Device, data structure and computer-implemented method for constructing a triple of a knowledge graph

Abstract

A device, a data structure, and a procedure are described for the automated construction (210) of triples for knowledge graphs. Starting from input data (item, property, headers for item and property), a semantic description of the headers, (204), a suitable ontology, (206), and a mapping specification, (208) which defines the relationship between the headers, are determined using large language models (LLMs). The resulting triple represents the relationship between item and property. Human experts or automated validation processes can check the outputs generated by the LLMs and thus improve the quality of the triple construction.

Description

State of the art

[0001] The invention relates to a device, a data structure and a computer-implemented method for constructing a triple of a knowledge graph. Description of the invention

[0002] A computer-implemented method for constructing a triple of a knowledge graph, characterized in that the method comprises providing input data, wherein the input data comprises an item and a property of the item, and a header for the item and a header for the property of the item; wherein the method comprises determining a semantic description of the headers depending on the input data; determining an ontology that defines a property for the header of the item's property, depending on the semantic description; determining a mapping specification depending on the semantic description and the ontology, wherein the mapping specification defines a relationship between the header for the item and the header for the item's property; and wherein the method comprises constructing the triple that includes the relationship between the item and the property.depending on the mapping specification and the input data, and wherein the method comprises at least one of the following: determining the semantic description by requesting a first major language model to produce an output that semantically describes the input data, in particular requesting a human expert or an automated machine validator to evaluate the semantically descriptive output of the input data, receiving a result of the evaluation of the semantically descriptive output of the input data, and determining the semantic description depending on the semantically descriptive output of the input data and the result of the evaluation of the semantically descriptive output of the input data, and determining the ontology by requesting the first or a second major language model to produce an output that ontologically describes the semantic description.In particular, requesting a human expert or an automated machine validator to review the semantic description ontologically describing output, receiving a result of the review of the semantic description ontologically describing output, and determining the ontology depending on the semantic description ontologically describing output and the result of the review of the semantic description ontologically describing output, and determining the mapping specification by requesting the first or second or a third major language model to output an output mapping for the semantic description and the ontology, in particular requesting a human expert or an automated machine validator to review the output mapping.Receiving the result of the review of the initial image and determining the ontology depending on the initial image and the result of the review of the initial image.

[0003] The process automates the construction of the knowledge graph with instructions from a human expert or with automated machine validation regarding syntax completeness and consistency without instructions from a human expert.

[0004] The process proceeds step-by-step through several stages of knowledge graph construction. An example sequence is determining a semantic description, determining an ontology, determining a mapping, and executing the mapping. At each step, a user, such as a human expert or automated machine validation, can review and / or revise the intermediate results—that is, the semantic description, the ontology, and the mapping—to refine the output of the large language model. The user can provide additional input, such as an existing ontology, based on which the process constructs a new, extended ontology.

[0005] The knowledge graph can encompass complex and heterogeneous data. It is based on semantic technologies, meaning it describes the data unambiguously and in a semantic language that is interpretable by both humans and machines. This semantic language can be standardized. The knowledge graph supports interoperability and knowledge sharing. It is designed for storing and discovering highly interconnected data and supports multi-step reasoning.

[0006] The method is based on formal languages ​​and semantics. The ontology, for example, is based on an ontology language, such as the Web Ontology Language OWL.

[0007] The mapping specification is based, for example, on the RDF mapping language RML, which is based on the W3C standard RDF (Resource Description Framework).

[0008] The semantic description and / or the mapping specification can be based on a constraint language, such as Shapes Constraint Language (SHACL).

[0009] The knowledge graph comprises knowledge graph data. A validation task can be performed on this knowledge graph data. This validation task can include a completeness check, a consistency check, and / or verification of compliance with standards. The knowledge graph provides a foundation for retrieving, discovering, and analyzing complex data and enables data analytics and decision-making in application areas such as finance, supply chain management, healthcare, and biotechnology.

