Knowledge graph visualization tool construction system

Abstract

The invention relates to a knowledge graph visualization tool construction system, and belongs to the field of knowledge graphs. The system comprises a data processing module, an algorithm management module, a graph design module, an extraction and mapping module and a data application module. The method comprises the following steps: acquiring a data source required for constructing a knowledge graph, and carrying out cleaning, format conversion and integration on data; knowledge is extracted and mapped through knowledge mapping, knowledge graph creation and structural design are supported, entity and relation management and visualized display of a knowledge graph are supported, and meanwhile, a graph knowledge question and answer function is provided. The modules cooperate with one another, so that the data processing efficiency, the knowledge extraction accuracy, the dynamic updating capability and the visual interactivity are improved, and the requirements of users in data analysis and mining are met.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of knowledge graph, and particularly relates to a knowledge graph visualization tool construction system. BACKGROUND

[0002] With the development of technology, the application of knowledge graph in the testing field is gradually becoming an important means to improve testing efficiency and optimize paths. By using the visualization and dynamic interaction technology of knowledge graph, complex testing information is structured and graphed to realize more efficient knowledge management and task execution. By extracting entities, relationships and attributes from technical documents, defect reports and use case libraries in the testing field, a knowledge network covering testing data is constructed. The knowledge graph visualization tool can present testing data and logical relationships in a multi-dimensional network structure, enabling testers to quickly understand the contextual relationships of testing data. The application of knowledge graph visualization tools in the testing field not only improves testing efficiency, but also promotes the deep integration of knowledge and data, providing a more intelligent solution for information system testing.

[0003] Current knowledge graph visualization tools have some problems, mainly in the following aspects: data processing, tools usually lack efficient cleaning, integration and format conversion capabilities for unstructured data, making it difficult to guarantee data quality; algorithm management, existing tools have black box algorithms that are difficult to configure and extend, and lack deep adaptation to domain knowledge; knowledge mapping, knowledge extraction and association accuracy is not high, and knowledge fragmentation or redundancy problems are prone to occur. In addition, these tools usually have weak interactivity and lack visualization analysis capabilities, making it difficult to meet the needs of testing field experts for intuitive exploration and rapid verification of knowledge. SUMMARY

[0004] (I) Technical problems to be solved The technical problem to be solved by the present application is how to provide a knowledge graph visualization tool construction system to solve the technical defects of low data processing efficiency, insufficient knowledge extraction accuracy, weak dynamic updating capability and poor visual interactive of knowledge graph visualization tools.

[0005] (ii) Technical solutions In order to solve the above technical problems, the present application provides a knowledge graph visualization tool construction system, which comprises: a data processing module, an algorithm management module, a graph design module, an extraction and mapping module, and a data application module. The data processing module is used to acquire and process multi-source data required for constructing knowledge graph, including structured, semi-structured and unstructured data; first, a unified data import method is used to realize multi-source data, and data is efficiently acquired from databases, file systems and network interfaces; then, data cleaning and verification are performed. An algorithm management module is used for comprehensively managing algorithm models in the knowledge graph visualization tool, covering knowledge extraction, data cleaning, and knowledge mapping models; flexible model management options are provided for users, including model training, model testing, and model publishing, and efficient configuration and control of each stage are supported; A graph design module is used for the structured design of a knowledge graph, unified domain ontology modeling is performed for concepts, attributes, and concept hierarchies in a test field, entity types and their relationships in the field are clearly defined, and the constructed domain ontology is supplemented, verified, and corrected based on business data, so that the integrity and accuracy of the knowledge graph are ensured; An extraction and mapping module extracts entities, attributes, and relationship information from field-related unstructured business documents in combination with manual annotation and automatic annotation methods, and ensures the accuracy and consistency of knowledge expression by using data dictionaries and disambiguation techniques; the extraction results are mapped to the designed domain ontology structure by using efficient knowledge mapping technology, and a required domain knowledge graph is generated; knowledge fusion and dynamic updating are supported, and the knowledge graph content is continuously optimized; A data application module realizes two functions of visualization and intelligent question answering based on the constructed domain knowledge graph; the visualization submodule dynamically displays the graph topology structure and entity relationships, and supports interactive exploration; the question answering submodule accurately generates answers by natural language analysis and graph reasoning, and supports multi-round interaction of users.

