Online reasoning method, device and equipment for constructing knowledge graph based on large model

Abstract

The embodiment of the invention provides an online reasoning method, device and equipment for constructing a knowledge graph based on a large model. The method comprises the steps of obtaining a preset database of a target object; wherein the preset database comprises a plurality of document slices; at least one piece of problem information corresponding to each document slice is determined, and at least one target slice meeting the similar requirement with each piece of problem information is determined from the multiple document slices; generating to-be-identified data based on the problem information and the at least one target slice; wherein the to-be-identified data indicates abstract data; and constructing a knowledge graph corresponding to the target object based on the to-be-identified data. According to the method, the knowledge graph can be automatically, comprehensively and accurately constructed.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to an online reasoning method, device and equipment for building a knowledge graph based on a large model. Background Art

[0002] With the development of computer technology and the popularization of intelligence, the application of knowledge graphs is becoming more and more extensive.

[0003] For example, the constructed knowledge graph can be used to assist the search engine in searching for search results that match the user's query content, and the constructed knowledge graph can also be used to quickly retrieve and answer the user's questions, etc. Therefore, it is very important to construct an accurate and comprehensive knowledge graph.

[0004] Generally, accurate and comprehensive knowledge graphs are constructed mainly by professional technicians. However, this implementation method is not only time-consuming but also costly. Summary of the invention

[0005] The embodiments of the present application provide an online reasoning method, device and equipment for constructing a knowledge graph based on a large model, which can quickly and accurately construct a knowledge graph, saving time and labor costs.

[0006] In a first aspect, the embodiment of the present application provides an online reasoning method for building a knowledge graph based on a large model, including:

[0007] Obtaining a preset database of the target object; wherein the preset database includes a plurality of document slices;

[0008] Determine at least one question information corresponding to each of the document slices, and determine at least one target slice from the multiple document slices that meets similar requirements as each of the question information;

[0009] Based on the problem information and the at least one target slice, generating data to be identified; wherein the data to be identified indicates summary data;

[0010] Construct a knowledge graph corresponding to the target object based on the data to be identified.

[0011] Optionally, determining at least one question information corresponding to each of the document slices includes:

[0012] Based on a first preset model, determining first semantic information corresponding to the document slice; wherein the first preset model indicates a semantic recognition model;

[0013] Based on the first semantic information, determining a first prompt word matching the document slice; wherein the first prompt word is used to prompt the generation of question information;

[0014] After the first prompt word and the document slice are combined into a first context, they are input into a second preset model to obtain at least one question information corresponding to the document slice; wherein the second preset model indicates a model that generates results based on the prompt word.

[0015] Optionally, generating the data to be identified based on the problem information and the at least one target slice includes:

[0016] Based on a first preset model, determining second semantic information corresponding to each of the target slices; wherein the first preset model indicates a semantic recognition model;

[0017] Based on the second semantic information, determining a second prompt word matching the at least one target slice; wherein the second prompt word is used to prompt generation of summary data;

[0018] After the second prompt word, the question information and the at least one target slice are combined into a second context, they are input into a second preset model to generate the data to be recognized; wherein the second preset model indicates a model that generates results based on the prompt word.

[0019] Optionally, constructing a knowledge graph corresponding to the target object based on the data to be identified includes:

[0020] Based on a first preset model, determining third semantic information corresponding to the data to be recognized; wherein the first preset model indicates a semantic recognition model;

[0021] Based on the third semantic information, determining a third prompt word matching the data to be recognized; wherein the third prompt word is used to prompt output of triple data;

[0022] After combining the third prompt word and the data to be recognized into a third context, the context is input into a second preset model to obtain triple data;

[0023] Construct a knowledge graph corresponding to the target object based on the triple data.

[0024] Optionally, constructing a knowledge graph corresponding to the target object based on the triple data includes:

[0025] Based on the triple data, construct a statement to be reviewed;

[0026] Based on the data of each word included in the sentence to be reviewed, the sentence to be reviewed is subjected to semantic evaluation processing to obtain a data evaluation result; wherein the data evaluation result is used to indicate the credibility of the triple data;

[0027] If it is determined that the credibility of the triple data meets the preset threshold requirement, a knowledge graph is constructed based on the triple data.

[0028] Optionally, the method further comprises:

[0029] If it is determined that the credibility of the triple data does not meet the preset threshold requirement, the triple data is sent to a reviewer for manual review.

[0030] Optionally, before obtaining the preset database of the target object, the method further includes:

[0031] Acquire document data associated with the target object;

[0032] Slicing the document data according to a preset slicing dimension to obtain the plurality of document slices; wherein the preset slicing dimension indicates a directory dimension and / or a semantic dimension;

[0033] The preset database is constructed based on the multiple document slices.

[0034] In a second aspect, the embodiment of the present application provides an online reasoning device for building a knowledge graph based on a large model, including:

[0035] An acquisition unit, used for acquiring a preset database of a target object; wherein the preset database includes a plurality of document slices;

[0036] A first generating unit, configured to determine at least one question information corresponding to each of the document slices, and determine at least one target slice from the multiple document slices that meets similar requirements as each of the question information;

[0037] A second generating unit, configured to generate data to be identified based on the problem information and the at least one target slice; wherein the data to be identified indicates summary data;

[0038] A construction unit is used to construct a knowledge graph corresponding to the target object based on the data to be identified.

[0039] Optionally, the first generating unit is configured to:

[0040] Based on a first preset model, determining first semantic information corresponding to the document slice; wherein the first preset model indicates a semantic recognition model;

[0041] Based on the first semantic information, determining a first prompt word matching the document slice; wherein the first prompt word is used to prompt the generation of question information;

[0042] After the first prompt word and the document slice are combined into a first context, they are input into a second preset model to obtain at least one question information corresponding to the document slice; wherein the second preset model indicates a model that generates results based on the prompt word.

