Devices, data structures, and computer-implemented methods for constructing knowledge graph triples.

Abstract

This invention provides a device, data structure, and computer-implemented method for constructing triples of a knowledge graph. [Solution] The method automates the construction of a knowledge graph using instructions from human experts or automates the construction of a knowledge graph using automated machine validation for syntactic completeness and consistency without instructions from human experts. This method can utilize existing ontologs and expert knowledge beforehand, which further enhances the suitability of the constructed knowledge graph for specific uses.

Description

Technical Field

[0001] The present invention relates to an apparatus, a data structure, and a computer-implemented method for constructing triples of a knowledge graph.

Summary of the Invention

Means for Solving the Problems

[0002] A computer-implemented method for constructing triples of a knowledge graph, wherein the method includes providing input data comprising items, properties of the items, a header for the item, and a header for the item's properties; depending on the input data, determining a semantic description of the headers; depending on the semantic description, determining an ontology that defines the properties for the item's property headers; depending on the semantic description and the ontology, determining a mapping specification that defines the relationship between the item's header and the item's property header; depending on the mapping specification and the input data, constructing triples comprising the relationship between the item and its properties, wherein the method determines the semantic description by instructing a first large-scale language model to output an output that semantically describes the input data, in particular by instructing a human expert or an automated validation by machine to review the output that semantically describes the input data; receiving the review results of the output that semantically describes the input data; and determining the semantic description of the input data The method is characterized by including at least one of the following: determining a semantic description depending on the result of reviewing an output that describes the input data semantically and an output that describes the input data semantically; determining an ontology by instructing a first or second large-scale language model to output an output that describes the semantic description ontologically, in particular by instructing a human expert or an automated machine to review the output that describes the semantic description ontologically; receiving the review result of the output that describes the semantic description ontologically; determining an ontology depending on the output that describes the semantic description ontologically and the review result of the output that describes the semantic description ontologically; and determining a mapping specification by instructing a first, second or third large-scale language model to output an output mapping for the semantic description and ontology, in particular by instructing a human expert or an automated machine to review the output mapping; receiving the review result or the output mapping; and determining an ontology depending on the output mapping and the review result of the output mapping.

[0003] This method automates the construction of a knowledge graph using instructions from human experts, or automates the construction of a knowledge graph using automated machine validation for syntactic completeness and consistency without instructions from human experts.

[0004] This method proceeds step-by-step through multiple work steps in building a knowledge graph. An exemplary set of steps includes determining semantic descriptions, determining ontologities, determining mappings, and executing the mappings. At each step, users, such as human experts or automated machine validation, can review and / or modify intermediate results—i.e., semantic descriptions, ontologities, and mappings—to refine the output of the large-scale language model. Users can, for example, provide existing ontologities as additional input, on which this method constructs novel, extended ontologities.

[0005] Knowledge graphs can contain complex and heterogeneous data. Knowledge graphs are based on semantic technology; that is, they describe data in a semantic language that is uniquely interpretable by humans and machines. This semantic language may be standardized. Knowledge graphs support interoperability and knowledge sharing. Knowledge graphs are configured for the storage and discovery of highly linked data. Knowledge graphs support multi-hop inference.

[0006] This method is based on a formal language and semantic content. The ontology is based on an ontology language, such as the Web Ontology Language (OWL). The mapping specification is based on an RDF mapping language, such as RML, which is built on the W3C standard RDF (Resource Description Framework). The semantic description and / or mapping specification may be based on a constraint language, such as the Shapes Constraint Language (SHACL).

[0007] A knowledge graph contains knowledge graph data. Validation tasks may be performed on the knowledge graph data. Validation tasks may include completeness checks and / or consistency checks and / or standards compliance verification. Knowledge graphs provide a foundation for searching, discovering, and analyzing complex data, enabling data analysis and decision-making in application areas such as finance, supply chain management, healthcare, and biotechnology.

[0008] A general advantage of employing large-scale language models in the knowledge graph construction process is that the considerable time and effort required from (skilled) human experts can be minimized when entrusted to the proposed method.

