Knowledge graph generation system

Abstract

This invention relates to a knowledge graph generation system (100), comprising an input unit (20) and a knowledge graph generator (10). The input unit is configured to receive multiple service reports and provide the multiple service reports to the knowledge graph generator; the knowledge graph generator is configured to deploy a large language model (LLM), identify information entries from the natural language in the multiple service reports using the LLM, form multiple rules using the identified information entries, generate multiple knowledge graph nodes using the multiple rules, and generate a knowledge graph using the multiple knowledge graph nodes, wherein generating the knowledge graph includes forming multiple node groups, wherein each node group includes multiple knowledge graph nodes linked together.

Description

Technical Field

[0001] This invention relates to a knowledge graph generation system, a knowledge graph generation method, and a computer program element. This is useful for the maintenance of electrical equipment such as low-voltage, medium-voltage, or high-voltage switchgear. Background Technology

[0002] As the required knowledge level for service personnel increases, a new wave of young and inexperienced personnel is entering the field, urgently needing the support of relevant tools. This phenomenon exists in many technical fields, one example being the field of switchgear (low-voltage, medium-voltage, and high-voltage).

[0003] Preventive and corrective maintenance of electrical equipment, such as switchgear, is a complex task that requires expertise to address a wide range of potential problems. Past service reports often contain valuable information and solutions for common faults, as well as potential "rules" for resolving such issues.

[0004] However, the relevant information contained in these reports—along with manuals and other knowledge bases—is scattered and disorganized. Furthermore, depending on the country of origin, this information may be in multiple different languages. This places a significant cognitive burden on future service technicians as they navigate through multiple documents and draw upon past knowledge. Building a unified knowledge base from this vast amount of existing information presents a technical challenge.

[0005] It is necessary to solve this problem. Summary of the Invention

[0006] Therefore, it would be beneficial to provide service personnel working with low-voltage, medium-voltage, and high-voltage switchgear with improved support tools.

[0007] The object of the invention is achieved by the subject matter of the independent claims, wherein further embodiments are included in the dependent claims.

[0008] In a first aspect, a knowledge graph generation system is provided, the system comprising: - Input unit; and - Knowledge graph generator.

[0009] The input unit is configured to receive multiple service reports, each including natural language. The input unit is configured to provide multiple service reports to a knowledge graph generator. The knowledge graph generator is configured to deploy / integrate a Large Language Model (LLM). The knowledge graph generator is configured to use the LLM to identify information entries / records / items from the natural language in the multiple service reports. The identified information entries include the name and type of the device, the problem these devices need to solve, and the actions / dispositions taken by service technicians to alleviate these problems. The knowledge graph generator is configured to form multiple rules using the identified information entries. Each rule contains a tuple consisting of a device type, a problem, and a disposition step or a series of disposition steps taken to alleviate the problem. The knowledge graph generator is configured to generate multiple knowledge graph nodes using the multiple rules, where each device type, problem, and disposition step is transformed into a separate knowledge graph node. The knowledge graph generator is configured to generate a knowledge graph using the multiple knowledge graph nodes, which includes forming multiple node groups. Each node group includes multiple knowledge graph nodes linked together.

[0010] It should be noted that the deployment of a large language model (LLM) may mean utilizing an existing large language model to meet specific needs, or it may mean that the large language model is actually built for specific needs.

[0011] Linked knowledge graph nodes can be considered as rules related to the steps required to resolve a problem about a device type.

[0012] In one example, multiple service reports include natural languages ​​in multiple different languages, including two or more of English, German, French, Italian, Spanish, Portuguese, Japanese, and Chinese.

[0013] The various languages ​​mentioned can be languages ​​other than those mentioned above, and can be any language in the service report.

[0014] Therefore, the system works using a single language, but it can also handle reports from multiple languages, or actually handle a single report containing more than one language.

[0015] Therefore, the new R&D advancements actually provide a workflow for extracting electrical equipment and maintenance rules from multilingual natural language text in service reports, and employ a knowledge graph to appropriately store the extracted rules. In this way, relevant information for current maintenance work can be retrieved and found more effectively in terms of locating and mitigating appropriate rules for current / on-the-fly issues of specific / specific equipment types.

[0016] In one example, each knowledge graph node includes at least one searchable attribute or characteristic, which includes at least one of device type, problem, or disposition steps.

[0017] This makes searching for information about devices, problems, and handling / operation more efficient.

