An automatic audit method and audit system based on knowledge graph

Abstract

The present invention discloses a system automatic audit method based on knowledge graph, which defines and automatically constructs the system knowledge graph, builds a suspected clue acquisition model based on the system knowledge graph, and establishes a system automatic audit module for receiving the audit instructions input by the user; the system automatic audit module obtains the audit results according to the audit instructions input by the user, and feeds back the design results to the user to complete the automatic audit work. The present invention designs and constructs the system knowledge graph from two dimensions: organizational hierarchy and label theme, quickly finds audit clues through the system knowledge graph, and improves audit efficiency.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and big data technology, and in particular to a system automatic audit method and audit system based on knowledge graph. Background Art

[0002] Enterprise system archives will become an increasingly important hidden asset of enterprises and a core data resource pool formed in the development of enterprises and organizations. How to give full play to the inherent constraint value of system documents and avoid process risks will be the core of the new generation of smart enterprise management. With the continuous development of organizations and the gradual informatization process over the years, various types and levels of internal systems are also increasing, and their data presents the following three characteristics:

[0003] 1. The total amount of system archives is large, the degree of digitization is high, there are a large number of electronic documents, and with the continuous expansion of business, the number of documents continues to expand;

[0004] 2. The knowledge value contained in the document is high. Not only does the file itself have voucher value, but the knowledge value density contained in a single file is also very high. For example, the production specification contains a large number of standardized knowledge points of the production process. These points are scattered in various documents and lack sufficient association;

[0005] 3. When faced with retrieval scenarios, due to the close knowledge associations between documents (e.g. a system of a subordinate organization is formulated based on a system of a superior organization), traditional knowledge retrieval technology only finds text-related documents and knowledge from the perspective of search matching, and cannot retrieve the associated knowledge between documents.

[0006] Among them, the audit of the system is an important measure to ensure the process assets of the enterprise organization. The system audit discovers the clues of the system audit by auditing the compliance, effectiveness and integrity of the system. The clues of the system audit include the conflict between the system formulated by the lower-level organization and the system of the upper level. At present, the audit inspection of the internal system documents of the enterprise organization is basically carried out by manual on-site audit. The auditor obtains the organization system library on-site and conducts manual audit on each audit point.

[0007] At present, the problems or difficulties in the internal system audit of enterprise organizations include:

[0008] 1. Manual auditing is time-consuming, inefficient and complex. Auditors need to learn the superior system and interpret the audit system one by one. When encountering audit questions, they need to confirm repeatedly.

[0009] 2. Manual auditing is difficult to be transformed into standardized and information-based auditing. It relies too much on the experience of auditors, and the auditors’ expert experience cannot be accumulated.

[0010] 3. There are knowledge islands between institutional documents and knowledge, and it is impossible to uniformly maintain and generate institutional knowledge associations. Summary of the invention

[0011] The purpose of the present invention is to overcome the shortcomings of the prior art and to provide an automatic system audit method and audit system based on a knowledge graph. The method of the present invention defines and constructs a system knowledge graph, fully explores the relevance and hierarchy of system norms, stores and analyzes system norms in a way that is easy for humans to understand, and ultimately achieves intelligent system audit capabilities: using machine learning algorithms to automatically conduct consistency checks between systems, achieve system rule and clause-level verification, and identify audit clues of inconsistencies between clauses of different organizations at the same level, different systems in the same organization, and similar systems at upper and lower levels.

[0012] In order to achieve the purpose of the above invention, the technical solution provided by the patent of this invention is as follows:

[0013] A system automatic audit method based on knowledge graph, the method specifically includes the following steps:

[0014] S1, define the institutional knowledge graph and establish an institutional knowledge graph model, which includes entity type definition and relationship type definition;

[0015] S2, extracts entity knowledge and institutional relations in the institutional knowledge graph model in S1;

[0016] S3, based on S2, constructs an institutional knowledge graph;

[0017] S4, based on the system knowledge graph constructed in S3, builds a suspected clue acquisition model, conducts automatic system audit to obtain audit clues, and feeds back the obtained audit clues to the auditors;

[0018] S5, auditors complete the automatic audit work based on the feedback audit results.