[0006] (Three) beneficial effects The present application proposes a knowledge graph visualization tool construction system, which is composed of five parts of data processing for multi-source data, algorithm management for field data, graph design for field knowledge, extraction and mapping based on field data, and data application based on field data, and each module cooperates with each other to improve the data processing efficiency, knowledge extraction accuracy, dynamic updating capability and visualizability, and meet the needs of users in data analysis and mining. The present application can significantly improve the data processing capability, knowledge extraction precision, dynamic maintenance capability and interactive experience of the knowledge graph visualization tool, and is suitable for various application scenarios requiring large-scale data analysis and knowledge mining. BRIEF DESCRIPTION OF DRAWINGS

[0007] Figure 1 The technical overall scheme diagram of the knowledge graph visualization tool of the present application is shown in the figure; Figure 2 The data processing design scheme diagram is shown in the figure; Figure 3 The algorithm management design scheme diagram is shown in the figure; Figure 4 The design scheme diagram of knowledge graph design, knowledge extraction and knowledge mapping is shown in the figure. DETAILED DESCRIPTION

[0008] In order to make the purpose, content and advantages of the present application clearer, the specific embodiments of the present application are described in further detail below in combination with the drawings and examples.

[0009] Based on the problems existing in the background art, the present application proposes a knowledge graph visualization tool construction method and system, aiming to improve the technical defects of the knowledge graph visualization tool such as low data processing efficiency, insufficient knowledge extraction accuracy, weak dynamic updating ability and poor visual interactive performance. Meet the needs of users for data analysis and mining. The knowledge graph visualization tool includes data processing, algorithm management, knowledge mapping, graph design, graph management and core modules. First, data processing is used to obtain the data source required for constructing the knowledge graph, and the data is cleaned, formatted and integrated to ensure data quality. Next, the algorithm management is responsible for managing the algorithms required for knowledge extraction, including algorithm model training, testing and publishing, while providing parameter configuration management, state monitoring and other functions to ensure efficient operation of the algorithm. Then, through knowledge mapping, knowledge is extracted and mapped, and data dictionary is used for knowledge disambiguation to ensure the accuracy and consistency of the knowledge structure. Graph design mainly supports knowledge graph creation and structured design, including entity type definition, relationship design and attribute configuration, laying the foundation for subsequent graph management. Finally, graph management supports entity and relationship management, visualization of knowledge graph, and provides graph knowledge question and answer function, effectively improving the answering ability to complex questions of users.

[0010] The present application proposes a knowledge graph visualization tool construction system, the overall technical scheme of which is shown in Figure 1 As shown in the figure, it includes a multi-source data-oriented data processing module, a field data-oriented algorithm management module, a field knowledge-oriented graph design module, an extraction and mapping module based on field data, and a data application module based on field data. Each module cooperates with each other to improve the data processing efficiency, knowledge extraction accuracy, dynamic updating ability and visual interactive performance, and meet the needs of users in data analysis and mining.

[0011] 1. Multi-source data-oriented data processing module The data processing module is used to obtain and process multi-source data required for constructing the knowledge graph, including structured, semi-structured and unstructured data. First, a unified data import method is used to realize multi-source data, and data is efficiently obtained from different sources such as databases, file systems and network interfaces. Subsequently, data cleaning and verification are performed to correct errors, remove duplicate data and format conversion, ensuring data integrity and consistency, and providing a high-quality data foundation for subsequent knowledge graph construction.