[0018] In one example, the input unit is configured to receive at least one new service report. Each new service report contains natural language. The input unit is configured to provide at least one new service report to a knowledge graph generator. The knowledge graph generator is configured to identify information entries from the natural language in the at least one new service report using a large language model. The identified information entries include the name and type of the device, the problem that the device needs to solve, and the actions taken by service technicians to alleviate these problems. The knowledge graph generator is configured to form one or more new rules using the identified information entries, wherein each new rule contains a tuple consisting of a device (type), a problem, and actions or a series of actions taken to alleviate the problem. For each new rule, the knowledge graph generator is configured to generate multiple new knowledge graph nodes using the new rule, wherein the device type, the problem, and each action are transformed into separate new knowledge graph nodes.

[0019] In one example, the knowledge graph generator is configured to search the knowledge graph to determine if a linked node group contains a knowledge graph node that has the same device type as the device type associated with the new rule. The knowledge graph generator is configured to link multiple new knowledge graph nodes associated with the new rule to knowledge graph nodes in a linked node group that includes knowledge graph nodes with the same device type as the device type associated with the new rule. If no knowledge graph node with the same device type as the device type associated with the new rule is found in the knowledge graph, the knowledge graph generator is configured to form a new node group in the knowledge graph that includes multiple new knowledge graph nodes.

[0020] In one example, the knowledge graph generator is configured to retrieve the knowledge graph to determine if a linked node group contains a knowledge graph node that has the same or similar question as the question associated with the new rule. The knowledge graph generator is configured to link multiple new knowledge graph nodes associated with the new rule to knowledge graph nodes in a linked node group within the knowledge graph that include knowledge graph nodes with the same or similar questions associated with the new rule. If no knowledge graph node with the same or similar questions associated with the new rule is found in the knowledge graph, the knowledge graph generator is configured to form a new node group in the knowledge graph that includes multiple new knowledge graph nodes.

[0021] Therefore, when new service reports become available, the knowledge graph can be updated efficiently and effectively to enhance the understanding of mitigation methods for specific device types and to continuously improve its usability by expanding the knowledge graph with questions related to new device types.

[0022] In one example, the input unit is configured to receive multimodal data. This multimodal data is not in text format. The input unit is configured to provide the multimodal data to a knowledge graph generator. The knowledge graph generator is configured to analyze the multimodal data and / or analyze information associated with the multimodal data to determine associated groups of nodes in the knowledge graph and append the multimodal data to the associated groups of nodes.

[0023] Therefore, knowledge graphs can be augmented with non-textual information, such as photos and body camera footage from service technicians, videos, sensor readings, and recordings from augmented reality (AR) and virtual reality (VR) conversations with remote experts.

[0024] In one example, the system includes an output unit. This output unit is configured to output a knowledge graph.

[0025] Therefore, this knowledge graph can be used on / by another system.

[0026] In a second aspect, a knowledge graph generation method is provided, which includes: - The input unit receives multiple service reports, each of which contains natural language. - The input unit provides multiple service reports to the knowledge graph generator; - Deploy large language models (LLMs) using a knowledge graph generator; - The knowledge graph generator uses a large language model to identify information entries from the natural language in multiple service reports. The identified information entries include the name and type of the device, the problems that these devices need to solve, and the measures taken by service technicians to alleviate these problems. - The knowledge graph generator uses the identified information entries to form multiple rules, where each rule includes a tuple consisting of a device type, a problem, and a set of disposal steps or a series of disposal steps to alleviate the problem; - A knowledge graph generator uses multiple rules to generate multiple knowledge graph nodes, where each device type, problem, and action step is transformed into a separate knowledge graph node; and - A knowledge graph is generated by a knowledge graph generator using multiple knowledge graph nodes. The generated knowledge graph includes forming multiple node groups, where each node group includes multiple knowledge graph nodes linked together.

[0027] It should be noted that the deployment of a large language model (LLM) may mean utilizing an existing large language model to meet specific needs, or it may mean that the large language model is actually built for specific needs.

[0028] In one example, multiple service reports include natural languages ​​in multiple different languages, including two or more of English, German, French, Italian, Spanish, Portuguese, Japanese, and Chinese.

[0029] Therefore, this method works using a single language, but it is also able to handle reports from multiple languages, or in fact, a single report containing more than one language.

[0030] In one example, each knowledge graph node includes at least one searchable attribute or characteristic, which includes at least one of device type, problem, and handling steps.

