Text auditing method and device, equipment and storage medium

Abstract

The embodiment of the invention discloses a text auditing method and device, equipment and a storage medium, and relates to the technical field of natural language processing. The method comprises the steps of obtaining an auditing target and a to-be-audited document, wherein the auditing target comprises a rule for auditing the to-be-audited document; obtaining a global knowledge graph corresponding to the to-be-audited document according to an association relationship among a plurality of target objects included in the to-be-audited document; and converting the auditing target into a query statement, performing retrieval in the global knowledge graph based on the obtained query statement, and obtaining and outputting an auditing report. According to the method, lossless structured representation can be realized, and active logic verification is supported, so that the efficiency and reliability of super-long and complex text auditing are improved.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of text processing, and relate to but are not limited to a text auditing method, device, equipment and storage medium. BACKGROUND

[0002] Existing text auditing tools are essentially passive information retrievers, and their core functions are limited to locating the positions of relevant content such as page numbers and paragraphs in documents according to keywords or query conditions input by users, and can only complete the task of "finding content".

[0003] Therefore, existing text auditing tools cannot understand the internal association relationships of target objects in documents, and cannot actively check content according to auditing rules. Users need to manually filter and integrate search results and make artificial judgments, and the whole process relies on artificial guidance, resulting in low efficiency and difficulty in realizing truly automated auditing. SUMMARY

[0004] Therefore, the text auditing method, device, equipment and storage medium provided by the embodiments of the present application can improve the efficiency and reliability of text auditing. The text auditing method, device, equipment and storage medium provided by the embodiments of the present application are implemented as follows: The first aspect of the embodiments of the present application discloses a text auditing method, which comprises: obtaining an auditing target and a document to be audited, wherein the auditing target comprises a rule for auditing the document to be audited; obtaining a global knowledge graph corresponding to the document to be audited according to association relationships between a plurality of target objects included in the document to be audited; converting the auditing target into a query statement, and searching in the global knowledge graph based on the obtained query statement to obtain and output an auditing report.

[0005] As an optional implementation, in the first aspect of the embodiments of the present application, the obtaining of the global knowledge graph corresponding to the document to be audited according to the association relationships between the plurality of target objects included in the document to be audited comprises: segmenting the document to be audited into a plurality of target text blocks; extracting association relationships between a plurality of target objects included in each target text block of the plurality of target text blocks according to a preset large language model and a preset ontology to obtain a plurality of local knowledge graphs corresponding to the plurality of target text blocks, wherein the preset ontology comprises types of target objects that need to be focused on and types of association relationships between any two target objects that need to be focused on in a field to which the document to be audited belongs, and the preset ontology is obtained by learning sample texts of the field to which the document to be audited belongs through a target model; merge the plurality of local knowledge graphs to obtain the global knowledge graph.

[0006] As an optional implementation, in the first aspect of the embodiment, the merging the plurality of local knowledge graphs to obtain the global knowledge graph comprises: unifying the name of each target object included in each of the plurality of local knowledge graphs, and determining a unique identifier of each target object in the global knowledge graph to obtain a plurality of processed local knowledge graphs; merging the plurality of processed local knowledge graphs according to the unique identifier of each target object in the plurality of processed local knowledge graphs to obtain the global knowledge graph.

[0007] As an optional implementation, in the first aspect of the embodiment, the method further comprises: storing the global knowledge graph in a graph database; The converting the audit target into a query statement comprises: translating the audit target into a query statement executable by the graph database.

[0008] As an optional implementation, in the first aspect of the embodiment, the retrieving in the global knowledge graph based on the obtained query statement to obtain and output an audit report comprises: retrieving in the graph database based on the obtained query statement to obtain a target retrieval result that does not satisfy the query statement; obtaining and outputting an audit report according to the target retrieval result that does not satisfy the query statement, the target retrieval result comprising a query object that does not satisfy the query statement and a target parameter corresponding to the query object, and the plurality of target objects comprising the query object.

[0009] As an optional implementation, in the first aspect of the embodiment, the obtaining and outputting an audit report according to the target retrieval result that does not satisfy the query statement comprises: counting and obtaining the number of all query objects included in the target retrieval result; obtaining the output order of each query object and the corresponding target parameter according to the position of each query object in the document to be audited, and generating a link identifier corresponding to each query object, the link identifier being used to jump to the position of the corresponding query object in the document to be audited; obtaining the reason for not satisfying the query statement according to the target parameter corresponding to each query object; According to the number of all the query objects, the output order of each query object and a corresponding target parameter, a link identifier corresponding to each query object, and the reason, the audit report is generated and output.

[0010] As an optional implementation, in the first aspect of the embodiment, the audit target is acquired, including: a rule interface is displayed, and the rule interface is used for inputting a rule described in a declarative programming language; The audit target is acquired in response to an input operation of a user on the rule interface.

[0011] The second aspect of the embodiment discloses a text audit device, and the system includes an audit data input module, a knowledge graph construction module, and an audit report output module. The audit data input module is configured to acquire an audit target and a document to be audited, and the audit target includes a rule for auditing the document to be audited. The knowledge graph construction module is configured to obtain a global knowledge graph corresponding to the document to be audited according to an association relationship between a plurality of target objects included in the document to be audited. The audit report output module is configured to convert the audit target into a query statement, and retrieve in the global knowledge graph based on the obtained query statement to obtain and output an audit report.

[0012] The third aspect of the embodiment discloses a computer device, including a memory and a processor, the memory stores a computer program that can run on the processor, and the processor implements the method of the embodiment when executing the program.

[0013] The fourth aspect of the embodiment discloses a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method provided by the embodiment.

[0014] Compared with the related art, the embodiment has at least the following beneficial effects: The text audit method, device, equipment, and storage medium provided by the embodiment can acquire an audit target and a document to be audited, the audit target includes a rule for auditing the document to be audited, obtain a global knowledge graph corresponding to the document to be audited according to an association relationship between a plurality of target objects included in the document to be audited, convert the audit target into a query statement, retrieve in the global knowledge graph based on the obtained query statement, and obtain and output an audit report. Therefore, the embodiment only needs to input the document to be audited and the audit target, and can automatically complete the audit whole process, and finally generate the audit report, get rid of the inefficiency and error of traditional manual audit, and greatly improve the audit efficiency and reliability. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description only constitute some embodiments of the present application. For those skilled in the field, other drawings can also be obtained from these drawings without any creative effort.