[0012] 2. Field data-oriented algorithm management module The algorithm management module is used for comprehensive management of algorithm models in the knowledge graph visualization tool, covering key models such as knowledge extraction, data cleaning, and knowledge mapping. The system provides users with flexible model management options, including model training, model testing, and model release, supporting efficient configuration and control of each stage.

[0013] The algorithm management module is used to manage various algorithms required for knowledge extraction and their running states. For unstructured data, the algorithm management module integrates natural language processing and knowledge extraction technology to convert information in business documents into structured data. At the same time, it provides training, testing, and release functions for algorithm models, supports parameter configuration management and state monitoring, and ensures efficient and stable operation of algorithms. In addition, through task unified management technology, it realizes centralized scheduling and control of different data analysis tasks, improving the efficiency and reliability of task execution.

[0014] 3. Domain-oriented graph design module The graph design module is used for the structured design of the knowledge graph, and performs unified domain ontology modeling for concepts, attributes, and concept hierarchies in the test domain, clearly defines entity types and their relationships within the domain, and supplements, verifies, and corrects the constructed domain ontology based on business data, thereby ensuring the completeness and accuracy of the knowledge graph.

[0015] The graph design module is used for the structured design of the knowledge graph, including entity type definition, relationship design, and attribute configuration. For a specific domain (such as the test domain), the graph design module can define key entities and their attributes and relationships, laying a structured foundation for the organization and display of knowledge. At the same time, the graph design results can directly provide references for knowledge extraction and mapping, ensuring that the generated graph is consistent with the domain semantics.

[0016] 4. Extraction and mapping module based on domain data The extraction and mapping module combines manual annotation and automatic annotation methods to extract entity, attribute, and relationship information from unstructured business documents related to the domain, and uses data dictionaries and disambiguation techniques to ensure the accuracy and consistency of knowledge expression. Through efficient knowledge mapping technology, the extraction results are mapped to the designed domain ontology structure to generate a domain knowledge graph that meets the requirements. At the same time, the module supports knowledge fusion and dynamic updating, solves the problem of data redundancy, and continuously optimizes the content of the knowledge graph.

[0017] 5. Data application module based on domain data The data application module is based on the constructed domain knowledge graph and realizes two major functions: visualization and intelligent question answering. The visualization submodule dynamically displays the graph topology structure and entity relationships, supporting interactive exploration; the question answering submodule accurately generates answers through natural language analysis and graph reasoning, and supports multi-round interaction with users.

[0018] Through the technical scheme, the application can significantly improve the data processing capability, knowledge extraction accuracy, dynamic maintenance capability and interactive experience of the knowledge graph visualization tool, and is suitable for various application scenarios requiring large-scale data analysis and knowledge mining.

[0019] Embodiment 1: First step: data processing The data processing is aimed to realize data reading, data cleaning and data checking functions, and to provide high-quality data support for subsequent applications. Considering that the test data is from various sources, involving local storage and remote database, and the data forms are various, including structured and unstructured data, the application designs multi-source data reading technology and data cleaning and checking technology. These technologies ensure that no matter the data storage form or structure, efficient reading, processing and verification can be achieved, providing a reliable foundation for knowledge graph construction and model training. The design scheme is as shown in Figure 2 . The data processing module of the application is aimed to realize data reading, data cleaning and data checking functions, and to provide high-quality, standardized data support for subsequent knowledge graph construction and model training. Given that the test data is from various sources, involving local storage and remote database, and the data forms cover structured and unstructured data, the application designs multi-source data reading technology and data cleaning and checking technology, ensuring that no matter the data storage method or structure form, efficient reading, processing and verification can be achieved. The structure design scheme is as shown in Figure 2 .

[0020] In the data reading stage, the module supports local file access and external database access. Local file access can handle multiple file formats, and for structured data (such as JSON, Excel files), it can efficiently read table-type and dictionary-type data, and for unstructured data (such as PDF, Word, TXT files), it can extract business documents, test reports and other text content, providing rich raw input for subsequent processing. External database access supports direct connection with mainstream database systems (such as MySQL), enabling fast access and unified extraction of remote data resources.