[0031] In one example, the method includes: - The input unit receives at least one new service report, wherein each new service report contains natural language; - The input unit provides at least one new service report to the knowledge graph generator; - Information entries are identified from the natural language of at least one new service report by a knowledge graph generator using a large language model. The identified information entries include the name and type of the device, the problems that the device needs to solve, and the measures taken by service technicians to alleviate these problems. - A knowledge graph generator uses the identified information entries to form one or more new rules, wherein each new rule includes a tuple consisting of a device (type), a problem, and a set of disposal steps or a series of disposal steps to alleviate the problem; and - For each new rule, the method includes generating multiple new knowledge graph nodes by a knowledge graph generator using the new rule, wherein the device type, problem, and each disposition step are transformed into separate new knowledge graph nodes.

[0032] In one example, the method includes: - The knowledge graph is retrieved by the knowledge graph generator to determine whether the linked node group has a knowledge graph node that has the same device type as the device type associated with the new rule; - The knowledge graph generator links multiple new knowledge graph nodes associated with the new rule to knowledge graph nodes in a group of linked nodes that include knowledge graph nodes with the same device type as the device type associated with the new rule; or - If no knowledge graph node with the same device type as the new rule is found in the knowledge graph, the knowledge graph generator will form a new node group in the knowledge graph that includes multiple new knowledge graph nodes.

[0033] In one example, the method includes: - The knowledge graph generator retrieves the knowledge graph to determine whether a group of linked nodes has a knowledge graph node that has the same or similar questions as the questions associated with the new rules. - The knowledge graph generator links multiple new knowledge graph nodes associated with the new rule to knowledge graph nodes in a linked node group within the knowledge graph, which includes knowledge graph nodes with the same or similar questions associated with the new rule; or - If no knowledge graph node with the same or similar problem as the new rule is found in the knowledge graph, the knowledge graph generator will form a new node group in the knowledge graph that includes multiple new knowledge graph nodes.

[0034] In one example, the method includes: - Receives multimodal data from the input unit, wherein the multimodal data is not in text form; - Input units provide multimodal data to the knowledge graph generator; and - The knowledge graph generator analyzes multimodal data and / or information associated with the multimodal data to identify associated node groups in the knowledge graph and appends the multimodal data to the associated node groups.

[0035] In one example, the method includes outputting a knowledge graph from the output unit.

[0036] In a third aspect, a computer program element is provided for controlling a processor to execute the methods of the second aspect and any associated examples thereof.

[0037] In one example, a computer-readable medium having computer program elements stored is provided.

[0038] Computer program elements can be, for example, software programs, but can also be FPGAs (Field Programmable Gate Arrays), PLDs (Programmable Logic Devices), or any other suitable digital devices.

[0039] The above aspects and examples will become clear and illustrated with reference to the embodiments described below. Attached Figure Description

[0040] Exemplary embodiments will now be described with reference to the accompanying drawings: Figure 1An exemplary knowledge graph generation system is shown, in which optional elements / features are shown in dashed lines; and Figure 2 A specific embodiment of the knowledge graph generation system is shown. Detailed Implementation

[0041] Figure 1 An exemplary knowledge graph generation system 100 is shown, which includes: - Input unit 20; and - Knowledge Graph Generator 10.

[0042] The input unit is configured to receive multiple service reports, each including natural language. The input unit is configured to provide multiple service reports to a knowledge graph generator. The knowledge graph generator is configured to deploy a Large Language Model (LLM). The knowledge graph generator is configured to use the LLM to identify information entries from the natural language in the multiple service reports. The identified information entries include the name and type of the device, the problem that the device needs to solve, and the actions taken by service technicians to alleviate these problems. The knowledge graph generator is configured to form multiple rules using the identified information entries. Each rule contains a tuple consisting of a device type, a problem, and actions taken to alleviate the problem, or a series of actions. The knowledge graph generator is configured to generate multiple knowledge graph nodes using the multiple rules, where each device type, problem, and action is transformed into a separate knowledge graph node. The knowledge graph generator is configured to generate a knowledge graph using the multiple knowledge graph nodes, which includes forming multiple node groups. Each node group includes multiple knowledge graph nodes linked together.

[0043] It should be noted that the deployment of a large language model (LLM) may mean utilizing an existing large language model to meet specific needs, or it may mean that the large language model is actually built for specific needs.