[0010] The general advantage of using the large language model in the steps of knowledge graph construction is that (experienced) human

[0011] Experts need a lot of time and effort, which can be minimized if the proposed procedure is adopted.

[0012] The image is a machine-readable intermediate work result that reduces the complexity of applying automatic verification and validation regarding the completeness and correctness of the produced graph.

[0013] Using the ontology supports the production of better results for specialized fields, such as manufacturing. The process can utilize previously existing ontologies and expert input, further improving the suitability of the constructed knowledge graph for a specific application.

[0014] Providing the input data may involve providing a table containing columns and rows, where the item header identifies the column containing the item, where the property header identifies the column containing the property, and where the table contains the item and the property in the same row.

[0015] Determining the semantic description of the headers can include determining a structured output that associates the item header with a description of the item, a semantics of the item, and a data type of the item, and determining a structured output that associates the property header with a description of the property, a semantics of the property, and a data type of the property.

[0016] Determining the ontology can include determining the property for the header of the item's property such that it includes the header of the item's property as an identifier for the header of the item's property.

[0017] Determining the mapping specification that defines the relationship between the header for the item and the header for the property of the item may involve providing a first mapping to determine a subject of the triple, wherein the first mapping includes a template that includes the header of the item, and providing a second mapping to determine a predicate of an object of the triple, wherein the second mapping includes the relationship and a template that includes the header of the property of the item.

[0018] Constructing the triple can involve providing the item as the subject of the triple, providing the relation as the predicate of the triple, and determining the object of the triple by finding the first mapping depending on the header of the item, finding the 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.

[0019] The procedure can include providing properties associated with the header of the property for the item, and where the properties include the property of the item; providing properties associated with a different header, and where finding the property of the item as the object depending on the second mapping; and determining a statement to find the properties associated with the header of the property of the item depending on the second mapping, in particular to search for the property of the item only in the properties associated with the header of the property of the item in the second mapping.

[0020] Determining the ontology may involve determining a class definition that includes an identifier for the class, and determining the property for the header of the item's property such that it includes the identifier for the class.

[0021] Determining the ontology can include determining the property for the header of the item's property such that it includes a scope and a data type of the second property.

[0022] The stepwise approach allows for better scaling to large datasets. Intermediate results can be shared, examined, and potentially rewritten, for example, using the large language model or another large language model. This promotes efficiency, such as better scaling and reduced costs when handling complex data.

[0023] The procedure can involve instructing the large language model to produce a mapping for the triple, instead of instructing the large language model to directly convert all data into triples. Using a mapping allows for easy scaling of knowledge graph construction while minimizing costs; for example, it is not necessary to send complete datasets to the large language model, which is costly and inefficient.

[0024] The procedure may include validation of the knowledge graph containing the constructed triple and, if validation of the knowledge graph is successful, construction of another triple of the knowledge graph depending on the input, the ontology, the semantic description and the mapping specification determined during the construction of the constructed triple, and removal of the constructed triple and the ontology, semantic description and mapping specification determined for the constructed triple before otherwise determining another triple depending on the input.

[0025] The produced map can be used with the input or with new data of the same structure as the data used to produce the map, without needing to reuse the large language model. In particular, the new data subsequently read from the map may contain sensitive protected data that does not need to be sent to a large language model using this method.

[0026] The procedure can involve constructing a different triple of the knowledge graph depending on a different input and the ontology, semantic description, and mapping specification determined during the construction of the constructed triple, or determining multiple triples depending on the mapping specification. Constructing triples from existing mappings reduces the time required for knowledge graph construction and improves scalability.

[0027] A device for constructing a triple of a knowledge graph, wherein the device comprises at least one processor and at least one memory, wherein the at least one memory stores instructions executable by the at least one processor which, when executed by the at least one processor, cause the device to execute the method.

[0028] A computer program product for constructing a triple of a knowledge graph, wherein the computer program product comprises computer-readable instructions which, when executed by the computer, cause the computer to perform the procedure.

