the art will readily understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the internal structure of the above-described device can be divided into different functional units or modules according to specific needs to complete all or part of the above functions. Furthermore, the number of functional units and modules in the embodiments can also be determined according to actual application needs. The names of the functional units and modules are for ease of distinction and are not intended to limit the scope of protection of this invention.

[0134] Example 3: This embodiment of the invention provides a domain knowledge graph visualization construction terminal device, which includes at least a processor and a memory. The above-mentioned program modules such as data source acquisition, data mapping, information extraction, knowledge fusion, and knowledge storage are stored in the memory. The processor executes the program modules in the memory to implement the steps in any of the above method embodiments.

[0135] Example 4: This embodiment of the invention provides a computer-readable storage medium storing program modules such as data source acquisition, data mapping, information extraction, knowledge fusion, and knowledge storage. When these program modules are executed by a processor, they implement the steps in any of the above method embodiments.

[0136] Example 5: The present invention also provides a computer software system that, when executed on a computer terminal device, is applicable to the following specific domain knowledge graph construction steps:

[0137] It continuously acquires structured and unstructured data from a specific domain for constructing a domain knowledge graph, and performs preprocessing operations on the data sources; optionally, it creates a domain ontology model to establish a model of concepts, attributes, and their relationships in the specific domain; for the original structured or semi-structured data in the specific domain, it provides data preprocessing operations to realize the correspondence between data structure columns and fields and domain-related concepts, attributes, and relationships; for unstructured data, it uses an extraction and conversion program to unify the file storage format to UTF-8 encoded text files.

[0138] Through a visual data mapping mechanism, entities and attributes in the domain knowledge graph of the data source are mapped to concepts and attributes in the pre-created domain ontology model. For structured domain data, the visualization mechanism maps data columns in tables or files to concepts and attributes in the domain ontology model. For unstructured data, the visualization mechanism maps specific domain entities and attributes in text data to concepts and attributes in the domain ontology model.

[0139] It continuously and automatically extracts entities, attributes, and relationships from data sources in a specific domain and generates domain knowledge graph triples (entity, relationship, attribute) or (entity, relationship, entity). For structured data, the information extraction module converts the structured data into knowledge graph triples based on the information extraction meta-model. For text data, it trains an information extraction model based on pre-labeled sequential text data to automatically extract domain knowledge graph triples from the text data. The pre-trained information extraction model is then used to automatically extract and generate domain knowledge graph triples from the continuously input text data.

[0140] The system utilizes a visualization-based domain knowledge graph to configure the fusion of similar entities or attributes. It calculates the similarity between entities and between attributes based on pre-created fusion rules, thus enabling knowledge fusion operations. The system also calculates the similarity between entity names and related attribute values, performs entity fusion according to entity fusion rules, and performs attribute fusion based on entity similarity and attribute relationships. The knowledge storage module stores the fused domain knowledge graph triples, enabling continuous incremental updates to the domain knowledge graph.

[0141] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for constructing a domain knowledge graph visualization, characterized in that, The steps of this method are as follows: Step S101: Obtain multi-source heterogeneous data sources for constructing the domain knowledge graph; Step S102: Establish a mapping between the data source and the domain ontology model through visualization, and generate an information extraction meta-model; Step S102 specifically includes: Step S201: Read the pre-created domain ontology model and visualize the domain concepts, attributes, and various relationships within it; Step S202: Read the data source used to construct the domain knowledge graph; Step S203: Map the entities and relationships in the data source to the concepts and relationships in the domain ontology model using visualization. This includes: For preprocessed structured data in a specific domain, the table or file structure corresponding to each type of entity is read, and the fields in the structure are mapped one by one to the concepts and attributes in the domain ontology model through visualization, generating a meta-model for extracting structured data information. For preprocessed text data in a specific domain, a certain number of text data are selected for visual sequence labeling to generate a text data information extraction meta-model. Based on the domain ontology model, the entity sequence in the text data is labeled as the corresponding concept entity in the domain ontology model, and the entity attribute sequence in the text data is labeled as the corresponding concept entity attribute in the domain ontology model. The relationship between entities is automatically generated through the corresponding concept relationship in the domain ontology model. Step S204: Extract the meta-model from the stored information; Each generated information extraction meta-model is applied to a data source with the same structure. Step S103: Based on the information extraction meta-model, automatically extract data from the data source to generate knowledge graph triples; specifically including: For structured data sources, by parsing the aforementioned generated information extraction meta-model, the information extraction model obtains the mapping relationship between the data source fields and the domain ontology model, and extracts data from the data source according to the mapping relationship to generate domain knowledge graph triples (entity, relation, attribute) or (entity, relation, entity). For unstructured data sources, the information extraction model automatically identifies entities, attributes, and relationships in a specific domain from the input text data, and combines the extracted information to generate domain knowledge graph triples (entity, relation, attribute) or (entity, relation, entity) based on the information extraction meta-model. One of the methods for visually training an information extraction model for text data in a specific domain includes: Step S301: Create an information extraction model for text data; The information extraction model can be either a sequential extraction model that extracts entities first and then relations, or a joint extraction model that integrates entity extraction and relation extraction. Step S302: Train the information extraction model based on the labeled domain-specific information extraction dataset; Based on the text data information extraction meta-model generated through visual annotation, the selected information extraction model is trained to obtain a domain-specific text data information extraction model. Step S303: Publish the domain-specific text data information extraction model; Select an information extraction model for domain-specific text data from a pre-created list of information extraction models using a visual method; Step S104: Perform knowledge fusion and storage on the generated domain knowledge graph triples to achieve continuous construction of a visualized domain knowledge graph.