[0044] Linked knowledge graph nodes can be considered as rules related to the steps required to resolve a problem about a device type.

[0045] In one example, multiple service reports include natural languages ​​in multiple different languages, including two or more of English, German, French, Italian, Spanish, Portuguese, Japanese, and Chinese.

[0046] The various languages ​​can be languages ​​other than those mentioned above, and can be any language in the service report.

[0047] Therefore, the system works using a single language, but it can also handle reports from multiple languages, or actually handle a single report containing more than one language.

[0048] Therefore, the new R&D advancements actually provide a workflow for extracting electrical equipment and maintenance rules from multilingual natural language text in service reports, and employ a knowledge graph to appropriately store the extracted rules. In this way, relevant information for current maintenance work can be retrieved and found more effectively in terms of locating and mitigating appropriate rules for current problems of specific equipment types.

[0049] In one example, each knowledge graph node includes at least one searchable attribute or characteristic, which includes at least one of device type, problem, or disposition steps.

[0050] This makes searches for information about devices, problems, and solutions more efficient.

[0051] In one example, the input unit is configured to receive at least one new service report. Each new service report contains natural language. The input unit is configured to provide at least one new service report to a knowledge graph generator. The knowledge graph generator is configured to identify information entries from the natural language in the at least one new service report using a large language model. The identified information entries include the name and type of the device, the problem that the device needs to solve, and the actions taken by service technicians to alleviate these problems. The knowledge graph generator is configured to form one or more new rules using the identified information entries, wherein each new rule includes a tuple consisting of a device (type), a problem, and a disposition step or a series of disposition steps to alleviate the problem. For each new rule, the knowledge graph generator is configured to generate multiple new knowledge graph nodes using the new rule, wherein the device type, the problem, and each disposition step are transformed into separate new knowledge graph nodes.

[0052] In one example, the knowledge graph generator is configured to search the knowledge graph to determine if a linked node group contains a knowledge graph node with the same device type as the device type associated with the new rule. The knowledge graph generator is configured to link multiple new knowledge graph nodes associated with the new rule to knowledge graph nodes in a linked node group that includes knowledge graph nodes with the same device type as the device type associated with the new rule. If no knowledge graph node with the same device type as the device type associated with the new rule is found in the knowledge graph, the knowledge graph generator is configured to form a new node group in the knowledge graph that includes multiple new knowledge graph nodes.

[0053] In one example, the knowledge graph generator is configured to retrieve knowledge graph nodes to determine if a linked node group contains knowledge graph nodes that have the same or similar questions as the questions associated with the new rule. The knowledge graph generator is configured to link multiple new knowledge graph nodes associated with the new rule to knowledge graph nodes in a linked node group that includes knowledge graph nodes with the same or similar questions as the questions associated with the new rule. If no knowledge graph node with the same or similar questions as the questions associated with the new rule is found in the knowledge graph, the knowledge graph generator is configured to form a new node group in the knowledge graph that includes multiple new knowledge graph nodes.

[0054] Therefore, when new service reports become available, the knowledge graph can be updated efficiently and effectively to enhance the understanding of mitigation methods for specific device types and to continuously improve its usability by expanding the knowledge graph with questions related to new device types.

[0055] In one example, the input unit is configured to receive multimodal data. This multimodal data is not in text format. The input unit is configured to provide the multimodal data to a knowledge graph generator. The knowledge graph generator is configured to analyze the multimodal data and / or analyze information associated with the multimodal data to determine associated groups of nodes in the knowledge graph and append the multimodal data to the associated groups of nodes.

[0056] Therefore, knowledge graphs can be augmented with non-textual information, such as photos and body cameras from service technicians, videos, sensor readings, and recordings of augmented reality (AR) and virtual reality (VR) conversations with remote experts.

[0057] In one example, the system includes an output unit 30. This output unit is configured to output a knowledge graph.

[0058] Therefore, this knowledge graph can be used on / by another system.