[0029] A data structure for constructing a triple of a knowledge graph, wherein the data structure comprises at least one data field for input data, wherein the input data comprises an item and a property of the item and a header for the item and a header for the property of the item, wherein the data structure comprises at least one data field for a semantic description of the headers, determined in particular with a large language model depending on the input data, wherein the data structure comprises at least one data field for an ontology that defines a property for the header of the property of the item, wherein the ontology is determined in particular with a large language model depending on the semantic description, wherein the data structure comprises at least one data field for a mapping specification.wherein the mapping specification defines a relationship between the header for the item and the header for the property of the item, wherein the mapping specification is determined in particular with a large language model depending on the semantic description and the ontology, and wherein the data structure includes at least one data field for the triple encompassing the relationship between the item and the property, wherein the triple is constructed depending on the mapping specification.

[0030] Further exemplary embodiments are derived from the following description and the drawings. These show: Fig. 1 schematically a device for constructing a triple of a knowledge graph, Fig. 2. A flowchart showing the steps of a procedure for constructing the triple of the knowledge graph. Fig. 3. Schematically, a data structure for constructing the triple of the knowledge graph.

[0031] Fig. Figure 1 schematically shows a device 100 for constructing a triple of a knowledge graph.

[0032] A knowledge graph is a collection of facts, also called triples, of the form (subject, predicate, object). The predicate defines the relationship between the subject and the object.

[0033] The device 100 comprises at least one processor 102 and at least one memory 104.

[0034] The at least one memory 104 stores instructions executable by the at least one processor 102 which, when executed by the at least one processor 102, cause the device 100 to execute a method for constructing the triple of the knowledge graph.

[0035] The procedure is described using the example of a large language model. This large language model is designed to understand and summarize data in multiple formats. It is also designed to be prompted to produce output in specific formats.

[0036] The method uses these features of the large language model by providing the large language model with input data and a query, the so-called request, to extract specific information about the input data in a structured format.

[0037] An example of a large language model is a generative pre-trained transformer model. Liu, Y., Han, T., Ma, S., Zhang, J., Yang, Y., Tian, ​​J., ... & Ge, B. (2023). “Summary of ChatGPT-Related Research and Perspective Towards the Future of Large Language Models”. Meta-Radiology, 100017, describes an example of a large language model.

[0038] The method is described using the example of an ontology. An ontology is a formal, explicit description of concepts, also called classes, in a field of discourse, where properties, also called relations, of each concept describe various features and attributes of the concept, as well as restrictions on concepts and properties.

[0039] Noy, Natalya F. and McGuinness, Deborah L. “Ontology Development 101: A Guide to Creating Your First Ontology” describes an example of an ontology.

[0040] The procedure is described using the example of constructing a sample triple. For knowledge graph construction, the triples of the knowledge graph are constructed in the same way as described for the sample triple.

[0041] Knowledge graph construction is a process for constructing a knowledge graph from input data sourced from one or more data sources, such as text or structured data. Knowledge graph construction is a time-consuming process that often requires an understanding of semantic web technologies and the relevant domain of interest. In an example environment, relational data (tabular data) is collected from various systems and used as input for constructing a unified knowledge graph.

[0042] Examples of steps for constructing a knowledge graph are: (1) Generating an ontology for the domain of input data; (2) Creating a mapping from tabular data to such an ontology, (3) Filling the knowledge graph with the tabular data about such a representation.

[0043] The procedure is described using the example of input data provided as a table, i.e., tabular data. The procedure is equally applicable to input data in certain other structured or semi-structured formats, such as key-value pairs.

[0044] The procedure may, if necessary, use additional text information about the input data in the prompt.

[0045] The procedure presupposes the existence of the large language model and access to it.

[0046] The procedure determines a triple that includes a subject, a predicate, and an object.

[0047] The procedure includes step 202.

[0048] Step 202 involves providing the input data.

[0049] The input data consists of a table.

[0050] The table is provided for a class. The table includes columns and rows.

[0051] The columns comprise a header in the first row of the table. The other rows of the table correspond to a specific instance of the class.