[0016] Figure 1 A flowchart of a text auditing method provided in the embodiments of the present application is shown in FIG. 1. Figure 2 A flowchart of constructing a global knowledge graph provided in the embodiments of the present application is shown in FIG. 2. Figure 3 A flowchart of generating an auditing report provided in the embodiments of the present application is shown in FIG. 3. Figure 4 A whole architecture diagram of a text auditing device provided in the embodiments of the present application is shown in FIG. 4. Figure 5 A structural block diagram of a text auditing device provided in the embodiments of the present application is shown in FIG. 5. Figure 6 A structural diagram of a computer device provided in the embodiments of the present application is shown in FIG. 6. DETAILED DESCRIPTION

[0017] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the specific technical solutions of the present application will be further described in detail below with reference to the drawings in the embodiments of the present application. The following embodiments are used to illustrate the present application, but not to limit the scope of the present application.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application, and are not intended to limit the present application.

[0019] In the following description, “some embodiments” are described, which describe a subset of all possible embodiments, but it can be understood that “some embodiments” can be the same subset or different subset of all possible embodiments, and can be combined with each other without conflict.

[0020] It should be noted that the terms “first\second\third” used in the embodiments of the present application are used to distinguish similar or different objects, and do not represent a specific order of the objects. Understandably, “first\second\third” can be interchanged with a specific order or sequence as allowed, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0021] Large language models refer to deep learning models trained using a large amount of text data, which can generate natural language text or understand the meaning of language text. Natural language processing is a branch of artificial intelligence and linguistics, which can realize various theories and methods of effective communication between people and computers using natural language. Therefore, the research in this field will involve natural language, that is, the language used in people's daily life, so it is closely related to the research of linguistics, but there are important differences. Natural language processing is not generally studying natural language, but developing computer systems, especially software systems, that can effectively realize natural language communication, so it is part of computer science.

[0022] With the continuous development of natural language processing, people can gradually perform information retrieval on the basis of natural language processing, thereby realizing the verification and audit of various text information. Traditional text audit relies on manual word-by-word reading, which is not only inefficient, but also easy to miss due to scattered information and complex logic. Natural language processing technology solves the problem of "accurate information extraction", which converts unstructured or complex structured text into searchable "tagged information" through semantic analysis, entity recognition, relationship extraction and other capabilities.

[0023] However, existing text audit tools are essentially passive information retrievers, and their core functions are limited to locating the positions of relevant content, such as page numbers and paragraphs, in documents based on user input keywords or query conditions, and can only complete the task of "finding content".

[0024] Therefore, the existing text audit tool cannot understand the internal relationship of the target objects in the document, and cannot actively check the content according to the audit rules. Users need to manually filter and integrate the search results and make manual judgments, relying on manual guidance throughout the process, resulting in low efficiency and difficulty in realizing truly automated audit. Therefore, how to improve the efficiency and reliability of long and complex text audit is a problem to be solved.

[0025] Therefore, the embodiments of the present application provide a text audit method and system, which can obtain an audit target and a document to be audited, the audit target including a rule for auditing the document to be audited; obtain a global knowledge graph corresponding to the document to be audited according to the association relationship between a plurality of target objects included in the document to be audited; convert the audit target into a query statement, and retrieve in the global knowledge graph based on the obtained query statement to obtain and output an audit report. Therefore, the embodiments of the present application only need to input the document to be audited and the audit target, and can automatically complete the entire audit process, and finally generate a clear and understandable audit report, getting rid of the inefficiency and errors of traditional manual audit, and greatly improving the audit efficiency and reliability.

[0026] In order to make the purpose, technical scheme of the present application more clear and intuitive, the text auditing method and system provided by the embodiments of the present application will be described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0027] Please refer to Figure 1 , the flowchart of the text auditing method provided in the embodiments of the present application is shown as Figure 1 , which can include the following S110 to S130: S110: obtaining an auditing target and a document to be audited, the auditing target including rules for auditing the document to be audited.

[0028] The execution subject of the method of the present embodiment is an electronic device, and it should be noted that the above-mentioned electronic device can include but is not limited to mobile phones, wearable devices (such as smart watches, smart bracelets, smart glasses, etc.), tablet computers, notebook computers, vehicle-mounted terminals, PC (Personal Computer), etc. The functions realized by the method can be realized by calling program codes in the processor of the electronic device, and of course the program codes can be saved in the computer storage medium, so the electronic device at least includes a processor and a storage medium.

[0029] In one embodiment, when auditing a text, the rules of the document to be audited and the auditing target need to be clarified first. The document to be audited usually refers to the text material that needs to be checked for compliance, logic or integrity, such as a construction engineering construction plan, which contains construction steps, material specifications, safety standards, etc.; or a business contract, which covers the rights and obligations of both parties, payment methods, breach of contract responsibilities, etc. The auditing target is the specific requirement that the user expects to achieve through auditing, and its core is a series of clear rules, which can be multi-dimensional, such as in the auditing of a construction engineering construction plan, the rules may include "all concrete strength grades must meet the C30 standard marked on the design drawing" and "high-altitude operation steps need to be accompanied by detailed safety protection measures description"; in the auditing of a business contract, the rules may be "the payment period shall not exceed 30 days" and "the breach of contract responsibility clause needs to clearly specify the compensation calculation method of both parties". Through the specific and clear scope of the document to be audited and the specific rules contained in the auditing target, clear guidance is provided for subsequent construction of knowledge graph and execution of retrieval.

[0030] In an embodiment, the way to obtain the document to be audited can be flexibly adapted according to the actual scene to meet diversified auditing needs. The user can actively paste the document content in the text box of the electronic device interface, or complete the submission by dragging the local document to the upload area. This method is suitable for the auditing of temporary and small amount of documents. If the document is pre-stored in the terminal device, the system can read the document under the specified path through the file browser interface without repeated uploading. When the target document is not stored locally on the device, the system can automatically download the document according to the user input document ID or path through the preset API interface connection enterprise server or cloud storage platform, and also supports batch synchronization of multiple associated documents. In the auditing scene embedded in the business process, the system can be real-time linked with the document management system and collaboration tools. For example, when the user triggers the "audit" button on the project management platform, the system will automatically pull the corresponding version of the document to be audited from the associated document library. In addition, by scanning the two-dimensional code corresponding to the document or inputting the online link of the document, the system can also directly obtain the document content after parsing, which is convenient for cross-device and cross-platform document sharing and auditing. These methods cover multiple dimensions of the scene and provide stable data source support for subsequent automated auditing processes.