[0059] In summary, it is clear that the associated exemplary knowledge graph generation method includes the following steps: - The input unit receives multiple service reports, each of which contains natural language. - The input unit provides multiple service reports to the knowledge graph generator; - Deploy large language models (LLMs) using a knowledge graph generator; - The knowledge graph generator uses a large language model to identify information entries from the natural language in multiple service reports. The identified information entries include the name and type of the device, the problems that these devices need to solve, and the measures taken by service technicians to alleviate these problems. - The knowledge graph generator uses the identified information entries to form multiple rules, where each rule includes a tuple consisting of a device type, a problem, and a solution or a series of solution steps to alleviate the problem; - A knowledge graph generator uses multiple rules to generate multiple knowledge graph nodes, where each device type, problem, and action step is transformed into a separate knowledge graph node; and - A knowledge graph is generated by a knowledge graph generator using multiple knowledge graph nodes. The generated knowledge graph includes forming multiple node groups, where each node group includes multiple knowledge graph nodes linked together.

[0060] It should be noted that the deployment of a large language model (LLM) may mean utilizing an existing large language model to meet specific needs, or it may mean that the large language model is actually built for specific needs.

[0061] In one example, multiple service reports include natural languages ​​in multiple different languages, including two or more of English, German, French, Italian, Spanish, Portuguese, Japanese, and Chinese.

[0062] Therefore, this method works using a single language, but it is also able to handle reports from multiple languages, or in fact, a single report containing more than one language.

[0063] In one example, each knowledge graph node includes at least one searchable attribute or characteristic, which includes at least one of device type, problem, and disposition steps.

[0064] In one example, the method includes: - The input unit receives at least one new service report, wherein each new service report contains natural language; - The input unit provides at least one new service report to the knowledge graph generator; - Information entries are identified from the natural language of at least one new service report by a knowledge graph generator using a large language model. The identified information entries include the name and type of the device, the problems that the device needs to solve, and the measures taken by service technicians to alleviate these problems. - A knowledge graph generator uses the identified information entries to form one or more new rules, wherein each new rule comprises a tuple consisting of a device (type), a problem, and a set of disposal steps or a series of disposal steps to alleviate the problem; and - For each new rule, the method includes generating multiple new knowledge graph nodes by a knowledge graph generator using the new rule, wherein the device type, problem, and each disposition step are transformed into separate new knowledge graph nodes.

[0065] In one example, the method includes: - The knowledge graph is retrieved by the knowledge graph generator to determine whether the linked node group has a knowledge graph node that has the same device type as the device type associated with the new rule; - The knowledge graph generator links multiple new knowledge graph nodes associated with the new rule to knowledge graph nodes in a group of linked nodes that include knowledge graph nodes with the same device type as the device type associated with the new rule; or - If no knowledge graph node with the same device type as the new rule is found in the knowledge graph, the knowledge graph generator will form a new node group in the knowledge graph that includes multiple new knowledge graph nodes.

[0066] In one example, the method includes: - The knowledge graph generator retrieves the knowledge graph to determine whether a group of linked nodes has a knowledge graph node that has the same or similar questions as the questions associated with the new rules. - The knowledge graph generator links multiple new knowledge graph nodes associated with the new rule to knowledge graph nodes in a linked node group within the knowledge graph, which includes knowledge graph nodes with the same or similar questions associated with the new rule; or - If no knowledge graph node with the same or similar problem as the new rule is found in the knowledge graph, the knowledge graph generator will form a new node group in the knowledge graph that includes multiple new knowledge graph nodes.

[0067] In one example, the method includes: - Receives multimodal data from the input unit, wherein the multimodal data is not in text form; - Input units provide multimodal data to the knowledge graph generator; and - The knowledge graph generator analyzes multimodal data and / or information associated with the multimodal data to identify associated node groups in the knowledge graph and appends the multimodal data to the associated node groups.

[0068] In one example, the method includes outputting a knowledge graph by output unit 30.

[0069] Therefore, this knowledge graph can be used on / by another system.

[0070] Therefore, new research and development progress provides a technical solution for how to use a large language model (LLM) to extract rules from service reports, store the generated set of rules, retrieve these rules, and enhance the collected knowledge by adding new rules or editing existing rules with the help of a knowledge graph.

[0071] The inventors arrived at this new research and development breakthrough because they recognized that retrieving relevant information is an important activity for most professional groups and a means of gaining a competitive advantage. In electrical equipment maintenance, retrieving relevant maintenance rules helps to troubleshoot problems more quickly, thereby reducing downtime for customer equipment and facilities.

[0072] However, the inventors realized that retrieving information most relevant to the current problem was a complex task. Related information was scattered across multilingual manuals and service reports, making it difficult to access in an integrated and easily searchable manner.