[0052] A column contains a header of an item in the first row and, in a second row, an item associated with that item's header. A column contains a header of an item's property in the first row and the property of the item associated with that property's header in the second row.

[0053] According to the 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.

[0054] This means that the header for the item identifies the column containing the item. The header for the property identifies the column containing the property. The table includes the item and the property in the same row.

[0055] The table can contain multiple rows for different instances of the class.

[0056] The table can include multiple columns for different properties, with a header of the respective property in the first row and the respective property for the respective instance in the other rows.

[0057] That is, the input data includes the item and the item's property, and the item's header and the item's property header.

[0058] This means the input data includes properties associated with the header of the item's property. The input data includes the item's property and properties associated with another header.

[0059] For example, the table contains production item numbers in the first column and temperature values ​​as the property in the second column. The header for the item is "Production Item". The header for the property is "Temperature".

[0060] The procedure includes step 204.

[0061] Step 204 involves determining a semantic description of the headers depending on the input data.

[0062] According to an example, the semantic description of the headers is determined using the large language model depending on the input data and depending on a request that asks the large language model to output the semantic description of the headers depending on the input data.

[0063] Determining the semantic description of the headers can include determining a structured output.

[0064] According to an example, the structured output associates the item header with a description of the item, a semantics of the item, and a data type of the item.

[0065] The large language model, for example, requires that the semantic description be output as a structured output, which in particular associates the header of the item with the description of the item, the semantics of the item and the data type of the item.

[0066] According to one example, the structured output associates the property header with a description of the property, a semantics of the property, and a data type of the property.

[0067] The large language model, for example, requires that the semantic description be output as a structured output, which specifically associates the property header with the property description, the property semantics, and the property data type.

[0068] Given the input data, such as the table, a portion of it can be sampled and included in a request for the large language model to recognize the data model and generate its semantic description. Optionally, a textual description of the input data can be provided in the request. The request is designed to return the structured output in a format defined within the request.

[0069] For the example table, which includes production items and temperatures, the semantic description includes, for example, "Production Item": { "description": "The number of the production item" "semantics": "Number" "data type": "integer"} "Temperature": { "description": "The temperature of the production item" "semantics": "Temperature" "data type": "integer"}

[0070] The data type can be different, e.g., "string".

[0071] Step 204 may involve revising the semantic description.

[0072] For example, step 204 includes - Requesting the large language model to produce output that semantically describes the input data, - In particular, requesting a human expert or an automated machine validation to review the output that semantically describes the input data, - Receiving a result of the review of the output semantically describing the input data, and - Determining the semantic description depending on the output that semantically describes the input data and the result of the review of the output that semantically describes the input data.

[0073] The output that semantically describes the input data is the semantic description to be revised.

[0074] The result of the review can confirm the output, which semantically describes the input data, as the semantic description. The result of the review can include changes made during the review of the output, which semantically describes the input data, as the semantic description.

[0075] The procedure includes step 206.

[0076] Step 206 involves determining an ontology that defines a first property for the header of the item and a second property for the header of the item's property, depending on the semantic description.

[0077] According to one example, the ontology is determined using the large language model depending on the semantic description and depending on a request that asks the large language model to output the ontology depending on the semantic description.

[0078] For example, the first property includes the header of the item as an identifier for the first property. The large language model, for instance, requests that the header of the item be output as an identifier for the first property.

[0079] For example, the second property includes the header of the property of the item as an identifier for the second property. The large language model, for instance, requires that the header of the property be output as an identifier for the second property.

[0080] The ontology can include a class definition that contains a class identifier. The first property and / or the second property can contain the class identifier. The large language model can be requested to provide the ontology with the class definition containing the class identifier. The large language model can also be requested to provide the first property or the second property in such a way that it contains the class identifier.

[0081] The first property can include a scope and a data type. The large language model can be requested to provide the first property with its scope and data type. The second property can include a scope and a data type. The large language model can be requested to provide the second property with its scope and data type.