[0031] In an embodiment, the way to obtain the auditing target can be flexibly adapted to different user habits and scene needs to ensure efficient input of rule information. The user can directly input the auditing rule through the text box, for example, "check whether the device parameters meet the industry standard GB / T 1234 - 2023". This method is suitable for scenes where the rule is clear and the expression is rigorous. If the user is not convenient to type, the voice input function can be used to speak the rule, such as "check whether the validity period of all payment clauses in the contract is consistent". The system will automatically convert the voice into text and parse it into a structured auditing target. For commonly used standardized auditing rules, the system also supports direct selection from the preset rule library. The user only needs to select the corresponding rule template in the drop-down menu, such as "engineering drawing size marking specification verification", and adjust the parameters according to the actual needs. In addition, for complex multi-dimensional auditing targets, a rule document such as an Excel table or a Word document containing multiple check conditions can be uploaded. The system will automatically extract the rule information and integrate it into a unified auditing target. These diversified acquisition methods not only reduce the user input threshold, but also accurately capture the auditing needs, laying a foundation for subsequent rule conversion and graph retrieval.

[0032] In an embodiment, the document to be audited can be a super-long document of tens of millions of words, a complex logical structure, or an unstructured document. Among them, the super-long document of tens of millions of words contains multiple chapters, multiple target objects, and multiple association relationships between the multiple target objects; the super-long document with complex logical structure has an explicit but tedious logical framework; the super-long unstructured document has no fixed format or logical framework, and the information is scattered. In practical applications, different types of documents can have at least one of the above characteristics, which are not limited here.

[0033] In an embodiment, the format of the document to be audited can include PDF, WORD, TXT, etc.; the field related to the document to be audited can include engineering technology field, financial field, legal field, medical field, etc.; the content of the document to be audited can include engineering design manual, legal contract, pharmaceutical process manual, etc.; which are not limited here, and can be set according to actual needs.

[0034] For example, the document to be audited can be an engineering design manual document related to the engineering technology field. The document is a PDF format super-long document of tens of millions of words with complex logical structure, totaling about 120 million words, containing 300 chapters, tens of thousands of specification parameters and operation procedures. Among them, the audit target can be "verify whether all component information, operation procedures and safety constraint content related to 'a certain coolant system' in the document are completely consistent within the entire document, and whether they meet the relevant provisions of safety regulations".

[0035] S120: Obtain a global knowledge graph corresponding to the document to be audited according to the association relationships between the multiple target objects included in the document to be audited.

[0036] It should be noted that the target object refers to a core concept carrier with practical significance in a specific field, which is a basic element for building a knowledge graph, equivalent to an "anchor point" for extracting key information from a document, which can help the system quickly locate core content and avoid missing or misjudgment.

[0037] And there is an association relationship between the target objects, because in the actual document content, these core concepts are not isolated, and there are various logical relationships between them. Only by associating them can a logical whole be formed, and the internal logic of the document information can be clearly sorted out to provide accurate guidance for subsequent information processing.

[0038] In an embodiment, the target objects in different fields are different, for example: in the field of engineering technology, there may be components, systems, procedures, specifications, parameters, positions, regulatory provisions, etc.; in the legal field, there may be parties, clauses, rights, obligations, breach of contract responsibilities, time limits, etc.; in the financial field, there may be accounts, transactions, interest rates, risk levels, financial products, regulatory policies, etc.; in the medical field, there may be diseases, drugs, treatment plans, examination items, patient information, medical standards, etc. It should be noted that due to the differences in characteristics and concerns of different fields, target objects are not universally present in all fields. Objects specific to a field, such as "statute of limitations" in the legal field, may not exist at all in other fields, such as the field of engineering technology. Therefore, the target objects in each field are not uniformly limited.

[0039] Corresponding to the target objects, the association relationships between the target objects in different fields also differ due to field characteristics. For example, in the engineering field, "system" and "component" may have a "containment" relationship, while in the legal field, "clause" and "right" may have a "definition" relationship. These relationships are also field-specific and are not uniformly regulated.

[0040] In an embodiment, the content containing specific field target objects and association relationships has flexibility and adaptability in acquisition methods: on the one hand, it can be pre-set by technical personnel in the field combined with industry experience and professional knowledge, for example, in the field of aerospace engineering, engineers can define core target objects such as "propulsion system", "fuel reserve", "orbit parameters" based on spacecraft design specifications, and define association relationships such as "propulsion system depends on fuel reserve" and "orbit parameters need to match task requirements". These contents are organized into structured documents, such as ontology definition files in JSON format, and stored in the system to ensure the professionalism and accuracy of the target objects and relationships. On the other hand, for fields that already have a large number of historical documents, such as financial contract review and medical case management, trained AI models, such as entity recognition models and relationship extraction models based on BERT, can be used to process historical documents in batches, automatically identify high-frequency core concepts such as "collateral" and "loan interest rate" in the financial field and "postoperative complications" and "drug contraindications" in the medical field, and mine the regular associations between these concepts, such as "the value of the collateral must cover the loan amount" and "postoperative complications must match the corresponding emergency plan". These automatically extracted objects and relationships are then integrated to generate the pre-set documents to be reviewed.

[0041] S130: converting the review target into a query statement, and retrieving in the global knowledge graph based on the obtained query statement to obtain and output a review report.

[0042] Transforming audit objectives into query statements essentially addresses the compatibility issue between "user requirement expression" and "knowledge graph retrieval logic." Audit objectives are typically user verification needs described based on domain rules or natural language, focusing on "what logical relationships need to be verified." Their expression is closer to human cognitive habits, including core elements such as entity types, association conditions, and verification rules. However, they do not follow the retrieval syntax of graph databases and cannot be directly parsed and executed by the knowledge graph. Query statements are "retrieval instructions" that graph databases can understand and execute. They must be written strictly according to the graph's storage structure (such as entity tags, relation types, and attribute definitions) and syntax rules (such as the format of entity matching, relation traversal, and conditional filtering) to ensure accurate location of target entities, extraction of relevant parameters, and execution of logical verification. Therefore, this conversion process essentially translates abstract, human-understandable audit requirements into structured instructions that conform to the retrieval logic of the knowledge graph, enabling the system to perform automated retrieval and verification within the global knowledge graph based on the core logic of the audit target.