[0009] The mapping is a machine-readable intermediate result, which reduces the complexity of automated verification and validation applied to ensure the completeness and accuracy of the generated graph.

[0010] The use of ontologs helps generate better results for specialized fields such as manufacturing. This method allows for the prior use of existing ontologs and expert knowledge, which further enhances the suitability of the constructed knowledge graph for specific applications.

[0011] Providing input data may include providing a table containing columns and rows, where a header for items identifies the columns containing the items, a header for properties identifies the columns containing the properties, and the table contains items and properties within the same row.

[0012] Determining the semantic description of a header may include determining the structured output that associates the item header with the item description, the item's semantic content, and the item's data type, and determining the structured output that associates the property header with the property description, the property's semantic content, and the property's data type.

[0013] Determining an ontology may involve determining the properties for the item's property headers so that the item's property headers are included as labels for the properties for the item's property headers.

[0014] Determining a mapping specification that defines the relationship between a header for an item and a header for an item's properties may include providing a first mapping for determining the subject of a triple, the first mapping including a template containing the item's header, and providing a second mapping for determining the predicate and object of a triple, the second mapping including the relationship and a template containing the item's property header.

[0015] Constructing a triple may involve providing an item as the subject of the triple, providing a relation as the predicate of the triple, and determining the object of the triple by finding a first mapping depending on the item's header, finding a second mapping depending on the relation, and finding the item's property as the object depending on the second mapping as defined by the mapping specification.

[0016] The method includes providing a property associated with a header of an item's properties, the property including the item's properties, and providing a property associated with a different header, and relying on a second mapping to find an item's property as an object may include determining an instruction to find a property associated with a header of an item's properties, relying on a second mapping, in particular determining an instruction to search for an item's property only within the properties associated with the header of an item's properties in the second mapping.

[0017] Determining an ontology may include determining a class definition that includes a label for the class, and determining the properties for the item's property headers so that they include the label for the class.

[0018] Determining an ontology may involve determining the properties for the header of an item's properties, such that they include the range and data type of a second property.

[0019] A stepwise approach allows for better scaling to larger datasets. Intermediate results may be shared, examined, and, in some cases, redescribed using, for example, a large language model or another large language model. This facilitates efficiency when handling complex data, such as better scaling and cost reduction.

[0020] This method may involve instructing a large-scale language model to generate mappings for triples, instead of instructing it to directly convert all data into triples. Using mappings allows for easy scaling of the knowledge graph while minimizing costs, for example, without the need to send the complete dataset to the large-scale language model, which would otherwise be costly and inefficient.

[0021] This method may include validating a knowledge graph containing the constructed triples, and if the validation of the knowledge graph is successful, constructing another triple in the knowledge graph depending on the inputs, ontology, semantic description, and mapping specifications determined when the constructed triples were constructed; otherwise, deleting the constructed triples and the ontology, semantic description, and mapping specifications determined for the constructed triples before determining another triple depending on the inputs.

[0022] The generated mapping may be used with the input, or with new data having a similar structure to the data used to generate the mapping, without the need to reuse the large-scale language model. In particular, the new data subsequently read using the mapping may include highly sensitive and protected data, but in this method, there is no need to transmit it to the large-scale language model.

[0023] This method may involve constructing another triple in the knowledge graph depending on other inputs, ontologities, semantic descriptions, and mapping specifications determined during the construction of a constructed triple, or determining multiple triples depending on the mapping specifications. By constructing triples from existing mappings, the time required for knowledge graph construction is reduced and scalability is improved.

[0024] A device for constructing a triple of a knowledge graph comprises at least one processor and at least one memory, the memory of which stores instructions executable by at least one processor, which cause the device to carry out the method when executed by at least one processor.

[0025] A computer program for constructing a knowledge graph triple, when executed by a computer, includes computer-readable instructions to cause that computer to perform this method.