[0021] In the data cleaning stage, the module improves data quality and consistency through various optimization strategies to ensure that the input data for subsequent processing meets standardized requirements. Cleaning operations include deduplication to eliminate duplicate records and ensure data uniqueness, text normalization to identify and correct misspelled words to improve text accuracy, and format unification to standardize date formats and full-width and half-width characters, thereby eliminating format differences between different data sources. Through the above cleaning methods, the accuracy and usability of the data can be significantly improved.

[0022] In the data verification phase, the module further verifies the quality of the cleaned data to ensure that it meets the system's usage requirements. The data verification technology can record data types, cleaning methods, and cleaning status, and automatically identify inconsistent or abnormal data, marking and describing them for subsequent manual review or automatic correction. This process can effectively prevent quality defect data from entering the knowledge graph construction and analysis model training stage, thereby improving the overall reliability and accuracy of the system.

[0023] Second step Algorithm management The algorithm management module is used to comprehensively manage the algorithm models in the knowledge graph visualization tool, covering key models such as knowledge extraction, data cleaning, and knowledge mapping. The system provides users with flexible model management options, including model training, model testing, and model publishing, supporting efficient configuration and control of each stage. As shown in Figure 3 .

[0024] Model training The system supports flexible model training options, allowing users to select different data sets and training devices according to their needs, and set key training parameters such as batch size and training period to optimize model performance. At the same time, the module supports multi-task training processes to meet the needs of various scenarios such as knowledge extraction and mapping.

[0025] Model testing The system provides comprehensive model testing functions, allowing users to select appropriate test data sets and devices, and flexibly adjust test parameters (such as batch size) to verify the effectiveness of the model in knowledge cleaning and knowledge mapping scenarios. In addition, the module supports multi-dimensional evaluation results to ensure the accuracy and reliability of model performance.

[0026] Model publishing The system provides a simple model publishing function, allowing users to flexibly configure publishing devices and target addresses, deploying optimized models to the knowledge graph visualization tool to ensure their efficiency and stability in actual applications. The module also provides version management and release record functions to facilitate the iteration and maintenance of models.

[0027] Third step Graph design The graph design module in the embodiment of the present invention is associated with the knowledge mapping module, and together they form the core support unit for test field knowledge graph construction and application. This module aims to abstract knowledge such as concepts, attributes, and concept hierarchies in the test field, and perform unified domain ontology modeling to clearly define entity types and their relationships within the domain. Based on business data, the constructed domain ontology is supplemented, verified, and corrected to ensure the integrity and accuracy of the knowledge graph. As shown in Figure 4 .

[0028] Domain ontology design is primarily used for formal modeling of shared knowledge within the testing domain. This process lays the foundation for knowledge graph construction through abstract descriptions of concepts, attributes, hierarchies, and relationships. The main data sources for testing scenarios include manually written documents such as test manuals, engineering specifications, and expert experience; these documents are characterized by their knowledge density and high degree of specialization. The focus of ontology construction is to extract key concepts and their hierarchical structures from these data sources, clarifying attribute features and relationships, thereby providing semantic support for subsequent knowledge mapping and reasoning.

[0029] In terms of functional design, the knowledge graph design module provides users with visual tools for defining entities and relationships, supporting the intuitive construction and adjustment of the knowledge graph structure. Specifically, it includes: Entity type design Users can create entity types through the system interface, setting entity names, disambiguation markers, and visual colors for categorized management and display. The system allows users to view and edit entity type information on the entity type and relationship type design page, and add attributes (including attribute names and data types) to complete the entity description.

[0030] Relationship type design Users can define relationship types, including relationship name, starting entity type, and ending entity type. The relationship design page provides editing capabilities for existing relationships, supporting adjustments to relationship names and their associated entity types; it also allows adding new relationship types to expand the semantic relationships between entities in the graph.