[0043] As an example, an engineering expert needs to verify a key safety margin rule: "Verify that all pressurized piping belonging to the 'primary coolant system' must be designed to a pressure at least 15% higher than the maximum output pressure of any pump directly connected to it." Based on this, the expert enters the rule into the interface: "FOR EACH p:Component WHERE p.type == 'Pipe' AND p.system == 'PrimaryCoolantSystem', FOR EACH pump:Component WHERE pump.type == 'Pump' AND pump CONNECTS_TO p, VERIFY THAT p.designPressure>= pump.maxOutputPressure * 1.15”. Furthermore, the system translates the above audit objectives into query statements supported by the graph database as follows: "MATCH(pump:Component{type:'Pump'})-[:CONNECTS_TO]->(pipe:Component{type: 'Pipe', system: 'PrimaryCoolantSystem'}); / / Obtain the pressure specifications for the pump and pipeline respectively; MATCH (pump)-[:HAS_SPEC]->(pump_spec:Specification {type: 'Maximum Output Pressure'}); MATCH (pipe)-[:HAS_SPEC]->(pipe_spec:Specification {type: 'Design stress'}); / / Convert the pressure value to a floating-point number for calculation; WITH pump, pipe, toFloat(pump_spec.value) AS pumpPressure, toFloat(pipe_spec.value) AS pipePressure; / / Find all combinations that do not meet the safety margin (less than 1.15 times); WHERE pipePressure <pumpPressure * 1.15; / / Returns the offending component and its location in the source documentation for further investigation; RETURN pump.name AS pumpName, pump.source_location AS pumpLocation, pipe.name AS pipeName, pipe.source_location AS pipeLocation, pumpPressure, pipePressure”. The above is just an example. Those skilled in the art can set different audit objectives according to actual use. Accordingly, if the format or standard used in the created global knowledge graph is different, the converted query statements may not be exactly the same. Examples will not be given here.

[0044] After obtaining the above query statement, the global knowledge graph can be searched according to the query statement. The specific search method can be similar to the search method in related technologies, and is not limited in the embodiments of this application.

[0045] As an example, the retrieval result can include all query objects and their associated information that do not satisfy the query statement: first, the total number of query objects involved is determined (such as 2), and the specific location of each object in the document to be audited (such as chapter, page number, table number) is recorded; at the same time, the target parameters of each object (such as the pressure value of the device, specification data, etc.) and the specific reasons for not satisfying the rules calculated by the parameters (such as the value does not meet the proportion requirement) are associated; in addition, a special link identifier is generated for each object to directly jump to the original document location. These information are associated with each other, which not only clearly presents the number, parameters and reasons of the violation objects, but also ensures the traceability of the results through location information and links.

[0046] After the retrieval result is retrieved in the global knowledge graph, a retrieval report can be generated according to the obtained retrieval result. For example, all the obtained retrieval results are combined to obtain a retrieval report, so that the user can obtain a more comprehensive retrieval situation. Or, from multiple retrieval reports, the retrieval results that do not meet the audit target are selected, and the retrieval results that do not meet the audit template are combined to obtain a retrieval report, so that the user can pay more attention to the content in the document to be audited that does not meet the audit target. Of course, other ways of generating a retrieval report can also be used, which are not exemplified here.

[0047] Therefore, the embodiments of the present application only need to input the document to be audited and the audit target, and can automatically complete the audit whole process, and finally generate an audit report, get rid of the inefficiency and error of traditional manual audit, and greatly improve the audit efficiency and reliability.

[0048] Figure 1 The method relies on the complete mapping and accurate association of the global knowledge graph to document information, and creating a complete and accurate global knowledge graph is a prerequisite for efficient and automated auditing: by analyzing the document to be audited, all core target objects are extracted, and the association between objects is constructed according to the relationship type defined by the domain ontology, and finally this structured knowledge graph is stored in a graph database. This global knowledge graph stored in the database not only provides a unified and traceable data source for queries, but also ensures that cross-object and cross-chapter association retrieval can accurately reflect the real logic of the document, laying a foundation for efficient execution of subsequent query statements and accuracy of the results.

[0049] Please refer to Figure 2 The flowchart for constructing the global knowledge graph provided in the embodiments of the present application is shown in FIG. 1. Figure 2 As shown in FIG. 1, the method can include the following S210 to S250: S210: Split the document to be audited into multiple target text blocks.

[0050] In an embodiment, the segmentation rules can include structural feature segmentation rules, content theme segmentation rules, length adaptation segmentation rules, etc., which are not specifically limited here and can be segmented according to actual needs.

[0051] The structural feature segmentation rule is segmented according to the inherent format structure of the document, for example, taking chapter titles, paragraph markers, hierarchical titles (such as "1.1 section" and "1.1.1 clause") as segmentation nodes to segment the document into text blocks corresponding to chapters and paragraphs.

[0052] The content theme segmentation rule is segmented according to the theme relevance of the text content, for example, by identifying the range of occurrence of core concepts (such as specific terms and object names) in the document, continuous text that develops around the same core concept is divided into independent text blocks.

[0053] The length adaptation segmentation rule is for long text, and a fixed character number or sentence number threshold is set as the segmentation basis. When the length of a single text exceeds the threshold, it is split according to the semantic complete sentence boundary to ensure that the length of each target text block adapts to the processing needs of subsequent entity recognition and relationship extraction.

[0054] As an example, the aforementioned 120 million word engineering design manual document can be segmented into thousands of text blocks according to chapters.

[0055] S220: Extract the association relationships between the multiple target objects included in each target text block in the multiple target text blocks according to the preset large language model and the preset ontology, obtain multiple local knowledge graphs corresponding to the multiple target text blocks, and the preset ontology includes the types of target objects that need to be concerned in the field to which the document to be audited belongs and the types of association relationships between any two target objects that need to be concerned. The preset ontology is obtained by learning sample texts in the field to which the document to be audited belongs through a target model.

[0056] Among them, the association relationships between the multiple target objects in the text block are constructed into a local knowledge graph by using a large language model, which is a further information structuring process based on text block segmentation.

[0057] Specifically, after the document to be audited is split into target text blocks according to logical boundaries, the LLM will first perform in-depth semantic analysis on a single target text block, and accurately identify the target objects in the text block according to the ontology defined in the field.

[0058] As an example, the text block of "reactor coolant system" in Chapter 5 obtained by segmentation in the engineering design manual, (1) the target object 1 of the audit is a component, the component name is the first primary coolant pump, and the model is RCP-01; (2) the target object 2 of the audit is a specification parameter, which indicates that the maximum allowable working pressure value is 17.5 MPa; (3) the relationship between the two target objects of the audit is "has". That is, the local knowledge graph A is that the first primary coolant pump of the model RCP-01 has a specification parameter, and the maximum working pressure value allowed by the specification parameter is 17.5 MPa.