[0043] Optionally, the second generating unit is used to:

[0044] Based on a first preset model, determining second semantic information corresponding to each of the target slices; wherein the first preset model indicates a semantic recognition model;

[0045] Based on the second semantic information, determining a second prompt word matching the at least one target slice; wherein the second prompt word is used to prompt generation of summary data;

[0046] After the second prompt word, the question information and the at least one target slice are combined into a second context, they are input into a second preset model to generate the data to be recognized; wherein the second preset model indicates a model that generates results based on the prompt word.

[0047] Optionally, the building unit includes a generation module for:

[0048] Based on a first preset model, determining third semantic information corresponding to the data to be recognized; wherein the first preset model indicates a semantic recognition model;

[0049] Based on the third semantic information, determining a third prompt word matching the data to be recognized; wherein the third prompt word is used to prompt output of triple data;

[0050] After combining the third prompt word and the data to be recognized into a third context, the context is input into a second preset model to obtain triple data;

[0051] Construct a knowledge graph corresponding to the target object based on the triple data.

[0052] Optionally, the building unit includes an audit module for:

[0053] Based on the triple data, construct a statement to be reviewed;

[0054] Based on the data of each word included in the sentence to be reviewed, the sentence to be reviewed is subjected to semantic evaluation processing to obtain a data evaluation result; wherein the data evaluation result is used to indicate the credibility of the triple data;

[0055] If it is determined that the credibility of the triple data meets the preset threshold requirement, a knowledge graph is constructed based on the triple data.

[0056] Optionally, the device further includes a sending unit, configured to:

[0057] If it is determined that the credibility of the triple data does not meet the preset threshold requirement, the triple data is sent to a reviewer for manual review.

[0058] Optionally, the device further comprises an initialization unit, configured to:

[0059] Before acquiring the preset database of the target object, acquiring document data associated with the target object;

[0060] Slicing the document data according to a preset slicing dimension to obtain the plurality of document slices; wherein the preset slicing dimension indicates a directory dimension and / or a semantic dimension;

[0061] The preset database is constructed based on the multiple document slices.

[0062] In a third aspect, an embodiment of the present application provides a computer device, including: a memory, a processor;

[0063] The memory stores computer-executable instructions;

[0064] The processor executes the computer-executable instructions stored in the memory, so that the processor executes the above first aspect and / or various possible implementations of the first aspect.

[0065] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the first aspect above and / or various possible implementations of the first aspect.

[0066] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the above first aspect and / or various possible implementation methods of the first aspect.

[0067] The online reasoning method, device and equipment for building a knowledge graph based on a large model provided in the embodiment of the present application can obtain a preset database of the target object, determine at least one question information corresponding to each document slice included in the preset database, and determine at least one target slice that meets the similarity requirements with each question information, and then generate the data to be identified based on the question information and at least one target slice. This implementation method can expand the content included in the document slice by generating question information on the basis of the original document slice, and filter out at least one target slice that meets the similarity requirements from multiple document slices according to the question information, that is, filter out at least one target slice that has an association relationship with the current document slice. At this time, the data to be identified generated based on the question information and at least one target slice can include more complete, accurate and concise entities and relationships. Afterwards, the knowledge graph corresponding to the target object is constructed based on the data to be identified, which can not only realize the automatic construction of a complete, accurate and concise knowledge graph, but also further improve the online reasoning speed of building a knowledge graph based on a large model. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0069] Figure 1 A flowchart of an online reasoning method for building a knowledge graph based on a large model provided in this application;

[0070] Figure 2 A flowchart of another online reasoning method for building a knowledge graph based on a large model provided in this application;

[0071] Figure 3 A schematic diagram of the process of an online reasoning system for building a knowledge graph based on a large model provided in this application;

[0072] Figure 4 A schematic diagram of the structure of an online reasoning device for building a knowledge graph based on a large model provided in this application;

[0073] Figure 5 A schematic diagram of the structure of another online reasoning device for building a knowledge graph based on a large model provided in this application;

[0074] Figure 6 A schematic diagram of the structure of a computer device provided in this application.

[0075] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0076] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0077] First, the terms involved in this application are explained:

[0078] Knowledge graph: A structured semantic knowledge base that is designed to store and represent real-world entities (such as people, places, organizations, etc.) and their relationships in a machine-readable manner. In this case, the knowledge graph organizes knowledge into a graph structure so that knowledge can be more effectively managed and reasoned by computer systems.

[0079] Triple: The most basic unit of knowledge in the knowledge graph, used to represent the association and attributes between entities. A triple is mainly composed of three parts: subject, predicate, and object. These three parts together describe a fact or relationship, and are usually used to express a certain connection between two entities.

[0080] As the application of knowledge graph reasoning in real-world scenarios continues to become more popular, how to build accurate and comprehensive knowledge graphs has become a technical problem that needs to be solved urgently.

[0081] For example, in the coal preparation industry, the coal preparation process involves many process steps, equipment, parameters, etc., and has high complexity. Therefore, effective management of the system in the coal preparation industry can be achieved by constructing a knowledge graph. At the same time, after constructing the corresponding knowledge graph, the complex production system can be intelligently analyzed and optimized through knowledge graph reasoning, thereby improving overall operational efficiency and competitiveness.

[0082] Generally, knowledge graphs can be constructed by professional technicians to ensure the accuracy and comprehensiveness of the constructed knowledge graphs, and further ensure the accuracy and reliability of reasoning based on the knowledge graphs.

[0083] However, this implementation method requires high manpower and time costs when the amount of data to be constructed of the knowledge graph is large, which in turn affects the efficiency of knowledge graph construction.

[0084] The online reasoning method for constructing a knowledge graph based on a large model provided in the present application can first generate corresponding question information according to the document slices associated with the target object, and then automatically construct the knowledge graph according to the generated question information and the corresponding target slices, thereby realizing the expansion of the data used to construct the knowledge graph, thereby ensuring the comprehensiveness, accuracy and efficiency of constructing the knowledge graph.