[0082] As an example, the semantic description is redirected back to the large language model, with a prompt to construct an OWL ontology with classes and relationships for the data in the semantic description. For more focused ontology generation, a textual description of how the input data will be used can be added to the prompt. The result of this step is an ontology in OWL format.

[0083] Since the result of the large language model may be incomplete and / or contain errors, a refinement step, i.e., an ontology refinement, can be performed in a loop. The refinement loop can be repeated until the resulting ontology satisfies a predetermined criterion.

[0084] For the example table, which includes production costs and temperatures, the ontology includes, for example, # class definition. <produktion>rdf:Type rdfs:Class ; rdfs:Identifier "Production". #Property definition <produktionsposten>rdf:Type rdf:Property ; rdfs:Identifier "Production Item"; rdfs:Area <produktion>; rdfs:scope xsd:string . <temperature> rdf:type rdf:property ; rdfs:label "temperature"; rdfs:area <produktion>; rdfs:scope xsd:string . Step 206 may include revising the ontology.

[0085] For example, step 206 includes - Requesting the large language model to output a description that ontologically describes the semantic description, - In particular, requesting a human expert or an automated machine validation to review the output that ontologically describes the semantic description, - Receiving a result of the review of the output that ontologically describes the semantic description, and - Determining the ontology depending on the output ontologically describing the semantic description and the result of the review of the output ontologically describing the semantic description.

[0086] The output that ontologically describes the semantic description is the ontology to be revised.

[0087] The review may confirm the output that ontologically describes the semantic description as the ontology. The review may also include changes made during the review of the output that ontologically describes the semantic description as the ontology.

[0088] The procedure includes step 208.

[0089] Step 208 involves determining a mapping specification depending on the semantic description and the ontology.

[0090] The mapping specification defines a relationship between the header for the item and the header for the item's property.

[0091] According to an example, the mapping specification is determined with the large language model depending on the semantic description and the ontology and depending on a request that asks the large language model to output the mapping specification depending on the semantic description and the ontology.

[0092] The mapping specification includes, for example, a first mapping to determine a subject of the triple. This first mapping includes, for example, a template containing the item's header. The large language model is then requested to output the first mapping with the template containing the item's header.

[0093] The mapping specification includes, for example, a second mapping for determining a predicate and an object of the triple. This second mapping includes, for example, the relationship and a template containing the header of the item's property. The large language model is requested, for example, to output the second mapping with the template containing the item's property header.

[0094] Given the ontology and semantic description, the large language model can be queried to generate the mapping specification in RML (RDF Mapping Language). RML is described in A. Dimou, M. Vander Sande, P. Colpaert, R. Verborgh, E. Mannens, and R. Van de Walle, "RML: A Generic Language for Integrated RDF Mappings of Heterogeneous Data," in Proceedings of the 7th Workshop on Linked Data on the Web, Volume 1184, CEUR Workshop Proceedings, 2014.

[0095] The large language model can be queried to transform the data into a graph that follows the ontology. The procedure can include automatic checking of the produced mapping specification for errors and completeness. The procedure can include a refinement step, i.e., mapping validation, to ensure that the large language model produces a correct mapping specification.

[0096] For the example table, which includes production costs and temperatures, the figure specification includes, for example, rr:subjectFigure [ rr:template "http: / / example.com / Production / {Production Items}"; rr:class <produktion>] rr:predicateObjectImage [ rr:predicate <temperatur>; rr:objectImage [rr:column "Temperature"] ]

[0097] Step 208 may include revising the mapping specification.

[0098] For example, step 208 includes - Requesting the large language model to output a mapping for the semantic description and ontology, - Requesting, in particular, a human expert or an automated machine validation to review the output image, - Receiving a result of the review of the output image and - Determining the ontology depending on the output image and the result of the review of the output image.

[0099] The output mapping for the semantic description and the ontology is the mapping specification to be revised.

[0100] The review result can confirm the output mapping for the semantic description and ontology as the mapping specification. The review result may include changes made during the review of the output mapping for the semantic description and ontology as the mapping specification.