[0059] The text block of "start-up and shutdown procedures" in Chapter 12 obtained by segmentation in the engineering design manual, (1) the target object 3 of the audit is a component, the component model is RCP-01; (2) the target object 4 of the audit is an operation procedure, the operation procedure number is P-22, and the name is primary loop preheating; (3) the relationship between the two target objects of the audit is "reference". That is, the local knowledge graph B is that a primary loop preheating operation procedure with a number of P-22 references a component with a model of RCP-01.

[0060] It should be noted that the preset ontology is obtained by learning sample texts belonging to the field of the document to be audited by the target model. The target model is specially designed for feature extraction of texts in a certain field, and generates a preset ontology template. Specifically, the target model will first learn various texts in the field (such as internal process specifications, project documents, business records, or industry-specific technical manuals, compliance files, etc.) systematically, accurately capture high-frequency entity types in the text (such as "department name", "product number", "operation steps", etc.), and the specific relationship between entities (such as "XX task is executed by XX department", "XX parameter is associated with XX standard", etc.), and based on these learning results, a structured framework that meets the characteristics of the field knowledge, i.e. the preset ontology, is constructed. The generated preset ontology template is not a fixed and unchangeable general format, but will be adjusted according to the specific type of the text (such as contract, report, log, etc.) and the content characteristics, for example, for technical documents, the extraction of "parameter-performance" type associations is strengthened, and for business documents, the capture logic of "responsibility-process" type relationships is highlighted. These preset ontology templates will serve as accurate guides for subsequent large model extraction of knowledge graph entities, effectively solving the problems of entity recognition deviation and missing association relationships that may occur when general large models lack domain-specific training when processing internal or vertical field documents, thereby laying a foundation for the construction of high-quality global knowledge graphs.

[0061] S230: Unify the name of each target object included in each knowledge graph of the plurality of local knowledge graphs, and determine a unique identifier of each target object in the global knowledge graph, to obtain the plurality of processed local knowledge graphs.

[0062] As an example, in the process of merging the above local knowledge graph A and local knowledge graph B into a global knowledge graph, the system can accurately determine that the "RCP-01" of target object 1 and target object 3 in the two sub-graphs points to the same physical object, i.e. No. 1 primary coolant pump. Based on this, the system will assign the two nodes related to "RCP-01" in the two sub-graphs as a unique node identifier, which can be set as urn:component:rcp-01, to obtain the plurality of processed local knowledge graphs.

[0063] It should be noted that determining that target objects with different names point to the same physical object is mainly achieved through multi-dimensional feature matching and rule checking: first, the system will extract the core attributes of the target objects for comparison, including but not limited to functional description (such as "primary coolant circulating pump"), system to which it belongs (such as "primary coolant system"), technical parameters (such as rated power, design pressure), association relationship (such as connected pipe number, control module model), etc. If the core attributes of the two objects are highly consistent (such as function, system, and key parameters are completely consistent), it is preliminarily determined that there is a possibility of pointing to the same physical object. Secondly, the synonym library or alias rule defined in the domain ontology is checked. For example, in the engineering field, "RCP-01" and "No. 1 primary coolant pump" may be explicitly marked as different names of the same device, and the system will associate objects with different names but equivalent semantics according to such preset rules. In addition, for the case of fuzzy matching (such as large difference in name but partial attribute overlap), the system will trigger an artificial confirmation mechanism to determine whether it is the same object by a domain expert, to ensure the accuracy of the unique identifier. Through this combination strategy, the system can accurately identify target objects with different names but pointing to the same physical object, laying a foundation for the unique identification of the global knowledge graph.

[0064] S240: Merge the plurality of processed local knowledge graphs according to the unique identifier of each target object in the plurality of processed local knowledge graphs, to obtain the global knowledge graph.

[0065] The system first fuses nodes with the same unique identifier in different local knowledge graphs into a single node in the global graph to avoid duplicate records. Then, all relationships associated with these nodes in each local graph, whether they are attribute parameters (such as pressure values), reference sources (such as procedure numbers), or connection objects (such as pipeline information), are uniformly aggregated to the fused node to form a complete association network. If the information of the same node in different local graphs conflicts, the system will coordinate according to pre-set rules (such as giving priority to the latest document data) to ultimately build a unified knowledge graph that contains both local details and global associations, ensuring that the description of each target object is comprehensive and consistent.

[0066] As an example, specifically, after completing the unification of target object names and determining the unique identifier, the two local knowledge graphs are merged according to the unique identifier. In the merged global knowledge graph, it contains the association relationship that the No. 1 primary coolant pump (model: RCP-01) has a maximum allowable working pressure value of 17.5 MPa, and it also contains the association relationship that the No. 1 primary coolant pump with model RCP-01 is referenced by the No. 1 loop preheating operation procedure with number P-22.

[0067] S250: store the global knowledge graph in a graph database; convert the audit target into a query statement, including: translating the audit target into a query statement that the graph database can execute.

[0068] In one embodiment, Neo4j is a high-performance graph database that uses a proprietary query language called Cypher. Neo4j is a native graph database that centers on "target objects and relationships", and its design philosophy is completely different from the table mode of traditional relational databases. It uses "nodes" to represent target objects in reality, and nodes can be classified and carry attributes through "labels"; it uses "relationships" to connect different nodes and clearly defines the association between target objects, and relationships have a clear direction and can also have attributes. This "node-relationship-attribute" model naturally fits the network association of the real world, and does not need to split data like relational databases with foreign keys, and can completely retain the semantics and structure of the association, so the graph database can execute the translated query statement as described in S130.

[0069] Based on Figure 2The method, a user needs to obtain a to-be-audited document through a document uploading entrance provided by the system, and directly inputs an audit target in a rule input interface, and the system can automatically complete full-process processing from knowledge graph construction, rule conversion to retrieval analysis, and finally output an audit report in real time. Such an extremely simple interaction mode greatly reduces the operation threshold of the user, and even non-technical personnel can quickly get started, completely getting rid of the cumbersome process of manually combing rules and repeatedly adjusting query conditions in traditional auditing, enabling the user to focus on the definition of the audit target and the application of the result, and significantly improving operation efficiency and user experience.

[0070] Please refer to Figure 3 The flowchart for generating an audit report provided in the embodiments of the present application is as shown in Figure 3 The method can include the following S310 to S370: S310: Based on the obtained query statement, retrieval is performed in the graph database to obtain a retrieval result that does not satisfy the query statement.

[0071] The retrieval of the query statement in the graph database is a precise traversal and screening process based on entities, relationships and constraint rules. First, the system will locate the target node in the knowledge graph according to the entity types (such as "pipeline" and "pump") and attribute conditions (such as "belongs to a loop coolant system") defined in the statement, and match the entity pairs that exist in association through the specified relationship type (such as "CONNECTS_TO"), to ensure that only objects that meet the logical association are included. Then, along the preset attribute association relationship (such as "HAS_SPEC"), the specific parameter value of the entity is extracted, and if the parameter is in text form, it is automatically converted into a calculable numerical value (such as converting "17.2 megapascals" to 17.2).