[0085] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0086] Figure 1 A flowchart of an online reasoning method for building a knowledge graph based on a large model provided in this application is shown in FIG. Figure 1 As shown, the method includes:

[0087] S101: Acquire a preset database of a target object.

[0088] The preset database includes multiple document slices.

[0089] In one example, the target object may indicate the application scenario of the knowledge graph. For example, the application scenario indicated by the target object includes but is not limited to the coal preparation scenario. Alternatively, the target object may also indicate the business system to which the knowledge graph is applied. For example, the business system indicated by the target object includes but is not limited to: enterprise management system, production system, etc. The target object is not specifically limited here, and is subject to meeting actual needs.

[0090] In one example, the document slices included in the preset database are obtained by slicing the document data related to the target object according to the text slicing model. At this time, the obtained document slices can be saved in the database to obtain the preset database. Among them, the text slicing model can be any model that can realize text slicing. The text slicing model is not limited here. For example, the text slicing model can be a TextTiling model.

[0091] In one example, after obtaining the document slices, the document slices can be converted into slice vectors according to the text embedding model and stored. At this time, the preset database may include not only the document slices but also the slice vectors corresponding to each document slice.

[0092] S102: Determine at least one piece of question information corresponding to each document slice, and determine at least one target slice from multiple document slices that meets similar requirements as each piece of question information.

[0093] In an example, the number of question information corresponding to each document slice may be 1, 5, 8, 10, etc., and the number of question information corresponding to the determined document slice is not limited here.

[0094] In one example, when determining at least one question information corresponding to each document slice, the initial questions corresponding to each document slice that meet the first quantity requirement can be determined first, and then the first number of initial questions can be merged to obtain at least one question information corresponding to the document slice. Optionally, when merging the first number of initial questions, the initial questions can be cleaned first, including but not limited to removing noise, errors and irrelevant information of the initial questions, so as to ensure the accuracy of the obtained initial questions. Afterwards, the cleaned initial questions are vectorized to determine the similarity between each initial question. At this time, the initial questions can be clustered according to the similarity between each initial question. Afterwards, the initial questions belonging to the same category can be semantically merged to obtain at least one question information.

[0095] In one example, after obtaining the question information, at least one target slice that meets the similarity requirement with each question information can be determined based on the similarity between the question information and each document slice. Optionally, when determining the similarity between the question information and each document slice, it can be determined by calculating the distance between the question vector corresponding to the question information and the slice vector corresponding to the document slice.

[0096] In one example, the similarity requirement can be understood as the N document slices that are most similar to the question information. At this time, after determining the similarity between each document slice and the question information, they can be sorted according to the similarity, and based on the sorting results, N target slices can be determined from multiple document slices, where the value of N is a natural number greater than 0. For example, N can be 10, or 5, etc. There is no specific limitation on the value of N here.

[0097] S103: Generate data to be identified based on the problem information and at least one target slice.

[0098] The data to be identified indicates summary data.

[0099] In one example, a target document may be combined according to question information and at least one target slice, and the data to be identified may be obtained by extracting a summary corresponding to the target document.

[0100] S104. Construct a knowledge graph corresponding to the target object based on the data to be identified.

[0101] In one example, entity relationships can be extracted from the data to be identified to construct a knowledge graph corresponding to the target object.

[0102] It can be seen from the above description that the online reasoning method for constructing a knowledge graph based on a large model provided in the embodiment of the present application can obtain a preset database of the target object, determine at least one question information corresponding to each document slice included in the preset database, and determine at least one target slice that meets similar requirements with each question information, and then generate data to be identified based on the question information and at least one target slice. This implementation method can expand the content included in the document slice by generating question information on the basis of the original document slice, and filter out at least one target slice that meets similar requirements from multiple document slices according to the question information, that is, filter out at least one target slice that has an association relationship with the current document slice. At this time, the data to be identified generated based on the question information and at least one target slice can include more complete, accurate and concise entities and relationships. Afterwards, the knowledge graph corresponding to the target object is constructed based on the data to be identified, which can not only realize the automatic construction of a complete, accurate and concise knowledge graph, but also further improve the online reasoning speed of constructing a knowledge graph based on a large model.

[0103] Figure 2 A flowchart of another online reasoning method for building a knowledge graph based on a large model provided in this application is as follows: Figure 2 As shown, the method includes:

[0104] S201: Acquire a preset database of a target object.

[0105] The preset database includes multiple document slices.

[0106] In one example, before obtaining the preset database of the target object, a preset database may be constructed first.

[0107] During specific implementation, the document data associated with the target object may be obtained first.

[0108] Then, the document data is sliced ​​according to a preset slice dimension to obtain a plurality of document slices, wherein the preset slice dimension indicates a directory dimension and / or a semantic dimension.

[0109] Finally, a preset database is built based on multiple document slices.

[0110] In one example, the document data associated with the target object may come from books, journals, papers, internal materials, network documents, etc. In this case, the document data may be data in Word format, data in PDF format, data in Excel format, etc. There is no limitation on the source and type of the document data.

[0111] In one example, when a directory exists in the document data, the document data may be sliced ​​according to the directory structure; or, when a target does not exist in the document data, the document data may be sliced ​​according to semantic information.

[0112] In one example, a slice length threshold may be pre-set so that the document data may be sliced ​​according to the slice length threshold. For example, if the document data has a directory and the length of the document in a single directory is relatively long, the document data may be sliced ​​according to the length threshold to be sliced ​​and the semantic information after the document data is sliced ​​according to the directory structure. Alternatively, if the document data does not have a directory and the document length is relatively long, the document data may be sliced ​​according to the length threshold to be sliced ​​and the semantic information of the document data.

[0113] In one example, after obtaining the document slices, a preset database may be constructed based on the document slices, or a preset database may be constructed based on the document slices and the slice vectors corresponding to the document slices.