[0031] Through the above functions, the knowledge graph design module can efficiently support ontological modeling of domain knowledge and form a collaborative working mechanism with the knowledge mapping module to realize the transformation from abstract knowledge to structured knowledge graphs, thereby improving the organizeability, retrieval, and reasonability of the test domain knowledge.

[0032] Step 4: Knowledge Mapping After completing the entity and relation design for the testing domain, this invention proposes a knowledge extraction and mapping method for the testing domain, which can efficiently convert semi-structured and unstructured evaluation data into formatted knowledge triples. Addressing the issues of scarce testing domain data, annotation relying on professional knowledge, and the high time consumption of manual annotation, the knowledge extraction function adopts a combination of manual and automatic annotation. First, domain experts perform high-quality manual annotation of representative business data based on the entity and relation types designed according to the domain ontology, ensuring the accuracy and completeness of the data. Then, using the model training function in the algorithm management module, the expert-annotated data is used as the training set, and combined with transfer learning methods, a domain knowledge extraction model is constructed to achieve efficient automatic annotation.

[0033] For high-quality structured test data, the application designs a knowledge mapping function to directly map structured data to the target knowledge graph. The function first converts the original data format to the format required by the knowledge graph, then aligns the entities to ensure consistency with the entities in the graph, and finally maps the relationships in the data to the relationship types of the knowledge graph, thereby ensuring the accuracy and completeness of the relationships.

[0034] Through the synergistic effect of the above knowledge extraction and knowledge mapping technologies, the application can construct a comprehensive and accurate test field knowledge graph, providing reliable semantic and data support for subsequent intelligent analysis, test design, and defect tracking applications based on the graph.

[0035] Knowledge extraction The test field contains a large amount of unstructured data, and extracting knowledge triples from these field data is an important task for building a field knowledge graph. Considering the professional nature of the knowledge in this field and the scarcity of data, both manual extraction and automatic extraction techniques are studied.

[0036] The application proposes a knowledge triple construction method combining manual extraction and automatic extraction for the construction of a field knowledge graph, considering the large amount of unstructured data in the test field, the strong professional nature of the field knowledge, and the scarcity of data.

[0037] In terms of manual extraction, domain experts label entities and relationships in representative documents based on a pre-designed domain ontology. To improve labeling efficiency, the application designs a visual labeling interface: users can select text fragments with a cursor and specify entity types; relationship labeling is based on labeled entities, selecting two target entities and specifying the relationship type that meets the ontology constraints, thereby reducing labeling errors and improving operational convenience.

[0038] In terms of automatic extraction, high-quality labeled data obtained in the manual extraction phase is used to train a knowledge extraction model to achieve large-scale automatic labeling. Entity recognition uses a named entity recognition technology based on a pre-trained model: a bidirectional LSTM is introduced on the basis of the BERT model to obtain sequence features, and CRF is used for sequence labeling, while a domain corpus and a domain term dictionary are used to optimize the model effect; through bootstrapping iteration and remote supervision learning, the constructed knowledge graph is used to continuously generate training data and optimize recognition results, thereby improving recognition quality under difficult data construction conditions.

[0039] In relation extraction, the application considers both template matching and deep learning methods. For non-template relations, a relation extraction model based on a convolutional neural network is used, and is extended to a distant supervision dataset. To reduce the noise introduced by distant supervision, a sentence-level attention mechanism is introduced, which assigns weights to sentences based on specific relations, so that effective sentences have higher weights and noise sentences have lower weights, thereby improving the accuracy of relation extraction.

[0040] Through the combination of the above artificial extraction and automatic extraction technologies, the application can efficiently and accurately extract entities and relations from unstructured data in the test field to form high-quality knowledge triples, providing a reliable data foundation for test field knowledge graph construction and subsequent applications.