[0072] Subsequently, according to the constraint rules (such as "pipeline design pressure < pump maximum output pressure x 1.15") in the statement, numerical operation and comparison are performed, and rule violation items that do not satisfy the conditions are screened out - for example, when the maximum output pressure of the pump is 17.2 MPa, its 1.15 times is 19.78 MPa, and if the pipeline design pressure is 19.5 MPa, the system will determine that the combination does not satisfy the rule and mark it as a violation. Finally, the names of the rule violation entities, the parameter values, and the specific reasons for not satisfying the rules are integrated, and are associated to the specific location (such as "Chapter 5, Section 3.2") of the original document through the "source_location" attribute stored in the knowledge graph, to form a retrieval result that can be directly traced. This process relies on the efficient processing capability of the graph database for entity relationships, realizes the full-process automation from entity matching to result output, and ensures accurate and traceable retrieval.

[0073] S320: obtaining and outputting an audit report according to the target retrieval result that does not satisfy the query statement, the target retrieval result including a query object that does not satisfy the query statement and a target parameter corresponding to the query object, and the plurality of target objects including the query object.

[0074] The "target retrieval result that does not satisfy the query statement" retrieved by the graph database is converted into an intuitive and usable audit report. Specifically, this step first determines that the core information source to be processed is the "target retrieval result", which includes two key contents: one is the "query object that does not satisfy the query statement", i.e. those entities that are judged as violations in the audit, such as the No. 2 primary coolant pump and the No. 2 loop coolant pipe that do not meet the pressure safety margin requirement in the example; the other is the "target parameter corresponding to the query object", i.e. the specific data related to these entities, such as the maximum output pressure of the pump 17.2 MPa and the design pressure of the pipe 19.5 MPa. At the same time, it is emphasized here that "the plurality of target objects includes the query object", in order to make it clear that these violating entities do not exist in isolation, but are selected from the global target objects of the document to be audited, ensuring the relevance of the retrieval result to the original document.

[0075] S330: counting and obtaining the number of all query objects included in the target retrieval result.

[0076] This step mainly counts the number of query objects in the target retrieval result, and through the analysis of all entities that do not satisfy the query statement, it determines the total number of objects involved in the violation and the specific number of inconsistencies. For example, in the example, 1 place is found to be inconsistent with the rules, involving 2 query objects (pump and pipe), so the statistical result is "1 inconsistency, involving 2 query objects", which can intuitively reflect the scale of the audit problem and let the user quickly understand the scope of the violation.

[0077] S340: obtaining the output order of each query object and the corresponding target parameter according to the position of each query object in the document to be audited, and generating a link identifier corresponding to each query object, the link identifier being used to jump to the position of the corresponding query object in the document to be audited.

[0078] This step mainly solves the problem of the presentation order and traceability of information in the audit report. First, according to the original position of each query object in the document to be audited (such as chapter, page number, table number), the output order is determined to ensure that the report content is consistent with the position logic of the original document. In the example, the pump is located in the 5th chapter and the pipe is located in the 8th chapter, so the information of the pump is output first and then the information of the pipe. Second, a special link identifier is generated for each query object. The identifier encodes the specific position in the document, such as "link: / / chap5 / sec3 / p281 / table5-3". Users can click on it to directly jump to the original position, greatly improving the convenience of problem checking.

[0079] S350: According to the target parameter corresponding to each query object, the reason for not meeting the query statement is obtained.

[0080] The core of this step is to analyze the specific reason for not meeting the query statement, which needs to be calculated and logically deduced based on the target parameter of each query object. For example, in the example, by comparing the 1.15 times of the maximum output pressure of the pump, which is 19.78 MPa, with the design pressure of the pipe, which is 19.5 MPa, the conclusion "19.5 MPa < 19.78 MPa" is obtained, and the reason is "the design pressure of the pipe does not meet the 115% safety margin of the maximum output pressure of the connected pump". This process converts parameter data into understandable violation reasons, allowing users to clearly understand the nature of the problem.

[0081] S360: According to the number of all query objects, the output order of each query object and the corresponding target parameter, the link identifier corresponding to each query object, and the reason, the audit report is generated and output.

[0082] As the final link of the audit process, this step is responsible for integrating all key information from the previous steps to generate and output a complete audit report. The integrated information includes: the number of query objects (such as 1 inconsistency, involving 2 components), the query objects and corresponding parameters sorted by position (such as pump 17.2 MPa, pipe 19.5 MPa), the link identifier of each object, and the specific violation reason. Through structured presentation (such as "audit task completed: found 1 inconsistency" + detailed content), a complete, logical, and directly traceable report is formed, providing clear evidence for subsequent rectification and verification by users.

[0083] S370: Display a rule interface for inputting rules described in a declarative programming language; in response to user input operations on the rule interface, obtain the audit target.

[0084] As an example, the system first displays a rule interface, which provides an input box and an example prompt, such as "Please enter the verification rule, format reference: [pipe design pressure] ≥ [maximum output pressure of the connected pump] x 1.15", to support the user to input the audit rule in a declarative programming language (such as a rule description language).

[0085] When the user needs to audit the "pressure matching of the pipe and the pump", the rule interface is input as "the pipe design pressure needs to reach 115% safety margin of the maximum output pressure of the connected pump", which corresponds to the logical relationship described in the declarative language. The system responds to the input operation, and determines the verification requirement corresponding to the rule as the audit target, that is, "check whether all pipe design pressures meet 115% safety margin of the maximum output pressure of the connected pump".

[0086] The embodiment of the present application is simple and easy to understand according to the interactive interface, and non-technical personnel can easily input the audit requirement; the structured retrieval of the graph database can accurately locate the violation item, and the link identifier is directly associated with the original text position to ensure that each problem can be traced back; the generated report information is clear, which helps the user to quickly understand the problem and rectify it. The whole process automation greatly improves the efficiency, adapts to the documents containing complex associations, and provides a reliable solution for cross-domain large-scale auditing.

[0087] Next, for the technical solutions in the embodiments of the present application, a framework diagram of a text auditing system is provided. Figure 4 For the overall architecture diagram of the text auditing system provided in the embodiments of the present application, please refer to Figure 4 , the document to be audited is input into the extensible knowledge graph constructor, and the constructed global knowledge graph is stored in the graph database. The domain expert or auditor can input the audit target in the interactive rule definition interface, obtain the query statement after translation, and retrieve in the graph database, and finally generate an inconsistency report and provide a visual audit result.