[0073] The inventors have learned that recently, due to the rapid development of generative artificial intelligence, the application of large language models (LLMs) for extracting information from natural language text has received considerable attention. These models are effective in scenarios requiring the parsing of large, unstructured texts in different languages.

[0074] The inventors realized that this latest large language model could be used in a new way to provide service personnel with an effective support tool, which led to new research and development progress.

[0075] In a specific embodiment, new R&D advancements effectively provide a solution processing flow so that: (i) Extract electrical equipment maintenance rules from the multilingual natural language text of the service report. (ii) Using knowledge graphs to appropriately store the extracted rules makes retrieval and information finding efficient, and (iii) When new rules are extracted from new service reports, update the knowledge graph appropriately.

[0076] also, (iv) Knowledge graphs can also be enhanced with non-textual information.

[0077] Figure 2 A specific embodiment of a knowledge graph generation system is shown, illustrating the steps for converting natural language text into a searchable knowledge graph about maintenance rules.

[0078] refer to Figure 2 The new solution process consists of the following stages.

[0079] Phase 1: Extracting rules from the multilingual natural language text in the service report: The maintenance rules are extracted from the multilingual service reports in two steps using the latest (SOTA) Large Language Model (LLM). (The latest Large Language Models are capable of parsing text in different languages ​​(such as English, German, Italian, Spanish, French, etc.) because they have been trained on a large amount of data.) The first step involves the large language model identifying and extracting appropriate entities from the natural language text of the report. The identified entities include the names and types of electrical installations mentioned in the report, the problems or issues that need to be addressed with these installations, and the actions taken by service technicians to alleviate the problem.

[0080] Then, in the second step, these identified entities are linked together to form rules. Each rule is a tuple consisting of a device / approach, a problem, and a series of actions to mitigate that problem. Multiple rules can be extracted from a single report, therefore the second step is needed to correctly identify these rules from all the entities identified in the first step.

[0081] Phase 2: Storing rules in the knowledge graph: After identifying the rule set from Phase 1, these rules are inserted into the knowledge graph. Each device, problem, and step in the disposal list is transformed into a knowledge graph node and linked together appropriately. Nodes in the knowledge graph can have different attributes and characteristics; for example, all "device nodes" belong to one type, all "problem nodes" belong to another type, and so on. This makes retrieval between devices or problems fast and efficient.

[0082] Phase 3: Update the knowledge graph when new rules are extracted: When a new rule is identified from a new service report, the first step is to check all existing device nodes in the knowledge graph to see if the device already exists in the graph or if it is a new device that has never been seen before. If the device already exists in the graph, the new rule is attached to the existing device node. Otherwise, a new node is created for the device.

[0083] Next, all "problem nodes" are retrieved based on the questions in the new rules. If a similar problem is found in the knowledge graph, the newly extracted problem will be linked to the existing problem. This makes it possible to retrieve "related problems," which is crucial for troubleshooting.

[0084] Retrieving relevant questions is a technical challenge because two questions may be semantically identical but use different terms, synonyms, and language patterns. State-of-the-art large language models help identify the most semantically similar search results for a given question. However, the final decision on whether two questions are similar is made by a domain expert, who is the "owner" of the rule-based knowledge graph.

[0085] Phase 4: Enhancing Knowledge Graphs with Multimodal Data Knowledge graphs can effectively store multimodal data sources. Therefore, the knowledge graph generated in the preceding steps can be enhanced by adding data sources, such as photos and body camera footage from service technicians, videos, sensor readings, and records from AR and VR sessions with remote experts (from ABB products RAISE and CLOSER).

[0086] With appropriate adjustments, the progress of new research and development enables lengthy, multilingual technical reports from different regions around the world to generate high-quality, self-updating knowledge graphs. This facilitates faster corrective maintenance and more accurate planning of preventative maintenance for electrical equipment, thereby reducing customer downtime.

[0087] Furthermore, the same solution process can be used in other industrial sectors to more effectively utilize unstructured data, potentially reducing reliance on human expertise and improving the organization of large-scale documents.

[0088] In another exemplary embodiment, a computer program or computer program element is provided, characterized in that it is configured to perform method steps of the method according to one of the foregoing embodiments on a suitable processor or system.

[0089] Therefore, computer program elements can be stored in a computer unit, which may also be part of the embodiments. The computing unit can be configured to execute or cause the execution of the steps of the described methods. Furthermore, it can be configured to operate components of the described system. The computing unit can be configured to automatically operate and / or execute user commands. The computer program can be loaded into the working memory of the data processor. Therefore, the data processor can be equipped to execute the method according to one of the foregoing embodiments.