[0101] The procedure includes step 210.

[0102] Step 210 involves constructing the triple encompassing the relationship between the item and the property, depending on the mapping specification. According to an example, the triple is constructed depending on the mapping specification and the input data, and depending on a request that asks the large language model to output the triple depending on the mapping specification and the input data.

[0103] According to an example, the item is provided as the subject of the triple.

[0104] According to an example, the relation is provided as the predicate of the triple.

[0105] According to an example, the object of the triple is determined by finding the first mapping depending on the header of the item, finding the second mapping depending on the relationship, and finding the property of the item as the object depending on the second mapping.

[0106] For example, to find the property of the item as the object dependent on the second mapping, a statement is required to find the properties associated with the header of the item's property, dependent on the second mapping. Specifically, the statement instructs the search for the item's property only among the properties associated with the header of the item's property in the second mapping.

[0107] For example, the large language model specifically requests the instruction to find the first mapping depending on the header of the item, to find the second mapping depending on the relationship, and to find the property of the item as the object depending on the second mapping.

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

[0109] The data structure 300 includes at least one data field 302 for the input data, the semantic description, the ontology of the mapping specification, and the triple. QUOTES INCLUDED IN THE DESCRIPTION

[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited non-patent literature

[0000] Liu, Y., Han, T., Ma, S., Zhang, J., Yang, Y., Tian, ​​J., ... & Ge, B. (2023). "Summary of ChatGPT-Related Research and Perspective Towards the Future of Large Language Models". Meta Radiology, 100017

[0037] Noy, Natalya F. and McGuinness, Deborah L. “Ontology Development 101: A Guide to Creating Your First Ontology

[0039] A. Dimou, M. Vander Sande, P. Colpaert, R. Verborgh, E. Mannens und R. Van de Walle. „RML: A Generic Language for Integrated RDF Mappings of Heterogeneous Data". In Proceedings of the 7th Workshop on Linked Data on the Web, Band 1184, der CEUR Workshop Proceedings. CEUR, 2014

[0094] < / temperatur> < / produktion> < / produktion> < / produktion> < / produktionsposten> < / produktion>