[0088] Therefore, the embodiments of the present application only need to input the document to be audited and the audit target, and can automatically complete the whole auditing process, and finally generate an audit report. This process completely eliminates the inefficiency and error problems of traditional manual auditing, and significantly improves the efficiency and reliability of the audit results.

[0089] It should be understood that although each step in the above flowcharts is shown in sequence according to the direction of the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless explicitly stated herein, there is no strict order limitation for the execution of these steps, and these steps can be executed in other orders. Moreover, at least part of the steps in the above flowcharts can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be alternately executed with at least part of other steps or sub-steps or stages of other steps.

[0090] Based on the foregoing embodiments, the embodiments of the present application provide a text auditing device, which comprises the modules included and the units included in the modules, and can be realized by a processor; of course, it can also be realized by a specific logic circuit; in the implementation process, the processor can be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA), etc.

[0091] Figure 5 For the structure block diagram of the text auditing device provided in the embodiments of the present application, please refer to Figure 5 , a text auditing device, comprising: an auditing data input module S510, a knowledge graph construction module S520, and an auditing report output module S530.

[0092] The auditing data input module S510 is configured to obtain an auditing target and a document to be audited, and the auditing target comprises a rule for auditing the document to be audited. The knowledge graph construction module S520 is configured to obtain a global knowledge graph corresponding to the document to be audited according to an association relationship between a plurality of target objects included in the document to be audited. The auditing report output module S530 is configured to convert the auditing target into a query statement, and retrieve in the global knowledge graph based on the obtained query statement to obtain and output an auditing report.

[0093] In an embodiment, the auditing data input module S510 is specifically configured to display a rule interface, the rule interface is used to input a rule described in a declarative programming language; and the auditing target is obtained in response to a user input operation on the rule interface.

[0094] In an embodiment, the knowledge graph construction module S520 is specifically configured to split the to-be-audited document into a plurality of target text blocks; extract, according to a preset large language model and a preset ontology, an association relationship between a plurality of target objects included in each target text block in the plurality of target text blocks, to obtain a plurality of local knowledge graphs corresponding to the plurality of target text blocks, the preset ontology including a type of a target object that needs to be focused on and a type of an association relationship of any two target objects that need to be focused on in a field to which the to-be-audited document belongs, the preset ontology being obtained by learning, by a target model, sample texts in the field to which the to-be-audited document belongs, and merge the plurality of local knowledge graphs to obtain a global knowledge graph.

[0095] In an embodiment, the knowledge graph construction module S520 is specifically further configured to unify a name of each target object included in each knowledge graph of the plurality of local knowledge graphs, and determine a unique identifier of each target object in the global knowledge graph, to obtain a plurality of processed local knowledge graphs; merge the plurality of processed local knowledge graphs according to the unique identifier of each target object in the plurality of processed local knowledge graphs, to obtain the global knowledge graph.

[0096] In an embodiment, the audit report output module S530 is specifically configured to translate the audit target into a query statement executable by a graph database.

[0097] In an embodiment, the audit report output module S530 is specifically further configured to perform retrieval in the graph database based on the obtained query statement, to obtain a target retrieval result that does not satisfy the query statement, the target retrieval result including a query object that does not satisfy the query statement and a target parameter corresponding to the query object, and the plurality of target objects including the query object.

[0098] In an embodiment, the audit report output module S530 is specifically further configured to count and obtain a number of all query objects included in the target retrieval result; obtain an output order of each query object and a corresponding target parameter according to a position of each query object in the to-be-audited document, and generate a link identifier corresponding to each query object, the link identifier being used to jump to the position of the corresponding query object in the to-be-audited document; obtain a reason for not satisfying the query statement according to the target parameter corresponding to each query object; and generate and output the audit report according to the number of all query objects, the output order of each query object and the corresponding target parameter, the link identifier corresponding to each query object, and the reason.

[0099] The text review device provided in this application embodiment can obtain review targets and documents to be reviewed. The review targets include rules for reviewing the documents. Based on the relationships between multiple target objects included in the documents to be reviewed, a global knowledge graph corresponding to the documents to be reviewed is obtained. The review targets are converted into query statements, and the global knowledge graph is searched based on the obtained query statements to obtain and output a review report. Therefore, this application embodiment only requires inputting the documents to be reviewed and the review targets to automatically complete the entire review process and ultimately generate a clear and easy-to-understand review report, eliminating the inefficiency and errors of traditional manual review and significantly improving review efficiency and reliability.

[0100] The description of the system embodiments above is similar to that of the method embodiments above, and has similar beneficial effects. For technical details not disclosed in the system embodiments of this application, please refer to the description of the method embodiments of this application for understanding.

[0101] It should be noted that, in the embodiments of this application... Figure 5 The module division of the text review device shown is illustrative and represents only one logical functional division; in actual implementation, other division methods may be used. Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, exist as separate physical units, or be integrated into one unit by two or more units. The integrated units can be implemented in hardware, as software functional units, or a combination of both.

[0102] It should be noted that, in the embodiments of this application, if the above-described methods are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an electronic device to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), magnetic disks, or optical disks. Thus, the embodiments of this application are not limited to any specific hardware and software combination.

[0103] Figure 6 This is a schematic diagram of the structure of the computer device provided in the embodiments of this application. Please refer to... Figure 6 This application provides a computer device, which can be an electronic device, such as a mobile phone, a watch, and a vehicle terminal, etc., and its internal structure diagram can be as follows. Figure 6As shown, the computer device includes a processor 620, memory, and a network interface 640 connected via a system bus 610. The processor 620 provides computing and control capabilities. The memory includes a non-volatile storage medium 631 and internal memory 632. The non-volatile storage medium 631 stores an operating system, computer programs, and a database. The internal memory 632 provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium 631. The database is used to store data. The network interface 640 is used to communicate with external terminals via a network connection. When the computer program is executed by the processor 620, it implements the methods described above.

[0104] This application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method provided in the above embodiments.

[0105] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0106] In one embodiment, the text review device provided in this application can be implemented as a computer program, which can be implemented in the form of, for example, Figure 6 The device operates on the computer device shown. The memory of the computer device can store the various program modules that make up the above-described apparatus. The computer program, composed of the various program modules, causes the processor to execute the steps of the methods in the various embodiments of this application described in this specification.