[0090] This exemplary embodiment of the invention covers computer programs that use the invention from the outset and computer programs that convert existing programs into programs that use the invention through updates.

[0091] Furthermore, computer program elements may be able to provide all the necessary steps to complete the procedures / steps of an embodiment of the method as described above.

[0092] According to another exemplary embodiment of the present invention, a computer-readable medium, such as a CD-ROM, a USB flash drive, etc., is provided, wherein the computer-readable medium has computer program elements stored thereon, the computer program elements being described in the foregoing portion.

[0093] Computer programs may be stored and / or distributed on suitable media, such as optical or solid-state media provided with or as part of other hardware, but may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunications systems.

[0094] However, computer programs can also be published via networks such as the World Wide Web and downloaded from such networks to the working memory of a data processor. According to another exemplary embodiment of the invention, a medium is provided for making computer program elements downloadable, the computer program elements being arranged to perform a method according to one of the foregoing embodiments of the invention.

Claims

1. A knowledge graph generation system (100), the knowledge graph generation system comprising: - Input unit (20); and - Knowledge graph generator (10); The input unit is configured to receive multiple service reports, each of which contains natural language. The input unit is configured to provide the multiple service reports to the knowledge graph generator; The knowledge graph generator is configured to deploy a large language model (LLM). The knowledge graph generator is configured to identify information entries from the natural language in the multiple service reports using the large language model. The identified information entries include the name and type of the device, the problem that the device needs to solve, and the measures taken by service technicians to alleviate the problem. The knowledge graph generator is configured to form multiple rules using the identified information entries, wherein each rule includes a tuple consisting of a device type, a problem, and a set of disposal steps or a series of disposal steps to alleviate the problem. The knowledge graph generator is configured to generate multiple knowledge graph nodes using the multiple rules, wherein each device type, problem, and handling step is converted into a separate knowledge graph node; and The knowledge graph generator is configured to generate a knowledge graph using the plurality of knowledge graph nodes. The generation of the knowledge graph includes forming a plurality of node groups, wherein each node group includes a plurality of knowledge graph nodes linked together.

2. The knowledge graph generation system of claim 1, wherein, The multiple service reports contain natural languages ​​in various languages, including two or more of English, German, French, Italian, Spanish, Portuguese, Japanese, and Chinese.

3. The knowledge graph generation system of claim 1 or 2, wherein, Each knowledge graph node includes at least one searchable attribute or characteristic, which includes at least one of device type, problem, and handling steps.

4. The knowledge graph generation system of any one of claims 1 to 3, wherein, The input unit is configured to receive at least one new service report, each new service report containing natural language, and the input unit is configured to provide the at least one new service report to the knowledge graph generator; the knowledge graph generator is configured to use the large language model to identify information entries from the natural language in the at least one new service report, the identified information entries including the name and type of the device, the problem that the device needs to solve, and the actions taken by service technicians to alleviate the problem; the knowledge graph generator is configured to use the identified information entries to form one or more new rules, each new rule including a tuple consisting of a device, a problem, and actions or a series of actions taken to alleviate the problem; for each new rule, the knowledge graph generator is configured to use the new rule to generate multiple new knowledge graph nodes, the device type, the problem, and each action step being transformed into a separate new knowledge graph node.

5. The knowledge graph generation system according to claim 4, wherein, The knowledge graph generator is configured to retrieve the knowledge graph to determine whether a linked node group has a knowledge graph node that has the same device type as the device type associated with the new rule, and to link the plurality of new knowledge graph nodes associated with the new rule to knowledge graph nodes in a linked node group that includes knowledge graph nodes with the same device type as the device type associated with the new rule. Alternatively, if no knowledge graph node with the same device type as the device type associated with the new rule is found in the knowledge graph, a new node group including the plurality of new knowledge graph nodes is formed in the knowledge graph.

6. The knowledge graph generation system according to claim 4 or 5, wherein, The knowledge graph generator is configured to retrieve the knowledge graph to determine whether a linked node group has a knowledge graph node that has the same or similar questions as the questions associated with the new rule, and to link the multiple new knowledge graph nodes associated with the new rule to knowledge graph nodes in a linked node group that includes knowledge graph nodes with the same or similar questions as the questions associated with the new rule. Alternatively, if no knowledge graph node with the same or similar questions as the questions associated with the new rule is found in the knowledge graph, a new node group including the multiple new knowledge graph nodes is formed in the knowledge graph.