Claims

[1] Computer-implemented method for constructing a triple of a knowledge graph, characterized by, that the procedure comprises providing (202) input data, wherein the input data comprises an item and a property of the item and a header for the item and a header for the property of the item, wherein the procedure comprises determining (204) a semantic description of the headers depending on the input data, determining (206) an ontology that defines a property for the header of the property of the item depending on the semantic description, determining (208) a mapping specification depending on the semantic description and the ontology, wherein the mapping specification defines a relationship between the header for the item and the header for the property of the item, and wherein the procedure comprises constructing (210) the triple that comprises the relationship between the item and the property depending on the mapping specification and the input data,and wherein the method comprises at least one of the following: determining the semantic description by requesting a first major language model to produce an output that semantically describes the input data, in particular requesting a human expert or an automated machine validator to evaluate the semantically describing input data output, receiving a result of the evaluation of the semantically describing input data output, and determining the semantic description depending on the semantically describing input data output and the result of the evaluation of the semantically describing input data output, and determining the ontology by requesting the first or a second major language model to produce an output that ontologically describes the semantic description, in particular requesting a human expert or an automated machine validator.to evaluate the ontologically describing output of the semantic description, receive a result of the evaluation of the ontologically describing output of the semantic description and determine the ontology depending on the ontologically describing output of the semantic description and the result of the evaluation of the ontologically describing output of the semantic description, and determine the mapping specification by requesting the first or the second or a third large language model to output an output mapping for the semantic description and the ontology, in particular requesting a human expert or an automated machine validation to evaluate the output mapping, receiving a result of the evaluation of the output mapping and determining the ontology depending on the output mapping and the result of the evaluation of the output mapping. [2] Method according to claim 1, characterized by , that providing (202) the input data includes providing a table with columns and rows, wherein the item header identifies the column containing the item, wherein the property header identifies the column containing the property, and wherein the table includes the item and the property in the same row. [3] Method according to any one of the preceding claims, characterized by , that determining (204) the semantic description of the headers includes: determining a structured output that associates the header of the item with a description of the item, a semantics of the item and a data type of the item, and determining a structured output that associates the header of the property with a description of the property, a semantics of the property and a data type of the property. [4] Method according to any one of the preceding claims, characterized by, that determining (206) the ontology includes: determining the property for the header of the item's property such that it includes the header of the item's property as an identifier of the property for the header of the item's property. [5] Method according to claim 4, characterized by , that determining (208) the mapping specification that defines the relationship between the header for the item and the header for the property of the item includes: providing a first mapping to determine a subject of the triple, wherein the first mapping includes a template that includes the header of the item, providing a second mapping to determine a predicate and an object of the triple, wherein the second mapping includes the relationship and a template that includes the header of the property of the item. [6] Method according to claim 5, characterized by, that constructing (210) the triple includes: providing the item as the subject of the triple, providing the relation as the predicate of the triple, and determining the object of the triple by finding the first mapping depending on the header of the item, finding the 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. [7] Method according to claim 6, characterized by, that the procedure comprises: providing (202) properties associated with the header of the property for the item, wherein the properties include the property of the item; providing (202) properties associated with another header, wherein finding (210) the property of the item as the object depending on the second mapping includes determining a statement to find the properties associated with the header of the property of the item depending on the second mapping, in particular to search for the property of the item only in the properties associated with the header of the property of the item in the second mapping. [8] Method according to any one of claims 4 to 7, characterized by, that determining (206) the ontology includes: determining a class definition that includes an identifier for the class, and determining the property for the header of the property of the item such that it includes the identifier for the class. [9] Method according to any one of claims 4 to 8, characterized by , that determining (206) the ontology includes: determining the property for the header of the property of the item such that it includes a scope and a data type of the second property. [10] Method according to any one of the preceding claims, characterized by, that the procedure includes: validating the knowledge graph containing the constructed triple, and constructing another triple of the knowledge graph depending on the input, the ontology, the semantic description, and the mapping specification determined during the construction of the constructed triple, if the validation of the knowledge graph is successful, and removing the constructed triple and the ontology, semantic description, and mapping specification determined for the constructed triple before otherwise determining another triple depending on the input. [11] Method according to any one of the preceding claims, characterized by, that the procedure includes: constructing another triple of the knowledge graph depending on a different input and the ontology, semantic description and mapping specification determined when constructing the constructed triple, or determining multiple triples depending on the mapping specification [12] Device (100) for constructing a triple of a knowledge graph, characterized by , that the device comprises at least one processor (102) and at least one memory (104), wherein the at least one memory (104) stores instructions executable by the at least one processor (102) which, when executed by the at least one processor (102), cause the device (100) to execute a method according to any one of claims 1 to 11. [13] Computer program product for constructing a triple of a knowledge graph, characterized bythat the computer program product comprises computer-readable instructions which, when executed by the computer, cause the computer to execute a method according to any one of claims 1 to 11. [14] Data structure (300) for constructing a triple of a knowledge graph, characterized bythat the data structure (300) comprises at least one data field (302) for input data, wherein the input data comprises an item and a property of the item and a header for the item and a header for the property of the item, wherein the data structure (300) comprises at least one data field (302) for a semantic description of the headers, determined in particular by a large language model depending on the input data, wherein the data structure (300) comprises at least one data field (302) for an ontology that defines a property for the header of the property of the item, wherein the ontology is determined in particular by a large language model depending on the semantic description, wherein the data structure (300) comprises at least one data field (302) for a mapping specification, wherein the mapping specification defines a relationship between the header for the item and the header for the property of the item,wherein the mapping specification is determined in particular with a large language model depending on the semantic description and the ontology and wherein the data structure (300) includes at least one data field (302) for the triple encompassing the relationship between the item and the property, wherein the triple is constructed depending on the mapping specification.

Images