[0026] A data structure for constructing a triple in a knowledge graph includes at least one data field for input data, which includes an item, properties of the item, a header for the item, and a header for the item's properties; the data structure includes at least one data field for a semantic description of the header, which is determined depending on the input data and in particular using a large-scale language model; the data structure includes at least one data field for an ontology that defines the properties for the item's property header, which is determined depending on the semantic description and in particular using a large-scale language model; the data structure includes at least one data field for a mapping specification that defines the relationship between the header for the item and the header for the item's properties, which is determined depending on the semantic description and the ontology and in particular using a large-scale language model; and the data structure includes at least one data field for a triple, which includes the relationship between the item and the property, which is constructed depending on the mapping specification.

[0027] Further exemplary embodiments are derived from the following description and the drawings.

Brief Description of the Drawings

[0028] [Figure 1] It is a schematic diagram showing an apparatus for constructing triples of a knowledge graph. [Figure 2] It is a flowchart including steps of a method for constructing triples of a knowledge graph. [Figure 3] It is a schematic diagram showing a data structure for constructing triples of a knowledge graph.

Modes for Carrying Out the Invention

[0029] FIG. 1 schematically shows an apparatus 100 for constructing triples of a knowledge graph.

[0030] A knowledge graph is a set of facts also referred to as triples in the form of <subject, predicate, object>. The predicate defines the relationship between the subject and the object.

[0031] The apparatus 100 includes at least one processor 102 and at least one memory 104. [[ID=XX]]

[0032] [[ID=XX]] When at least one memory 104 is executed by at least one processor 102, it stores instructions executable by at least one processor 102 for causing the apparatus 100 to implement a method for constructing triples of a knowledge graph.

[0033] This method is described using an example of a large language model. This large language model is configured to understand and summarize multiple forms of data. The large language model is configured to be instructed to generate an output in a specific format.

[0034] Note: There are some consecutive lines with only "<0000xxx>" tags which seem to be placeholders without specific content. I've left them as they are in the translation. Also, in the original text, lines 34 - 36 and lines 38 - 40 have only "<0000xxx>" tags, and in the translation, I've added "XX" tags for those lines to maintain the line numbering consistency as required. If there's any specific meaning or correction needed for those lines, please let me know.This method utilizes the features of a large-scale language model by providing it with input data and questions (so-called prompts), thereby extracting specific information about the input data in a structured format.

[0035] Examples of large-scale language models include the generative pre-trained transformer model by Liu, Y., Han, T., Ma, S., Zhang, J., Yang, Y., Tian, ​​J., ... and Ge, B. (2023). An example of a large-scale language model is presented in "Summary of ChatGPT-related research and future prospects for large-scale language models" (Meta-Radiology, 100017).

[0036] This method is explained using the example of an ontology. An ontology is a formal and explicit description of a concept (also called a class) in a given field, consisting of properties (also called relations) of each concept that describe the various characteristics and attributes of the concept, as well as constraints on the concepts and properties.

[0037] "Ontology Development 101: A Guide to Creating Your First Ontology" by Noy, Natalya F. and McGuinness, Deborah L. provides an example for developing an ontology.

[0038] This method will be explained using an exemplary triple construction example. In knowledge graph construction, the triples of the knowledge graph are constructed as described for the exemplary triple.

[0039] Knowledge graph construction is the process of building a knowledge graph from input data originating from one or different data sources, such as text or structured data. Knowledge graph construction is a time-consuming process that requires an understanding of both semantic web technologies and the target domain. In an exemplary setup, relevant data (tabular data) is collected from different systems and used as input for building an integrated knowledge graph.

[0040] The following are exemplary steps for building a knowledge graph: (1) Step of creating an ontology for the domain of input data (2) Step of creating a mapping from tabular data to such ontology (3) The step of incorporating tabular data into a knowledge graph via such a mapping.

[0041] This method is explained using an example of input data provided as a table, i.e., tabular data. This method is similarly applicable to other structured or semi-structured input data, such as key-value pairs.

[0042] This method allows for the optional use of additional text information regarding the input data within the prompt.

[0043] This method assumes the existence of a large-scale language model and access to that large-scale language model.