2. The method according to claim 1, characterized in that, Step S101 specifically includes: Based on specific domain business scenarios, create domain-related business concepts and relationships, and generate a domain ontology model, which includes concepts, attribute types, and relationships between concepts. Data sources are categorized according to specific domain business scenarios, including structured and unstructured data; For structured data sources, data preprocessing is performed based on the domain ontology model, representing each business concept as a structure, where the fields correspond one-to-one with a concept and its attributes in the domain ontology model; one or more related structured data tables or files in the data source are preprocessed into the aforementioned structure. For unstructured data, it is uniformly preprocessed into UTF-8 encoded text file data; When creating a domain concept, the concept's associated attribute types are created simultaneously.

3. The method according to claim 1, characterized in that, Step S104 specifically includes: Based on a pre-created domain ontology model, and according to concepts and their attributes, the fusion rules for entities corresponding to each domain concept are set in a visual manner; for each concept in the domain ontology model, attributes for knowledge fusion are selected, and a similarity measurement function and a similarity measurement threshold are set for each attribute value; the relationship between multiple knowledge fusion measurement attributes under the same concept entity is a logical AND relationship; The knowledge fusion module reads the knowledge fusion rules, calculates the entity or attribute similarity based on the fusion rules, and performs knowledge fusion processing on the knowledge graph triples. For the knowledge graph triples obtained after knowledge fusion, the knowledge storage module stores them in the corresponding database; For knowledge graph data sources that are continuously generated in a specific domain, the information extraction module continuously extracts information from data sources with the same structure based on the information extraction meta-model and generates domain knowledge graph triples.

4. A domain knowledge graph visualization construction device, characterized in that: include: The data source acquisition module is used to continuously acquire structured and unstructured data from a specific domain for building a domain knowledge graph, and to perform preprocessing operations on the data source. The data source acquisition module includes: The domain ontology model creation unit is used to provide a visualization mechanism for building models of concepts, attributes, and their relationships in a specific domain. The structured / semi-structured data preprocessing unit provides data preprocessing operations for original structured or semi-structured data in a specific domain, enabling the mapping of data structure semantics to domain-related concepts, attributes, and relationships. The unstructured data preprocessing unit provides an extraction and transformation program to unify the file storage format of text files into UTF-8 encoded text data; The data mapping module provides a visual data mapping mechanism, mapping entities and attributes in the domain knowledge graph from the data source to concepts and attributes in a pre-created domain ontology model. The data mapping module includes: The structured data mapping unit is used to provide a visualization mechanism to map data columns in a table or file structure to concepts and attributes in the domain ontology model; The text data mapping unit is used to provide a visualization mechanism to map specific domain entities and attributes in text data to concepts and attributes in the domain ontology model; An information extraction module is used to continuously and automatically extract entities, attributes, and relationships from a specific domain data source, and generate domain knowledge graph triples (entity, relationship, attribute) or (entity, relationship, entity); the information extraction module includes: The structured data extraction unit is used to convert structured data into knowledge graph triples based on the information extraction meta-model. The text data information extraction model training unit is used to train an information extraction model for automatically extracting domain knowledge graph triples from text data based on pre-labeled sequential text data. The text data information extraction unit is used to automatically extract and generate domain knowledge graph triples from continuously input text data using a pre-trained information extraction model. For structured data sources, the generated information extraction meta-model is parsed. The information extraction model obtains the mapping relationship between the data source fields and the domain ontology model, and extracts data from the data source according to the mapping relationship to generate domain knowledge graph triples (entity, relation, attribute) or (entity, relation, entity). For unstructured data sources, the information extraction model automatically identifies entities, attributes, and relationships in a specific domain from the input text data, and combines the extracted information to generate domain knowledge graph triples (entity, relation, attribute) or (entity, relation, entity) based on the information extraction meta-model. One of the methods for visually training an information extraction model for text data in a specific domain includes: Step S301: Create an information extraction model for text data; The information extraction model can be either a sequential extraction model that extracts entities first and then relations, or a joint extraction model that integrates entity extraction and relation extraction. Step S302: Train the information extraction model based on the labeled domain-specific information extraction dataset; Based on the text data information extraction meta-model generated through visual annotation, the selected information extraction model is trained to obtain a domain-specific text data information extraction model. Step S303: Publish the domain-specific text data information extraction model; Select an information extraction model for domain-specific text data from a pre-created list of information extraction models using a visual method; The knowledge fusion module provides a visualized configuration mechanism for fusion of similar entities in a domain knowledge graph, and calculates the similarity between entities and entities, and between attributes, according to pre-created fusion rules, thereby realizing knowledge fusion operations. The knowledge storage module provides a mechanism for persistently storing domain knowledge graph triples; the knowledge fusion module includes: The entity fusion unit is used to calculate the similarity between entity names and related attribute values, and provides entity fusion functionality according to entity fusion rules. The attribute fusion unit is used to provide attribute fusion functionality for the same entity based on entity similarity and attribute relationships.

5. A domain knowledge graph visualization construction terminal device, characterized in that, It includes at least one processor and a memory, and multiple computer instructions stored in the memory and capable of running on the processor, wherein the processor implements the domain knowledge graph visualization construction method as described in any one of claims 1-3 when executing the computer instructions in the memory.

6. A computer-readable storage medium, characterized in that, Used to store computer-readable instructions, wherein when the computer-readable instructions stored in the storage medium are loaded and executed by a processor, the domain knowledge graph visualization construction method as described in any one of claims 1-3 is implemented.

Images