[0041] Knowledge mapping The application proposes a multi-strategy knowledge extraction and mapping method. For different forms of unstructured data, corresponding knowledge extraction strategies are used to efficiently convert data to information and information to knowledge. For structured data in the test field that has been constructed, a knowledge mapping method combining heuristic rules and distant supervision is designed to quickly import structured data into the knowledge graph.

[0042] Specifically, the knowledge mapping module parses the structured test data output by the data processing module based on a heuristic rule library, and automatically compares the data table headers or column names with the entity types and relationship types defined in the knowledge graph design module through a keyword similarity matching algorithm. The system supports users to manually confirm or adjust the mapping relationship in the visual interface, and generates a mapping table of entities and relationships. After mapping is completed, the data will be automatically imported into the entity list and relationship list of the test field knowledge graph, realizing batch and one-key knowledge import, ensuring the accuracy and efficiency of mapping.

[0043] Step 5: Data application Knowledge graph visualization After constructing the entity and relationship lists of the test field knowledge graph, to facilitate users to intuitively understand the complex knowledge graph, the application realizes the visualization function of the knowledge graph based on the high-performance graph database management system Neo4j, and intuitively displays the relationship between entities in the form of visual nodes and edges. This function uses the graph structure (nodes, relationships and attributes) supported by Neo4j to store and represent data, helping users quickly understand the connection, interaction and dependency relationship between different equipment.

[0044] In a specific implementation, for the test field knowledge graph, after the user fully understands the test related device types, performance parameters, test results and related operation procedures, the user can create device, test and parameter node types and connect them through relationships such as performing a test or having a parameter. After completing the knowledge graph mode design and data import, the system supports multiple quick and convenient graph retrieval and display through a predefined query instruction set, allows the user to visually browse entity nodes, relationship connections and attribute details in a graphical manner. The visualization function significantly improves the cognitive efficiency of the user on the complex relationship network, enabling the user to quickly understand the connection mode, interaction mechanism and dependency relationship between entities. In addition, the module also integrates advanced interaction functions such as view adjustment (such as zooming, panning, filtering), accurate search positioning of specific nodes or relationships, and exporting the current view into multiple standard formats (such as PNG, SVG, JSON), thereby providing a powerful visualization support tool for the depth analysis and application of the knowledge graph.

[0045] Knowledge graph question answering On the basis of the completed constructed knowledge graph, in order to efficiently realize the knowledge question answering function, the application designs an intelligent question answering system based on a knowledge graph. The system takes the test field knowledge graph as the core knowledge base, analyzes the natural language question input by the user through a natural language processing technology, accurately identifies the entities, relationships and semantic intentions contained therein. The system automatically converts the parsed query condition into a structured graph query statement (such as Cypher), and performs efficient retrieval and reasoning in the knowledge graph, locates the entity nodes and related relationship paths matched with the question. The retrieval result is processed by a semantic analysis and answer generation module, and finally a precise answer conforming to the natural language expression habit is generated. The system supports multi-round interactive question answering, can dynamically adjust the query strategy based on the context, and can mine the implicit knowledge by means of the graph reasoning mechanism (such as path traversal, relationship transmission, etc.), thereby significantly improving the ability to answer complex questions of the user.

[0046] The application provides a knowledge graph visualization tool construction method and system, which is composed of five parts, i.e., data processing for multi-source data, algorithm management for field data, graph design for field knowledge, extraction and mapping based on field data, and data application based on field data, and the modules cooperate with each other to improve the data processing efficiency, knowledge extraction accuracy, dynamic updating capability and visualization interactivity, and meet the needs of users in data analysis and mining.

[0047] The above only describes the preferred embodiments of the application, and it should be noted that, for those skilled in the art, without departing from the technical principles of the application, a number of improvements and modifications can be made, and these improvements and modifications should also be considered as the protection scope of the application.