[0044] This method determines a triple that includes a subject, predicate, and object.

[0045] This method includes step 202.

[0046] Step 202 includes the step of providing input data.

[0047] The input data includes tables.

[0048] The table is provided for the class. The table includes columns and rows.

[0049] Each column, within the first row of the table, includes its respective header. The other rows of the table correspond to each instance of the class.

[0050] One column contains the item header in the first row and the item associated with the item header in the second row. Another column contains the item property header in the first row and the item property associated with the item property header in the second row.

[0051] In this example, the item represents the subject, the item's property represents the object, and the column header containing the item's property represents the relationship.

[0052] This means that a header for an item identifies the column containing the item. A header for a property identifies the column containing the property. A table contains both items and properties within the same entity.

[0053] A table can contain multiple rows for different instances of a class.

[0054] A table may contain multiple columns for different properties, with these columns containing the header for each property in the first row and the respective property for each instance in the other rows.

[0055] This means that the input data includes fields, field properties, headers for the fields, and headers for the field properties.

[0056] This means that the input data includes multiple properties associated with the header of the item's properties. The input data includes the item's properties and multiple properties associated with different headers.

[0057] For example, the table might contain the production item number in the first column and the temperature value as a property in the second column. The header for the item is "Production Item". The header for the property is "Temperature".

[0058] This method includes step 204.

[0059] Step 204 includes the step of determining the semantic description of the header, depending on the input data.

[0060] For example, the semantic description of a header is determined using a large-scale language model, depending on the input data and on prompts that request the large-scale language model to output a semantic description of the header, again depending on the input data.

[0061] The step of determining the semantic description of the header may include determining the structured output.

[0062] For example, structured output associates item headers with item descriptions, item meanings, and item data types.

[0063] Large-scale language models require, for example, that semantic descriptions be output as structured output that associates item headers with item descriptions, item semantics, and item data types.

[0064] For example, structured output associates property headers with property descriptions, property meanings, and property data types.

[0065] Large-scale language models require, for example, that semantic descriptions be output as structured output that associates property headers with property descriptions, property semantics, and property data types.

[0066] Given input data, such as a table, a portion of the input data may be sampled and included in a prompt for a large-scale language model to detect the data model and generate a semantic description of that data model. Optionally, a text description of the input data may also be provided within the prompt. The prompt is designed to return structured output in a format defined within the prompt.

[0067] If the example table includes production items and temperature, the semantic description might include, for example, the following: “Production Item”: { “description”: “The Number of the production item” “semantics”: “Number” “datatype”: “integer” } “Temperature”: { “description”: “The temperature of the production item” “semantics”: “Temperature” “datatype”: “integer” } It includes.

[0068] Data types are often different; for example, there is a "string".

[0069] Step 204 may include a step of modifying the semantic description.

[0070] For example, step 204 is the following step: - A step of instructing a large-scale language model to output an output that semantically describes the input data, - In particular, a step of instructing a human expert or an automated machine to review the output which semantically describes the input data, - A step of receiving the review result of the output which semantically describes the input data, - A step of determining the semantic description based on the output that semantically describes the input data and the review result of the output that semantically describes the input data, Includes.

[0071] The output, which semantically describes the input data, is a semantic description that needs to be corrected.

[0072] The review results allow us to verify the output, which semantically describes the input data, as a semantic description. The review results may also include, as a semantic description, any changes made to the output, which semantically describes the input data, during the review process.

[0073] This method includes step 206.

[0074] Step 206 includes determining an ontology that, depending on the semantic description, defines a first property for the item header and a second property for the item property header.

[0075] For example, the ontology is determined using a large-scale language model, depending on the semantic description, and depending on prompts that request the large-scale language model to output the ontology, also depending on the semantic description.

[0076] For example, the first property includes the item header as the label for the first property. A large language model might require, for instance, that the item header be output as the label for the first property.

[0077] For example, the second property includes the header of the item's property as the label of the second property. A large-scale language model might require, for instance, that the property header be output as the label of the second property.