[0114] S202: Determine first semantic information corresponding to the document slice based on a first preset model.

[0115] Among them, the first preset model indicates a semantic recognition model.

[0116] Optionally, the semantic recognition model indicated by the first preset model may be a word embedding model, in which case the document slice may be recognized and processed according to the word embedding model to obtain the first semantic information corresponding to the document slice. The specific model structure of the word embedding model is not limited here, and is subject to implementation.

[0117] S203: Determine a first prompt word that matches the document slice based on the first semantic information.

[0118] The first prompt word is used to prompt the generation of question information.

[0119] In one example, the keyword corresponding to the document slice can be determined based on the first semantic information corresponding to the document slice, and the domain information to which the document slice belongs can be determined based on the keyword. Then, the first prompt word can be determined based on the preset prompt word corresponding to the domain information to which the document slice belongs and used to prompt the generation of questions.

[0120] In one example, the preset prompt word for prompting the generation of a question may include at least one prompt item, and each prompt item may include corresponding prompt content. For example, the prompt item may be at least one of: a role and a task. At this time, the preset prompt word for prompting the generation of a question may be obtained by setting the prompt content corresponding to each prompt item.

[0121] For example, if the target object indicates a coal preparation scenario, then the domain information to which the document slice of the target object belongs may include, but is not limited to: coal quality domain, production domain, electromechanical domain, safety domain, etc. At this time, a corresponding preset prompt word can be set for each domain information. For example, if the domain information is "electromechanical domain", then the preset prompt word corresponding to "electromechanical domain" for generating questions (assuming that the preset prompt word includes two prompt items: role and task) can be:

[0122] “Role”: An expert in mechanical and electrical questions at a coal preparation plant.

[0123] "Task":

[0124] 1) You are good at knowledge questions and answers in the electromechanical field of coal preparation plants. Please generate 8 questions based on the content.

[0125] 2) Read and understand the text content related to the field of electromechanical in the original text.

[0126] 3) Identify key information and potential electromechanical problems in the text. When you identify relevant problems, you should ask questions in the structure of (equipment name + fault phenomenon + what is the cause of the fault, what are its solutions and preventive measures). For example: What are the reasons for the occurrence of X problem in X equipment? What are its solutions and preventive measures?

[0127] 4) Extract specific issues based on different aspects of the equipment in the coal preparation plant, including mechanical, electrical and automation.

[0128] 5) Prioritize questions based on the main issues addressed in the article.

[0129] 6) Organize and present the list of questions.

[0130] 7) In the content, if there is a closer description of the problem, please try to output the relevant issues in accordance with the original text.

[0131] S204: After combining the first prompt word and the document slice into a first context, the first context is input into a second preset model to obtain at least one question information corresponding to the document slice.

[0132] Among them, the second preset model indicates a model that generates results according to the prompt word.

[0133] In one example, the first prompt word and the document slice can be combined according to the prompt content corresponding to the first prompt word to obtain a first context. Then, the first context is input into the second preset model to generate a first number of initial questions corresponding to the document slice. After that, the initial questions of the first data are merged to obtain at least one question information corresponding to the document slice. The merging process can refer to the content described in S102 above, which will not be described in detail here.

[0134] Optionally, the second preset model indicates a model that can generate text based on the text. For example, the second preset model can include but is not limited to the Qianwen model.

[0135] In the above implementation, the first semantic information of the document slice can be extracted by the first preset model, and the first prompt word matching the document slice can be determined according to the domain information to which the first semantic information belongs, so that the first prompt word that better matches the document slice can be determined. Afterwards, the second preset model can be accurately guided to generate question information corresponding to the document slice according to the first prompt word, thereby improving the accuracy of the generated question information.

[0136] S205: Determine at least one target slice from the multiple document slices that meets similar requirements as each question information.

[0137] In an example, this step can refer to the content described in S102 above, and will not be described in detail here.

[0138] S206: Determine second semantic information corresponding to each target slice based on the first preset model.

[0139] Among them, the first preset model indicates a semantic recognition model.

[0140] The first preset model here is the same as the first preset model mentioned above, and will not be repeated here.

[0141] S207: Determine a second prompt word matching at least one target slice based on the second semantic information.

[0142] The second prompt word is used to prompt the generation of summary data.

[0143] In one example, the domain information to which at least one target slice belongs may be determined based on the second semantic information, and then the second prompt word may be determined based on a preset prompt word corresponding to the domain information and used to prompt generation of summary data.

[0144] In one example, the preset prompt word for prompting the generation of summary data may include at least one prompt item, and each prompt item may include corresponding prompt content. For example, the prompt item may be at least one of: a role and a task. At this time, the preset prompt word for prompting the generation of summary data may be obtained by setting the prompt content corresponding to each prompt item.

[0145] For example, if the target object indicates a coal preparation scene, then the domain information to which at least one target slice belongs may include but is not limited to: coal quality domain, production domain, electromechanical domain, safety domain, etc. At this time, if the domain information to which at least one target slice belongs is "electromechanical domain", then the second prompt word matching at least one target slice (that is, the preset prompt word corresponding to "electromechanical domain" for prompting the generation of summary data, assuming that the preset prompt word includes two prompt items, namely, role and task) may be:

[0146] “Role”: Summary expert of mechanical and electrical engineering in coal preparation plant.

[0147] "Task":

[0148] 1) You are good at summarizing the knowledge in the electromechanical field of coal preparation plants. Please generate a summary sentence of no more than 300 words based on the content.

[0149] 2) Identify the key information in the text. Once the question is identified, fill in the answer and then describe it.

[0150] S208: After combining the second prompt word, the question information and at least one target slice into a second context, the context is input into a second preset model to generate data to be recognized.

[0151] Among them, the second preset model indicates a model that generates results according to the prompt word.