[0019] The entity types in S1 above include system, unit, clause and audit tag, among which the attributes of the system entity include system name, system issuance number, system issuance time, system summary and system text, and the attributes of the unit entity include unit name.

[0020] The relationship types in S1 above include the abolition and dependency relationships between systems, the relationship between system issuing units, and the superior-subordinate relationship and inclusion relationship between units.

[0021] The above-mentioned inclusion relationships include the inclusion relationship between system and clauses, the inclusion relationship between system and labels, and the inclusion relationship between clauses and labels.

[0022] The specific steps for extracting entity knowledge and institutional relationships in the institutional knowledge graph model in S2 above are:

[0023] S11, establish a field extraction and alignment module, use the BMESO annotation set to annotate the system, extract the name, issuance number, issuance time field and issuing unit field of the system through the BILSTM+CRF extraction model, and establish a system library. After the field extraction is completed, align the issuing unit to complete the entity knowledge extraction of the system and the unit, as well as the extraction of the issuing unit relationship and subordinate relationship;

[0024] S12, establish an institutional relationship mining module, through which the institutional relationship mining module is used to extract institutional relationships of abolition and dependency between institutions. The institutional summary contains institutions that have dependency and abolition relationships with the target institution. The BILSTM+CRF extraction model is used to extract the name and document number of the dependent or abolished institution for the dependent and abolished institution, where the document number is directly located after the institution name; the institutional entities of the dependent or abolished institution and the institutional entities in the institutional database are aligned, and the institutions in the institutional database are matched by the institutional names and document numbers of the dependent or abolished institutions. At the same time, the relationship between the target institution and the associated institution is judged based on the key trigger words; the key trigger words for the dependency relationship include "according to", "based on", "based on" and "according to", and the key trigger words for the abolition relationship include "abolition", "invalidation" and "repealing";

[0025] S13, establish a clause parsing module, and complete the entity knowledge extraction of clauses and the extraction of the relationship between the system and clauses through the clause parsing module; parse the system summary file and the system text file through the clause parsing module to obtain the basic information and clause information of the system; the basic information of the system includes the system name, system issuance number, system issuance unit, system issuance time, system summary and system text file, and the system clause information includes each clause in the system text. The system clause is a collection of several continuous natural paragraphs in the system text, and the system clause is the semantic unit with the smallest strength; the clause parsing can be divided according to the rules, such as segmenting according to the characteristics of "the first article", "the second article"... "the Nth article";

[0026] S14, establish a label module, through which the entity knowledge of the audit label and the relationship between the system and the label, as well as the relationship between the clause and the label, are extracted, and the audit label is extracted through the entity extraction model; the sentences or contents containing the audit label are used as audit items, and the audit items are audited for audit scenarios with inconsistent values ​​and audit scenarios with inconsistent scope and process.

[0027] The extraction of the issuing time and issuing unit in the above S11 is performed by limiting the extraction and filtering of the two fields of issuing time and issuing unit in the last several short text contents of the summary file. The relationship between the system name and the issuing unit is located in the header short text or the header table of the summary file. The extraction of the relationship between the system name and the issuing unit is completed by limiting the relative position of the fields of the relationship between the system name and the issuing unit.

[0028] The audit label in the audit scenario with inconsistent values ​​in the above S14 is the specific system extracted by the entity extraction model as the audit subject. For the clauses containing the same audit label, the candidate audit item statements are located through the trigger words with inconsistent values, and the sentences in the candidate audit item statements whose subsequent words contain numerical features are used as audit item statements; the audit label in the audit scenario with inconsistent scope processes is the specific scope process name extracted by the entity extraction model, and the number of steps in the sub-process contained in the scope class clause is used as the audit item pointing to the relationship attribute of the audit subject.

[0029] The above-mentioned numerical inconsistency trigger words include "less than", "more than", "at most", "at least", "lower than", "higher than", "no less than", "not higher than", "not higher than" and "not lower than"; the said valid range clauses include multiple consecutive steps and the sequence numbers at the beginning of the steps are increasing.