[0078] An ontology may include class definitions that include labels for classes. The first and / or second properties may include labels for classes. A large-scale language model may require that an ontology be provided that has class definitions that include labels for classes. A large-scale language model may require that the first and / or second properties be provided to include labels for classes.

[0079] The first property may include the range and data type of the first property. A large-scale language model may require that a first property be provided, having the range and data type of the first property. A second property may include the range and data type of the second property. A large-scale language model may require that a second property be provided, having the range and data type of the second property.

[0080] For example, the semantic description is passed back to a large language model with prompts to construct an OWL ontology that has classes and relationships for the data within the semantic description. For more focused ontology creation, a text description of how the input data is used may be added to the prompts. The result of this step is an ontology in OWL format.

[0081] Since the results from large-scale language models may be incomplete and / or contain errors, a refinement step, i.e., ontology refinement, may be performed within a loop. This refinement loop may be repeated until the resulting ontology meets predetermined criteria.

[0082] If the exemplary table includes production items and temperature, the ontology would include, for example, the following: #Class definition <production>rdf:type rdfs:Class ; rdfs:label "Production” . #Property definition <Production Item> rdf:type rdf:Property ; rdfs:label " Production Item ”; rdfs:domain <production>; rdfs:range xsd:string . < Temperature > rdf:type rdf:Property ; rdfs:label " Temperature”; rdfs:domain <production>; rdfs:range xsd:string . It includes.

[0083] Step 206 may include a step of modifying the ontology.

[0084] For example, step 206 is the following step: - A step of instructing a large-scale language model to output an output that describes semantic descriptions into an ontological manner, - In particular, a step of instructing a human expert or an automated machine to review the output that ontologically describes the semantic description, - A step of receiving the review result of the output which describes the semantic description intotonically, - A step of determining the ontology based on the output that describes the semantic description ontologically and the review result of the output that describes the semantic description ontologically, Includes.

[0085] Output that describes semantic descriptions into an ontology is an ontology that should be modified.

[0086] The review results allow us to verify the output, which describes the semantic description ontologically, as an ontology. The review results may include changes made during the review of the output that describes the semantic description ontologically, as part of the ontology.

[0087] This method includes step 208.

[0088] Step 208 includes the step of determining the mapping specification based on the semantic description and ontology.

[0089] The mapping specification defines the relationship between a header for an item and a header for the item's properties.

[0090] For example, the mapping specification is determined using a large-scale language model, depending on semantic descriptions and ontologities, and depending on prompts that request the large-scale language model to output the mapping specification, depending on semantic descriptions and ontologities.

[0091] A mapping specification includes, for example, a first mapping for determining the subject of a triple. The first mapping includes, for example, a template containing item headers. A large-scale language model is required to output, for example, a first mapping having a template containing item headers.

[0092] A mapping specification includes, for example, a second mapping for determining the predicate and object of a triple. The second mapping includes, for example, a relation and a template containing headers for the item's properties. A large-scale language model is required to output, for example, a second mapping having a template containing headers for the item's properties.

[0093] Given an ontology and semantic description, a large-scale language model may be asked to generate a mapping specification in the RDF Mapping Language (RML) manner. RML is described in "RML: A Generic Language for Integrated RDF Mappings of Heterogeneous Data" by A. Dimou, M. Vander Sande, P. Colpaert, R. Verborgh, E. Mannens, and R. Van de Walle, Proceedings of the 7th Workshop on Linked Data on the Web, volume 1184 of CEUR Workshop Proceedings. CEUR, 2014.

[0094] A large-scale language model may be questioned to transform data into a graph according to an ontology. The method may include a step to automatically check the generated mapping specification for errors and completeness. The method may include an elaboration step, i.e., mapping validity checking, to guide the large-scale language model to generate a proper mapping specification.

[0095] If the example table includes production items and temperature, the mapping specification would include, for example, the following: rr:subjectMap [ rr:template "http: / / example.com / Production / {Production Item}"; rr:class <production>] rr:predicateObjectMap [ rr:predicate <temperature>; rr:objectMap [rr:column "Temperature"] ] It includes.