[0152] In one example, at least one target slice and at least one target slice may be combined according to the second prompt word to obtain a second context.

[0153] In one example, the second preset model can be understood as a model having the same structure as the second preset model for generating question information described above, but is used to generate different results, which will not be described in detail here.

[0154] In the above implementation, the second semantic information corresponding to the at least one target slice can be used to determine the second prompt word matching the at least one target slice, so that the second preset model can be accurately guided to generate the data to be identified according to the customized second prompt word, thereby improving the accuracy of the generated data to be identified.

[0155] S209: Determine third semantic information corresponding to the data to be identified based on the first preset model.

[0156] Among them, the first preset model indicates a semantic recognition model.

[0157] The first preset model here is the same as the first preset model mentioned above, and will not be repeated here.

[0158] S210: Determine a third prompt word that matches the data to be recognized based on the third semantic information.

[0159] The third prompt word is used to prompt the output of triple data.

[0160] In one example, the domain information to which the data to be identified belongs may be determined based on the third semantic information, and then the third prompt word may be determined based on a preset prompt word corresponding to the domain information and used to prompt the output of triple data.

[0161] In one example, the preset prompt word for prompting the output of triple data may include at least one prompt item, and each prompt item may include corresponding prompt content. For example, the prompt item may be at least one of a role and a task. At this time, the preset prompt word for prompting the output of triple data may be obtained by setting the prompt content corresponding to the role and the task respectively.

[0162] For example, if the target object indicates a coal preparation scene, then the domain information to which the data to be identified belongs may include but is not limited to: coal quality domain, production domain, electromechanical domain, safety domain, etc. At this time, if the domain information to which the data to be identified belongs is "electromechanical domain", then the third prompt word matching the data to be identified (that is, the preset prompt word corresponding to "electromechanical domain" for prompting the output of triple data, assuming that the preset prompt word only includes task prompt items) can be:

[0163] "Task": Based on the given text content, extract all qualified head entities, head entity types, relationship types, tail entities, and tail entity types. Your generated content must meet all of the following conditions:

[0164] 1. The head entity and the tail entity must be original text fragments clearly mentioned in the document. They cannot be randomly summarized or elaborated, and the intermediate content cannot be omitted.

[0165] 2. The entity type only includes Entity_fault phenomenon, Entity_fault cause, Entity_solution, Entity_cause category, Entity_preventive measures, Entity_phenomenon category, Entity_equipment type, and Entity_equipment.

[0166] (1).Entity_Fault phenomenon: The specific fault phenomenon of a certain device or device type. For example: xx current is overloaded, xx has a pressure drop.

[0167] (2).Entity_Fault Cause: The specific cause of a fault phenomenon. For example: xx fault, xx bearing damage.

[0168] (3). Entity_Solution: The solution to a specific cause. For example: Replace xx, clear xx, adjust xx.

[0169] (4).Entity_Reason Category: xx.

[0170] (5). Entity_Preventive measures: preventive measures for a certain fault phenomenon. For example: check xx regularly and replace xx regularly.

[0171] (6).Entity_phenomenon category: xx.

[0172] (7).Entity_Equipment Type: Equipment type name. For example: horizontal centrifuge, heavy medium cyclone.

[0173] (8).Entity_Equipment: The equipment number in actual production.

[0174] 3. Relationship types only include preventive measures, causes, belonging to, solutions, and phenomena.

[0175] 4. The entity type corresponding to the head entity, the tail entity type corresponding to the tail entity, and the relationship type can only have the following sets of construction patterns. You must strictly output according to the following construction patterns. The types of the same entity in different JSONs must be the same.

[0176] (1). Head entity type: Entity_Fault phenomenon, relationship type: cause, tail entity type: Entity_Fault cause.

[0177] (2). Head entity type: Entity_Failure Cause, relationship type: Solution, tail entity type: Entity_Solution.

[0178] (3). Head entity type: Entity_Failure Cause, relationship type: Belongs to, tail entity type: Entity_Cause Category.

[0179] (4). Head entity type: Entity_Failure Cause, relationship type: Preventive Measures, tail entity type: Entity_Preventive Measures.

[0180] (5). Head entity type: Entity_fault phenomenon, relationship type: belongs to, tail entity type: Entity_phenomenon category.

[0181] (6). Head entity type: Entity_Equipment type, relationship type: phenomenon, tail entity type: Entity_fault phenomenon.

[0182] (7). Head entity type: Entity_Equipment, relationship type: phenomenon, tail entity type: Entity_fault phenomenon.

[0183] 5. The head entity type, relationship type, and tail entity type you output must belong to one of the items in 3. They cannot be freely combined. You must fully understand the given text content and output it strictly according to the construction mode.

[0184] 6. For each JSON line in the output, you must be able to find the corresponding head entity, tail entity, and relationship type in the given text to prove that the relationship between them exists. Your output must be a correct pair.

[0185] 7. Output format: a list composed of json. Be careful not to output duplicate content and output a complete list: [

[0187] {"Head Entity": "", "Head Entity Type": "", "Relationship Type": "", "Tail Entity": "", "Tail Entity Type": ""},

[0188] {"Head Entity": "", "Head Entity Type": "", "Relationship Type": "", "Tail Entity": "", "Tail Entity Type": ""}, ...

[0190] {"Head Entity": "", "Head Entity Type": "", "Relationship Type": "", "Tail Entity": "", "Tail Entity Type": ""} ]

[0192] It should be noted here that the prompt items included in the above-mentioned preset prompt words and the prompt content corresponding to each prompt item are only exemplary, and this application does not make any specific limitations on this, and is subject to meeting actual needs.

[0193] S211 , after combining the third prompt word and the data to be recognized into a third context, input it into a second preset model to obtain triple data.

[0194] In one example, the third prompt word and the data to be recognized may be concatenated to obtain a third context, and the third context may be input into the second preset model for processing.