Claims

1. A knowledge graph visualization tool construction system, characterized in that, The system includes: a data processing module, an algorithm management module, a graph design module, an extraction and mapping module, and a data application module; The data processing module is used to acquire and process multi-source data required for building the knowledge graph, including structured, semi-structured, and unstructured data. First, it uses a unified data import method to efficiently acquire data from databases, file systems, and network interfaces. Then, it performs data cleaning and verification. The algorithm management module is used to comprehensively manage the algorithm models in the knowledge graph visualization tool, covering knowledge extraction, data cleaning, and knowledge mapping models; it provides users with flexible model management options, including model training, model testing, and model deployment, and supports efficient configuration and control of each stage; The knowledge graph design module is used for the structured design of knowledge graphs. It performs unified domain ontology modeling for concepts, attributes, and concept levels in the test domain, clarifies the entity types and their relationships within the domain, and supplements, verifies, and corrects the constructed domain ontology based on business data, thereby ensuring the integrity and accuracy of the knowledge graph. The extraction and mapping module combines manual and automatic annotation methods to extract entity, attribute, and relationship information from domain-related unstructured business documents. It also utilizes a data dictionary and disambiguation technology to ensure the accuracy and consistency of knowledge representation. Through efficient knowledge mapping technology, the extraction results are mapped to the designed domain ontology structure to generate a domain knowledge graph that meets the requirements. The module supports knowledge fusion and dynamic updates, and continuously optimizes the knowledge graph content. The data application module, based on the constructed domain knowledge graph, realizes two major functions: visualization and intelligent question answering. The visualization sub-module dynamically displays the graph topology and entity relationships, supporting interactive exploration. The question answering sub-module accurately generates answers through natural language parsing and graph reasoning, and supports users to conduct multiple rounds of interaction.

2. The knowledge graph visualization tool construction system as described in claim 1, characterized in that, During the data reading phase, the data processing module supports two access methods: local file access and external database access. Local file access can handle various file formats, efficiently reading tabular and thesaurus data for structured data, and extracting business documents and test report text content for unstructured data. External database access supports direct connection with mainstream database systems, enabling fast access and unified extraction of remote data resources.

3. The knowledge graph visualization tool construction system as described in claim 2, characterized in that, During the data cleaning phase, cleaning operations include deduplication to eliminate duplicate records and ensure data uniqueness, text normalization to identify and correct typos and improve text accuracy, and format standardization to standardize date formats and full-width / half-width characters, thereby eliminating format differences between different data sources. During the data verification phase, the cleaned data is quality verified to ensure that it meets the system's usage requirements. Data types, cleaning methods, and cleaning status are recorded, and inconsistent or abnormal data is automatically identified, marked, and described to facilitate subsequent manual review or automatic correction.

4. The knowledge graph visualization tool construction system as described in claim 1, characterized in that, Model training includes: users selecting different datasets and training devices according to their needs, and setting key training parameters to optimize model performance; at the same time, it supports multi-task training processes to meet the needs of knowledge extraction and mapping. Model testing includes: allowing users to select suitable test datasets and devices, and flexibly adjust test parameters to verify the model's effectiveness in knowledge cleaning and knowledge mapping scenarios; in addition, it supports multi-dimensional evaluation of results to ensure the accuracy and reliability of model performance. Model deployment includes: providing model deployment functionality, supporting users to flexibly configure deployment devices and target addresses, and deploying optimized models to knowledge graph visualization tools; it also provides version management and deployment record functionality to facilitate model iteration and maintenance.

5. The knowledge graph visualization tool construction system as described in claim 1, characterized in that, Domain ontology is used to formally model shared knowledge within the testing domain. This process lays the foundation for knowledge graph construction through the abstract description of concepts, attributes, levels, and relationships. Data sources for testing scenarios include test manuals, engineering specifications, and expert experience documents, which are characterized by their knowledge density and high degree of specialization. Key concepts and their hierarchical structures are extracted from the above data sources, and attribute features and relationships are clarified, thereby providing semantic support for subsequent knowledge mapping and reasoning.