[0096] Step 208 may include a step to modify the mapping specification.

[0097] For example, step 208 is the following step: - A step of instructing a large-scale language model to output output maps for semantic descriptions and ontologities, - In particular, the step of instructing a human expert or an automated validation by machine to review the output map, - A step of receiving the review result or output map, - A step in which the ontology is determined depending on the output map and the review result of the output map, Includes.

[0098] The output mappings for semantic descriptions and ontologities are mapping specifications that need to be modified.

[0099] The review results allow us to verify the output maps for semantic descriptions and ontologities as mapping specifications. The review results may include changes made to the output maps for semantic descriptions and ontologities during the review process as part of the mapping specifications.

[0100] This method includes step 210.

[0101] Step 210 involves constructing a triple that includes the relationship between items and properties, depending on the mapping specification.

[0102] For example, a triple is constructed depending on the mapping specification and input data, and also depending on prompts that request a large-scale language model to output a triple, depending on the mapping specification and input data.

[0103] For example, an item is provided as the subject of a triple.

[0104] For example, the relation is provided as a predicate of a triple.

[0105] For example, the object of a triple is determined by finding a first mapping depending on the item's header, finding a second mapping depending on the relation, and finding the item's property as the object depending on the second mapping.

[0106] For example, in order to find an item's property as an object, depending on the second mapping, an instruction is determined to find the property associated with the header of the item's property, depending on the second mapping. This instruction specifically instructs to search for the item's property only within the properties associated with the header of the item's property in the second mapping.

[0107] For example, large-scale language models are required using instructions that, in particular, rely on the headers of items to find a first mapping, rely on relations to find a second mapping, and rely on the second mapping to find the properties of items as objects.

[0108] Figure 3 shows an exemplary data structure 300 for constructing a knowledge graph triple.

[0109] This data structure 300 includes input data, semantic description, ontology, mapping specification, and at least one data field 302 for triples.< / temperature> < / production> < / production> < / production> < / production>

Claims

1. In a computer-implemented method for constructing triples of a knowledge graph, The aforementioned method, (202) Provide input data including an item, properties of the item, a header for the item, and a header for the properties of the item. Includes, The aforementioned method, (204) Determining the semantic description of the header based on the input data, (206) Determining an ontology that defines properties for the property header of the item, depending on the semantic description, (208) Determining a mapping specification that defines the relationship between the header for the item and the header for the property of the item, depending on the semantic description and the ontology, Includes, The aforementioned method, (210) Constructing a triple that includes the relationship between the item and the property, depending on the mapping specification and the input data. Includes, The aforementioned method, The semantic description is determined by instructing the first large-scale language model to output an output that semantically describes the input data. In particular, instructing human experts or automated machine validation to review the output that semantically describes the input data, Receiving the review result of the output, which semantically describes the input data. The semantic description is determined based on the output that semantically describes the input data and the review result of the output that semantically describes the input data. and, The ontology is determined by instructing the first or second large-scale language model to output an output that describes the semantic description ontologically. In particular, instructing a human expert or an automated machine to review the output that ontologically describes the aforementioned semantic description, To receive the review result of the output that describes the aforementioned semantic description into an ontological manner, The ontology is determined based on the output that describes the semantic description ontologically and the review result of the output that describes the semantic description ontologically. and, The mapping specification is determined by instructing a first, second, or third large-scale language model to output an output mapping for the semantic description and the ontology. In particular, instructing a human expert or an automated machine to review the output map, Receiving the aforementioned review results or the aforementioned output map, Determining the ontology depending on the output map and the review result of the output map, Including at least one of the following: A method characterized by the following:

2. Providing the aforementioned input data (202) is, This includes providing a table that includes columns and rows, The header for the item identifies the column containing the item, the header for the property identifies the column containing the property, and the table includes the item and the property in the same row. The method according to claim 1.