[0195] In one example, when the third prompt word and the data to be recognized are concatenated, the concatenation may be performed in the following format: [

[0197] The third prompt word

[0198] text:

[0199] {Input Document}

[0200] Output: ]

[0202] The content corresponding to the "{input document}" part is the data to be identified.

[0203] For example, assume that the data to be identified is: the most common inspection component of the vibrating screen is the vibrator, which belongs to the drive part. The inspection items include the bearing seat and the bearing cover. The inspection standard is no cracks and tight assembly. The inspection cycle is 1 week and the inspection requirement is that the machine is in the shutdown state.

[0204] Then, after the above prompt words and the data to be recognized are input into the second preset model, the output may be as follows: [

[0206] {

[0207] "Head Entity": "Vibrating Screen",

[0208] "Header Entity Type": "Entity_Device Type",

[0209] "Relation Type": "Inspection Parts",

[0210] "Tail Entity": "Exciter",

[0211] "Tail Entity Type": "Entity_Inspection Parts"

[0212] },

[0213] {

[0214] "Head Entity": "Exciter",

[0215] "Header Entity Type": "Entity_Inspection Part",

[0216] "Relation Type": "Belongs To",

[0217] "Tail Entity": "Drive Part",

[0218] "Tail Entity Type": "Entity_Inspection Part Category"

[0219] },

[0220] {

[0221] "Head Entity": "Exciter",

[0222] "Header Entity Type": "Entity_Inspection Part",

[0223] "Relation Type": "Inspection Content",

[0224] "Tail Entity": "Bearing Seat, Bearing Cover",

[0225] "Tail Entity Type": "Entity_Inspection Content"

[0226] },

[0227] {

[0228] "Head entity": "Bearing seat, bearing cover",

[0229] "Header Entity Type": "Entity_Inspection Content",

[0230] "Relationship Type": "Standard Compliant",

[0231] "Tail Entity": "No cracks, assembly is tight",

[0232] "Tail Entity Type": "Entity_Inspection Standard"

[0233] },

[0234] {

[0235] "Head entity": "Bearing seat, bearing cover",

[0236] "Header Entity Type": "Entity_Inspection Content",

[0237] "Relationship Type": "Inspection Cycle",

[0238] "Tail Entity": "1 Week",

[0239] "Tail Entity Type": "Entity_Inspection Cycle"

[0240] },

[0241] {

[0242] "Head entity": "Bearing seat, bearing cover",

[0243] "Header Entity Type": "Entity_Inspection Content",

[0244] "Relation Type": "Inspection Request",

[0245] "Tail Entity": "Stop",

[0246] "Tail Entity Type": "Entity_Inspection Requirement"

[0247] } ]

[0249] In the above implementation, the preset prompt word can be determined by the domain information to which the data to be identified belongs, and then the third prompt word can be determined, so that the format and description of the generated triple data can be standardized according to the domain information, so that accurate triple data can be obtained while making the obtained triple data more standardized, thereby making the generated triple data more readable.

[0250] S212. Construct a knowledge graph corresponding to the target object based on the triple data.

[0251] In one example, triple data can be mapped into points and edges according to a graph data structure (e.g., a graph database) to obtain a knowledge graph.

[0252] In a possible implementation manner, the above S212 may specifically include the following process:

[0253] First, based on the triple data, construct the statement to be reviewed.

[0254] In one example, the head entity, relationship, and tail entity included in the triple data can be spliced ​​to obtain a statement to be reviewed. For example, if the head entity is "heavy medium cyclone", the relationship is "phenomenon", and the tail entity is "pressure instability", then the statement to be reviewed after splicing can be "heavy medium cyclone phenomenon pressure instability".

[0255] Then, based on the data of each word included in the sentence to be reviewed, a semantic evaluation process is performed on the sentence to be reviewed to obtain a data evaluation result.

[0256] The data evaluation result is used to indicate the credibility of the triplet data.

[0257] In one example, the sentence to be reviewed can be input into the first preset model, so that the sentence to be reviewed is encoded based on the first preset model to obtain the word vectors corresponding to the words included in the sentence to be reviewed. After that, the similarity of each word vector corresponding to the sentence to be reviewed can be judged to obtain the data evaluation result. For example, the similarity between each word vector can be calculated first, and then the weighted average of each similarity can be obtained to obtain the data evaluation result. Optionally, the similarity between word vectors can be determined based on cosine similarity.

[0258] In one example, after obtaining the data evaluation result, the data evaluation result may be converted (for example, the data evaluation result may be converted through a fully connected layer and a sigmoid function) to obtain a data evaluation result between 0 and 1, thereby making the data evaluation result more concise and intuitive.

[0259] Finally, if it is determined that the credibility of the triple data meets the preset threshold requirement, a knowledge graph is constructed based on the triple data.

[0260] If it is determined that the credibility of the triplet data does not meet the preset threshold requirement, the triplet data is sent to the auditor for manual review.

[0261] Optionally, the preset threshold requirement may indicate that the credibility of the triplet data is greater than a preset threshold. For example, the preset threshold may be set to 0.5 or 0.8, etc., and the value of the preset threshold is not limited here.

[0262] In the above implementation, the accuracy of the triple data used to construct the knowledge graph can be improved by calculating the credibility of the triple data, thereby improving the accuracy of the knowledge graph. At the same time, when the triple data does not meet the preset threshold requirements, the triple data is sent to the reviewer for manual review. If the reviewer determines that the triple data is accurate, a second preset model can be trained based on the triple to further improve the accuracy of the second preset model. On the other hand, if the reviewer determines that the triple data is wrong, the prompt word can be adjusted according to the triple data to ensure that it helps to further improve the accuracy of the generated triple data, thereby improving the accuracy of the constructed knowledge graph.