6. The knowledge graph visualization tool construction system as described in claim 5, characterized in that, The knowledge graph design module provides users with visual tools for defining entities and relationships, supporting the intuitive construction and adjustment of knowledge graph structures; specifically including: Entity type design: Users create entity types through the system interface, setting entity names, disambiguation markers, and visual colors for categorization, management, and display; users view and edit entity type information on the entity type and relationship type design page, and add attributes to improve entity descriptions; Relationship type design: Users define relationship types, including relationship name, starting entity type, and ending entity type; the relationship design page provides the ability to edit existing relationships, supporting adjustments to relationship names and their associated entity types; it also allows the addition of new relationship types to expand the semantic associations between entities in the graph.

7. The knowledge graph visualization tool construction system as described in claim 1, characterized in that, Knowledge extraction includes: For manual extraction, domain experts annotate representative documents with entities and relationships based on a pre-designed domain ontology; a visual annotation interface is designed: users select text fragments with a cursor and specify entity types; relationship annotation is based on the annotated entities, selecting two target entities and specifying relationship types that conform to ontology constraints; For automatic extraction, a knowledge extraction model is trained using high-quality labeled data obtained in the manual extraction stage to achieve large-scale automatic labeling. For entity recognition, named entity recognition technology based on a pre-trained model is adopted: bidirectional LSTM is introduced on the basis of BERT model to obtain sequence features, and CRF is combined for sequence labeling. At the same time, domain corpus and domain terminology dictionary are used to optimize model performance. Through bootstrapping and remote supervised learning, training data is continuously generated and recognition results are optimized with the help of the constructed knowledge graph, thereby improving recognition quality under the condition of difficult data construction. In terms of relation extraction, both template matching and deep learning methods are considered. For non-template relations, a relation extraction model based on convolutional neural networks is adopted and extended to remote supervised datasets. A sentence-level attention mechanism is introduced to assign weights to sentences according to specific relations, so that effective sentences receive higher weights and noisy sentences receive lower weights, thereby improving the accuracy of relation extraction. By combining manual and automatic extraction techniques, entities and relationships are extracted from unstructured data in the testing domain to form high-quality knowledge triples.

8. The knowledge graph visualization tool construction system as described in claim 7, characterized in that, Knowledge mapping includes: Based on a heuristic rule base, the structured test data output by the data processing module is parsed, and the data table headers or column names are automatically compared with the entity types and relationship types defined in the knowledge graph design module using a keyword similarity matching algorithm. Users can manually confirm or adjust the mapping relationship in the visual interface to generate a mapping table of entities and relationships. After the mapping is completed, the entity list and relationship list of the test domain knowledge graph are automatically imported according to the mapping results, realizing batch and one-click knowledge import and ensuring the accuracy and efficiency of the mapping.

9. The knowledge graph visualization tool construction system as described in claim 1, characterized in that, Knowledge graph visualization includes: After constructing the entity and relationship list of the knowledge graph in the testing domain, the visualization function of the knowledge graph was implemented based on the high-performance graph database management system Neo4j, which intuitively displays the relationships between entities in the form of visualized nodes and edges; After completing the knowledge graph pattern design and data import, a variety of fast and convenient graph retrieval and display methods are achieved through a predefined set of query commands, allowing users to intuitively browse entity nodes, relationship connections and their attribute details in a graphical manner.

10. The knowledge graph visualization tool construction system as described in claim 1, characterized in that, Knowledge graph question answering includes: using a knowledge graph of the testing domain as the core knowledge base, parsing the natural language questions input by users through natural language processing technology, accurately identifying the entities, relationships, and semantic intent contained therein; converting the parsed query conditions into structured graph query statements, and performing efficient retrieval and reasoning in the knowledge graph to locate entity nodes and related relationship paths that match the question; the retrieval results are processed by the semantic parsing and answer generation module to finally generate accurate answers that conform to the habits of natural language expression.

Images