7. The knowledge graph generation system according to any one of claims 1 to 6, wherein, The input unit is configured to receive multimodal data, which is not in text form, and is configured to provide the multimodal data to the knowledge graph generator, wherein the knowledge graph generator is configured to analyze the multimodal data and / or analyze information associated with the multimodal data to determine associated node groups in the knowledge graph, and to append the multimodal data to the associated node groups.

8. The knowledge graph generation system according to any one of claims 1 to 7, wherein, The knowledge graph generation system includes an output unit (30), wherein the output unit is configured to output the knowledge graph.

9. A knowledge graph generation method, the method comprising: - The input unit receives multiple service reports, each of which contains natural language. - The input unit provides the multiple service reports to the knowledge graph generator; - Deploy large language models (LLMs) using the knowledge graph generator; - The knowledge graph generator uses a large language model to identify information entries from the natural language in the multiple service reports, wherein the identified information entries include the name and type of the device, the problem that the device needs to solve, and the measures taken by service technicians to alleviate the problem; - The knowledge graph generator uses the identified information entries to form multiple rules, wherein each rule includes a tuple consisting of a device type, a problem, and a set of disposal steps or a series of disposal steps to alleviate the problem; - The knowledge graph generator generates multiple knowledge graph nodes using the multiple rules, wherein each device type, problem, and handling step is transformed into a separate knowledge graph node; and - The knowledge graph generator generates a knowledge graph using the plurality of knowledge graph nodes, wherein generating the knowledge graph includes forming a plurality of node groups, wherein each node group includes a plurality of knowledge graph nodes linked together.

10. The method according to claim 9, wherein, Each knowledge graph node includes at least one searchable attribute or characteristic, which includes at least one of device type, problem, and handling steps.

11. The method according to claim 9 or 10, wherein, The method includes: - The input unit receives at least one new service report, wherein each new service report contains natural language; - The input unit provides the knowledge graph generator with the at least one new service report; - The knowledge graph generator uses a large language model to identify information entries from the natural language in the at least one new service report, wherein the identified information entries include the name and type of the device, the problem that the device needs to solve, and the measures taken by the service technician to alleviate the problem; - The knowledge graph generator uses the identified information entries to form one or more new rules, wherein each new rule includes a tuple consisting of a device, a problem, and a disposal step or a series of disposal steps to alleviate the problem; - For each new rule, the method includes generating multiple new knowledge graph nodes by the knowledge graph generator using the new rule, wherein the device type, problem, and each disposition step are transformed into separate new knowledge graph nodes.

12. The method according to claim 11, wherein, The method includes: - The knowledge graph is retrieved by the knowledge graph generator to determine whether the linked node group has a knowledge graph node that is a device type that is the same as the device type associated with the new rule; - The knowledge graph generator links the plurality of new knowledge graph nodes associated with the new rule to knowledge graph nodes in a group of linked nodes that include knowledge graph nodes with the same device type as the device type associated with the new rule; or - If no knowledge graph node with the same device type as the new rule is found in the knowledge graph, the knowledge graph generator forms a new node group in the knowledge graph that includes the multiple new knowledge graph nodes.

13. The method according to claim 11 or 12, wherein, The method includes: - The knowledge graph is retrieved by the knowledge graph generator to determine whether the linked node group has a knowledge graph node that has the same or similar questions as the questions associated with the new rules; - The knowledge graph generator links the plurality of new knowledge graph nodes associated with the new rule to knowledge graph nodes in a linked node group that includes knowledge graph nodes with the same or similar questions associated with the new rule; or - If no knowledge graph node with the same or similar question as the new rule is found in the knowledge graph, the knowledge graph generator forms a new node group in the knowledge graph that includes the multiple new knowledge graph nodes.

14. The method according to any one of claims 9 to 13, wherein, The method includes: - The input unit receives multimodal data, wherein the multimodal data is not in text form; - The input unit provides the multimodal data to the knowledge graph generator; - The knowledge graph generator analyzes the multimodal data and / or analyzes information associated with the multimodal data to determine associated node groups in the knowledge graph, and appends the multimodal data to the associated node groups.

15. A computer program element for controlling a processor to perform the method of any one of claims 9 to 14.

Images