[0263] Figure 3 A flowchart of an online reasoning system for building a knowledge graph based on a large model provided in this application is as follows: Figure 3 As shown, the online reasoning system for constructing a knowledge graph based on a large model may include: a database construction module, a question generation module, a question merging module, a question and answer aggregation module, a data extraction module, and an audit entry module.

[0264] The database construction module is used to obtain document data associated with the target object, and to slice the document data to obtain multiple document slices. After that, the document slices can be vectorized to obtain slice vectors. Then, a preset database can be constructed based on the document slices and slice vectors.

[0265] The question generation module is used to determine the domain information to which the document slice belongs according to the first preset model, and determine the first prompt word matching the document slice according to the domain information. After that, after combining the first prompt word and the document slice into a first context, the second preset model is queried to obtain a plurality of initial questions corresponding to the document slice.

[0266] The question merging module is used to perform vectorization processing on the cleaned initial questions after cleaning the multiple initial questions, so as to determine the similarity between the initial questions. At this time, the initial questions can be clustered according to the similarity between the initial questions. After that, the initial questions belonging to the same category can be semantically merged to obtain at least one question information.

[0267] The question-answer aggregation module is used to determine N target slices that meet similar requirements with the question information based on multiple slice vectors in the preset database after vectorizing the question information. After that, the second prompt word matching the N target slices can be determined based on the first preset model, and after the second prompt word, the N target slices, and the question information are combined into a second context, the second preset model is queried to generate data to be recognized.

[0268] The data extraction module is used to determine the domain information to which the data to be identified belongs according to the first preset model, and then determine the third prompt word matching the data to be identified. After that, the data to be identified and the third prompt word can be combined into a third context, and then the second preset model can be queried to obtain triple data.

[0269] The review input module is used to construct the statement to be reviewed based on the triple data, so as to obtain the data evaluation result indicating the credibility of the triple data after evaluating the statement to be reviewed. At this time, if it is determined based on the data evaluation result that the credibility of the triple data meets the preset threshold requirement, a knowledge graph is constructed based on the triple data. If it is determined based on the data evaluation result that the credibility of the triple data does not meet the preset threshold requirement, the triple data is sent to the reviewer for manual review.

[0270] Figure 4 A schematic diagram of the structure of an online reasoning device for building a knowledge graph based on a large model provided in this application, such as Figure 4 As shown, the online reasoning device 400 for building a knowledge graph based on a large model provided in this embodiment includes:

[0271] The acquisition unit 401 is used to acquire a preset database of the target object; wherein the preset database includes a plurality of document slices.

[0272] The first generating unit 402 is used to determine at least one question information corresponding to each document slice, and determine at least one target slice that meets similar requirements as each question information from multiple document slices.

[0273] The second generating unit 403 is used to generate data to be identified based on the problem information and at least one target slice; wherein the data to be identified indicates summary data.

[0274] The construction unit 404 is used to construct a knowledge graph corresponding to the target object based on the data to be identified.

[0275] Figure 5 A schematic diagram of the structure of another online reasoning device for building a knowledge graph based on a large model provided in this application, such as Figure 5 As shown, the online reasoning device 500 for building a knowledge graph based on a large model provided in this embodiment includes:

[0276] The acquisition unit 501 is used to acquire a preset database of a target object; wherein the preset database includes a plurality of document slices.

[0277] The first generating unit 502 is used to determine at least one question information corresponding to each document slice, and determine at least one target slice that meets similar requirements as each question information from multiple document slices.

[0278] The second generating unit 503 is used to generate data to be identified based on the problem information and at least one target slice; wherein the data to be identified indicates summary data.

[0279] The construction unit 504 is used to construct a knowledge graph corresponding to the target object based on the data to be identified.

[0280] Optionally, the first generating unit 502 is configured to:

[0281] Based on a first preset model, determining first semantic information corresponding to the document slice; wherein the first preset model indicates a semantic recognition model;

[0282] Based on the first semantic information, determining a first prompt word that matches the document slice; wherein the first prompt word is used to prompt the generation of question information;

[0283] After the first prompt word and the document slice are combined into a first context, they are input into a second preset model to obtain at least one question information corresponding to the document slice; wherein the second preset model indicates a model that generates results based on the prompt word.

[0284] Optionally, the second generating unit 503 is configured to:

[0285] Based on the first preset model, determining the second semantic information corresponding to each target slice; wherein the first preset model indicates a semantic recognition model;

[0286] Based on the second semantic information, determining a second prompt word matching at least one target slice; wherein the second prompt word is used to prompt generation of summary data;

[0287] After the second prompt word, question information and at least one target slice are combined into a second context, they are input into a second preset model to generate data to be recognized; wherein the second preset model indicates a model that generates results based on the prompt word.

[0288] Optionally, the construction unit 504 includes a generation module 5041, which is used to:

[0289] Based on the first preset model, determining third semantic information corresponding to the data to be recognized; wherein the first preset model indicates a semantic recognition model;

[0290] Based on the third semantic information, determining a third prompt word matching the data to be recognized; wherein the third prompt word is used to prompt output of triple data;

[0291] After combining the third prompt word and the data to be recognized into a third context, the context is input into a second preset model to obtain triple data;

[0292] Construct the knowledge graph corresponding to the target object based on triple data.

[0293] Optionally, the construction unit 504 includes an audit module 5042, which is used to:

[0294] Based on the triple data, construct the statement to be reviewed;

[0295] Based on the data of each word included in the sentence to be reviewed, semantic evaluation processing is performed on the sentence to be reviewed to obtain a data evaluation result; wherein the data evaluation result is used to indicate the credibility of the triple data;

[0296] If it is determined that the credibility of the triple data meets the preset threshold requirement, a knowledge graph is constructed based on the triple data.

[0297] Optionally, the device further includes a sending unit 505, configured to:

[0298] If it is determined that the credibility of the triplet data does not meet the preset threshold requirement, the triplet data is sent to the auditor for manual review.