[0107] It should be noted that the descriptions of the storage medium and device embodiments above are similar to those of the method embodiments above, and have similar beneficial effects. For technical details not disclosed in the storage medium, storage medium, and device embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0108] It should be understood that the term "one embodiment" or "an embodiment" or "some embodiments" as used herein means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the application. Thus, the appearances of the phrase "in one embodiment" or "in an embodiment" or "in some embodiments" in various places throughout the specification are not necessarily referring to the same embodiment. Further, the particular features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that the sequence of steps in the above-described processes is not meant to be limiting, and that the steps can be executed in any suitable order, as determined by the function and logic of the steps, and that the execution of the steps should not limit the implementation of the embodiments of the application. The sequence of the above-described embodiments of the application is merely for description, and does not represent the advantages or disadvantages of the embodiments. The above description of the various embodiments tends to emphasize differences between the various embodiments, and the same or similar parts can be mutually referred to, and are not described again for the sake of brevity.

[0109] The term "and / or", as used herein, merely describes association between associated objects, and can represent three conditions: A and / or B; both A and B exist together, A exists alone, and B exists alone.

[0110] It should be noted that the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0111] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The above-described embodiments are merely illustrative. For example, the division of the modules is merely a logical functional division, and actual implementation can have another division manner, such as: a plurality of modules or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed components can be indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0112] The modules described above as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules; they can be located in one place or distributed on multiple network units; and some or all of the modules can be selected as needed to achieve the purposes of the embodiments.

[0113] In addition, the functional modules in the embodiments of the present application can be integrated in one processing unit, or each module can be a separate unit, or two or more modules can be integrated in one unit; the integrated modules can be realized in the form of hardware or in the form of hardware plus software functional units.

[0114] Those skilled in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction related hardware, and the foregoing program can be stored in a computer readable storage medium, and the program executes the steps including the above-mentioned method embodiments when executed; and the foregoing storage medium includes mobile storage devices, read only memory (ROM), magnetic discs or optical discs and various storage medium that can store program codes.

[0115] Alternatively, the integrated units of the present application, if implemented in the form of software functional modules and sold or used as independent products, can also be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present application can be embodied in the form of software products, and the computer software products are stored in a storage medium, including a number of instructions to make an electronic device execute all or part of the methods described in the embodiments of the present application. The foregoing storage medium includes mobile storage devices, ROM, magnetic discs or optical discs and various storage medium that can store program codes.

[0116] The methods disclosed in the several method embodiments provided by the present application can be combined arbitrarily without conflict to obtain new method embodiments.

[0117] The features disclosed in the several product embodiments provided by the present application can be combined arbitrarily without conflict to obtain new product embodiments.

[0118] The features disclosed in the several method or device embodiments provided by the present application can be combined arbitrarily without conflict to obtain new method embodiments or device embodiments.

[0119] The above merely provides the implementation of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of the change or replacement within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A text review method, characterized in that, The method includes: Obtain the audit objectives and the documents to be audited, wherein the audit objectives include the rules for auditing the documents to be audited; Based on the relationships between the multiple target objects included in the document to be reviewed, a global knowledge graph corresponding to the document to be reviewed is obtained; each target object refers to a core concept carrier that has practical significance in a specific domain. The audit objective is converted into a query statement, and the audit report is obtained and output based on the obtained query statement in the global knowledge graph.

2. The text review method according to claim 1, characterized in that, The step of obtaining a global knowledge graph corresponding to the document to be reviewed based on the relationships between multiple target objects included in the document to be reviewed includes: The document to be reviewed is divided into multiple target text blocks; Based on the preset large language model and the preset ontology, the association relationship between multiple target objects included in each of the multiple target text blocks is extracted to obtain multiple local knowledge graphs corresponding to the multiple target text blocks. The preset ontology includes the type of target objects that need to be paid attention to in the domain to which the document to be reviewed belongs, as well as the type of association relationship between any two target objects that need to be paid attention to. The preset ontology is obtained by learning the sample text of the domain to which the document to be reviewed belongs through the target model. The multiple local knowledge graphs are merged to obtain the global knowledge graph.

3. The text review method according to claim 2, characterized in that, The process of merging the multiple local knowledge graphs to obtain the global knowledge graph includes: The names of each target object included in each of the multiple local knowledge graphs are unified, and a unique identifier for each target object in the global knowledge graph is determined to obtain the multiple processed local knowledge graphs. Based on the unique identifier of each target object in the processed multiple local knowledge graphs, the processed multiple local knowledge graphs are merged to obtain the global knowledge graph.

4. The text review method according to claim 1, characterized in that, The method further includes: The global knowledge graph is stored in a graph database; The step of converting the audit target into a query statement includes: The audit objectives are translated into query statements that the graph database can execute.

5. The text review method according to claim 4, characterized in that, The process of retrieving and outputting an audit report based on the obtained query statement in the global knowledge graph includes: Based on the obtained query statement, a search is performed in the graph database to obtain search results that do not satisfy the query statement; Based on the target retrieval results that do not satisfy the query statement, an audit report is obtained and output. The target retrieval results include query objects that do not satisfy the query statement and target parameters corresponding to the query objects. The multiple target objects include the query objects.

6. The text review method according to claim 5, characterized in that, The step of obtaining and outputting an audit report based on the target retrieval results that do not satisfy the query statement includes: The total number of all query objects included in the target retrieval results is counted and obtained. According to the position of each query object in the document to be reviewed in the target retrieval results, the output order of each query object and the corresponding target parameters is obtained, and a link identifier corresponding to each query object is generated. The link identifier is used to jump to the position of the corresponding query object in the document to be reviewed. Based on the target parameters corresponding to each query object, the reasons why the query statement is not satisfied are obtained; The audit report is generated and output based on the total number of all query objects, the output order of each query object and its corresponding target parameters, the link identifier corresponding to each query object, and the reason.

7. The text review method according to claim 4, characterized in that, The acquisition of audit targets includes: The rules interface is used to input rules described in a declarative programming language. In response to the user's input on the rules interface, the audit target is obtained.

8. A text verification device, characterized in that, include: The audit data input module is used to obtain the audit target and the document to be audited, wherein the audit target includes the rules for auditing the document to be audited; The knowledge graph construction module is used to obtain a global knowledge graph corresponding to the document to be reviewed based on the relationships between multiple target objects included in the document to be reviewed; each target object refers to a core concept carrier with practical significance in a specific domain. The audit report output module is used to convert the audit target into a query statement, and retrieve and output the audit report based on the obtained query statement in the global knowledge graph.

9. A computer device comprising a memory and a processor, the memory storing a computer program executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7.

Images