3. Determining the semantic description of the header (204) is, Determine a structured output that associates the header of the item with the item's description, semantic content, and data type. Determine the structured output that associates the header of the property with the property description, the meaning of the property, and the data type of the property. including, The method according to claim 1 or 2.

4. Determining the aforementioned ontology (206) The property for the header of the property of the item is determined to be included as the label for the property for the header of the property of the item, The method according to any one of claims 1 to 3.

5. Determining a mapping specification that defines the relationship between the header for the item and the header for the properties of the item (208) The present invention provides a first mapping for determining the subject of the triple, wherein the first mapping includes a template that includes the header of the item. The present invention provides a second mapping for determining the predicate and object of the triple, wherein the second mapping includes the relation and a template including a header for the properties of the item. including, The method according to claim 4.

6. Constructing the aforementioned triple (210) is, The above item is provided as the subject of the above triple, The above relationship is provided as a predicate of the triple, The object of the triple is determined by finding a first mapping depending on the header of the item, finding a second mapping depending on the relation, and finding the property of the item as the object depending on the second mapping as defined by the mapping specification, including, The method according to claim 5.

7. The aforementioned method, (202) To provide a property associated with the header of the property of the item, wherein the property includes the property of the item, Providing properties associated with different headers (202), Includes, Finding the property of the item as the object by relying on the second mapping (210) The process involves determining an instruction to find a property associated with the header of the item's properties, depending on the second mapping, and in particular determining an instruction to search for the item's properties only within the properties associated with the header of the item's properties in the second mapping. The method according to claim 6.

8. Determining the aforementioned ontology (206) Determining a class definition that includes a label for the class, The property for the header of the property of the aforementioned item is determined to include the label for the class, including, The method according to any one of claims 4 to 7.

9. Determining the aforementioned ontology (206) This includes determining the properties for the header of the property of the aforementioned item such that they include the range and data type of the second property, The method according to any one of claims 4 to 8.

10. The aforementioned method, Validating the knowledge graph, which includes the constructed triples, If the validation of the knowledge graph is successful, another triple in the knowledge graph is constructed, depending on the input, ontology, semantic description, and mapping specification determined when the constructed triple was constructed; otherwise, the constructed triple and the ontology, semantic description, and mapping specification determined for that constructed triple are deleted before determining another triple depending on the input. including, The method according to any one of claims 1 to 9.

11. The aforementioned method, Constructing another triple in the knowledge graph, depending on different inputs, ontologities, semantic descriptions, and mapping specifications determined during the construction of the constructed triple. Or, Determining multiple triples depending on the aforementioned mapping specification. including, The method according to any one of claims 1 to 10.

12. A device (100) for constructing a triple of a knowledge graph, The aforementioned device is At least one processor (102), At least one memory (104), Equipped with, The at least one memory (104) stores instructions that can be executed by the at least one processor (102) to cause the device (100) to carry out the method according to any one of claims 1 to 11 when executed by the at least one processor (102). A device characterized by the following features.

13. A computer program for constructing triples of a knowledge graph, The computer program is characterized in that, when executed by a computer, it includes computer-readable instructions causing the computer to carry out the method described in any one of claims 1 to 11.

14. In the data structure (300) for constructing a triple of the knowledge graph, The aforementioned data structure (300) is Includes at least one data field (302) for input data, which includes an item, properties of the item, a header for the item, and a header for the properties of the item, The aforementioned data structure (300) is It includes at least one data field (302) for the semantic description of the header, which is determined depending on the input data, particularly using a large-scale language model, The aforementioned data structure (300) is Includes at least one data field (302) for an ontology that defines properties for the property header of the item, which are determined depending on the semantic description, particularly using a large-scale language model, The aforementioned data structure (300) is A mapping specification that defines the relationship between a header for the item and a header for the properties of the item, comprising at least one data field (302) for the mapping specification, which is determined depending on the semantic description and the ontology, particularly using a large-scale language model, The aforementioned data structure (300) is Includes at least one data field (302) for a triple that includes the relationship between the item and the property, which is constructed depending on the mapping specification, A data structure (300) characterized by the following.

Images