[0299] Optionally, the device further includes an initialization unit 506, configured to:

[0300] Before obtaining the preset database of the target object, obtaining document data associated with the target object;

[0301] Slice the document data according to a preset slicing dimension to obtain a plurality of document slices; wherein the preset slicing dimension indicates a directory dimension and / or a semantic dimension;

[0302] Build a preset database based on multiple document slices.

[0303] The online reasoning device for building a knowledge graph based on a large model provided in this embodiment can execute the online reasoning method for building a knowledge graph based on a large model provided in the above method embodiment. Its implementation principle and technical effects are similar, and this embodiment will not be repeated here.

[0304] Figure 6 This is a schematic diagram of the structure of a computer device provided in this application. Figure 6 As shown, the computer device 600 provided in this embodiment includes: at least one processor 601 and a memory 602. Optionally, the computer device 600 also includes a communication component 603. The processor 601, the memory 602 and the communication component 603 are connected via a bus 604.

[0305] In a specific implementation process, at least one processor 601 executes the computer execution instructions stored in the memory 602, so that at least one processor 601 executes the above method.

[0306] The specific implementation process of the processor 601 can be found in the above method embodiment, and its implementation principle and technical effect are similar, so this embodiment will not be repeated here.

[0307] In the above embodiments, it should be understood that the processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the invention can be directly implemented as a hardware processor, or can be implemented by a combination of hardware and software modules in the processor.

[0308] The memory may include a high-speed memory (Random Access Memory, RAM), and may also include a non-volatile memory (NVM), such as at least one disk storage.

[0309] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of this application is not limited to only one bus or one type of bus.

[0310] The present application also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0311] The present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the above method is implemented.

[0312] The above-mentioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0313] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (Application Specific Integrated Circuits, referred to as: ASIC). Of course, the processor and the readable storage medium can also exist in the device as discrete components.

[0314] The division of units is only a logical function division, and there may be other divisions in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0315] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0316] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0317] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0318] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0319] Finally, it should be noted that those skilled in the art will readily conceive of other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention, which follow the general principles of the present invention and include common knowledge or customary technical means in the art not disclosed by the present invention, are not limited to the precise structure described above and shown in the drawings, and may be modified and changed in various ways without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. An online reasoning method for building a knowledge graph based on a large model, characterized in that: include: Obtaining a preset database of the target object; wherein the preset database includes a plurality of document slices; Determine at least one question information corresponding to each of the document slices, and determine at least one target slice from the multiple document slices that meets similar requirements as each of the question information; Based on the problem information and the at least one target slice, generating data to be identified; wherein the data to be identified indicates summary data; Construct a knowledge graph corresponding to the target object based on the data to be identified.

2. The method according to claim 1, characterized in that Determining at least one question information corresponding to each of the document slices includes: Based on a first preset model, determining first semantic information corresponding to the document slice; wherein the first preset model indicates a semantic recognition model; Based on the first semantic information, determining a first prompt word matching the document slice; wherein the first prompt word is used to prompt the generation of question information; After the first prompt word and the document slice are combined into a first context, they are input into a second preset model to obtain at least one question information corresponding to the document slice; wherein the second preset model indicates a model that generates results based on the prompt word.

3. The method according to claim 1, characterized in that Generating to-be-identified data based on the problem information and the at least one target slice includes: Based on a first preset model, determining second semantic information corresponding to each of the target slices; wherein the first preset model indicates a semantic recognition model; Based on the second semantic information, determining a second prompt word matching the at least one target slice; wherein the second prompt word is used to prompt generation of summary data; After the second prompt word, the question information and the at least one target slice are combined into a second context, they are input into a second preset model to generate the data to be recognized; wherein the second preset model indicates a model that generates results based on the prompt word.

4. The method according to claim 1, characterized in that: Constructing a knowledge graph corresponding to the target object based on the data to be identified, including: Based on a first preset model, determining third semantic information corresponding to the data to be recognized; wherein the first preset model indicates a semantic recognition model; Based on the third semantic information, determining a third prompt word matching the data to be recognized; wherein the third prompt word is used to prompt output of triple data; After combining the third prompt word and the data to be recognized into a third context, the context is input into a second preset model to obtain triple data; Construct a knowledge graph corresponding to the target object based on the triple data.

5. The method according to claim 4, characterized in that Constructing a knowledge graph corresponding to the target object based on the triple data includes: Based on the triple data, construct a statement to be reviewed; Based on the data of each word included in the sentence to be reviewed, the sentence to be reviewed is subjected to semantic evaluation processing to obtain a data evaluation result; wherein the data evaluation result is used to indicate the credibility of the triple data; If it is determined that the credibility of the triple data meets the preset threshold requirement, a knowledge graph is constructed based on the triple data.

6. The method according to claim 5, characterized in that The method further comprises: If it is determined that the credibility of the triple data does not meet the preset threshold requirement, the triple data is sent to a reviewer for manual review.

7. The method according to any one of claims 1 to 6, characterized in that Before acquiring the preset database of the target object, the method further includes: Acquire document data associated with the target object; Slicing the document data according to a preset slicing dimension to obtain the plurality of document slices; wherein the preset slicing dimension indicates a directory dimension and / or a semantic dimension; The preset database is constructed based on the multiple document slices.

8. An online reasoning device for building a knowledge graph based on a large model, characterized in that: include: An acquisition unit, used for acquiring a preset database of a target object; wherein the preset database includes a plurality of document slices; A first generating unit, configured to determine at least one question information corresponding to each of the document slices, and determine at least one target slice from the multiple document slices that meets similar requirements as each of the question information; A second generating unit, configured to generate data to be identified based on the problem information and the at least one target slice; wherein the data to be identified indicates summary data; A construction unit is used to construct a knowledge graph corresponding to the target object based on the data to be identified.

9. A computer device, characterized in that: include: Memory, processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory, so that the processor performs the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 7 when executed by a processor.

Images