[0030] The construction of the suspected clue acquisition model in S4 above specifically includes the following steps:

[0031] S21, establish a system pair screening module, and screen the current systems through system relationships. The specific screening conditions are as follows: the system pairs are systems that are formulated by the superior and subordinate units and have a dependent relationship, and the two clauses from the two systems contain the same audit tag; the two clause pairs containing the same audit tag serve as audit clues to assist auditors in quickly locating audit clues; audit item clues are screened through the system pair screening module; among them, the abolished system is not used as the basis for system relationship screening;

[0032] S22, establish a suspected numerical inconsistency audit module, extract specific numerical values ​​from two audit items according to the numerical inconsistency trigger word, and judge whether the numerical values ​​are the same; if the numerical values ​​are different, the two audit items are used as audit clues; if the numerical values ​​are the same, the two audit items are not used as audit clues;

[0033] S23, establish a scope process inconsistency audit module to determine whether the number of steps of the two audit item attributes is the same. When the number of steps is the same, it is not used as an audit clue; when the number of steps is different, the Jaccard similarity of the two clauses after word segmentation and the similarity of the semantic vectors after Bert encoding are calculated, and then the final similarity value is generated by weighting. When the final similarity value is higher than 0.8, the two audit items are used as audit clues; the formula for calculating the similarity of clause pairs is as follows, where,

[0034] sim(i,j)=α*Jaccard(i,j)+(1-α)Bert_sim(i,j), i and j represent two terms, s i is the set of participles of clause i;

[0035] Bert sim(i，j) = cosine(V i , V j ), V i is the sentence vector of clause i after Bert encoding, where α is 0.8.

[0036] A knowledge graph-based system automatic audit system is implemented based on any of the above-mentioned knowledge graph-based system automatic audit methods, including the following:

[0037] The institutional knowledge graph module is used to establish an institutional knowledge graph model, which includes entity type definitions and relationship type definitions, and extract entity knowledge and institutional relationships from the institutional knowledge graph model;

[0038] The institutional knowledge graph module is used to construct the institutional knowledge graph based on the entity knowledge and institutional relations extracted from the institutional knowledge graph model;

[0039] The suspected clue acquisition module is used to build a suspected clue acquisition model based on the system knowledge graph, conduct automatic system audit to obtain audit clues, and feed back the obtained audit clues to the auditors;

[0040] The audit clue sending module is used to feed back the obtained audit clues to the auditors, and the auditors complete the automatic audit work according to the fed-back audit clue results.

[0041] The method of the present invention has achieved the following positive and beneficial effects through practice:

[0042] 1. The present invention automatically establishes a system knowledge graph. At the same time, the system knowledge graph can be combined with expert experience to complete the precipitation and update of expert experience. Through graph-based association queries, audit clues can be quickly found to improve efficiency.

[0043] 2. The network-based audit clues of the present invention are easy to interpret and trace, and combined with knowledge association, the comprehensiveness and accuracy of the audit are improved.

[0044] 3. The present invention avoids information islands through a unified and standardized system knowledge graph, making it convenient for relevant personnel to quickly upload, query and update. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flow chart for constructing a system knowledge graph in a system automatic audit method based on a knowledge graph of the present invention.

[0046] Figure 2 It is a flow chart for constructing a suspected clue model in a knowledge graph-based system automatic audit method of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention is described below by the specific examples shown in the accompanying drawings. However, it should be understood that these descriptions are only exemplary and are not intended to limit the scope of the present invention. In addition, in the following description, the description of well-known structures and technologies is omitted to avoid unnecessary confusion of the concept of the present invention.

[0048] like Figure 1 As shown, a system automatic audit method based on knowledge graph, the method specifically includes the following steps:

[0049] S1, define the institutional knowledge graph and establish an institutional knowledge graph model, which includes entity type definition and relationship type definition;

[0050] S2, extracts entity knowledge and institutional relations in the institutional knowledge graph model in S1;

[0051] S3, based on S2, constructs an institutional knowledge graph;

[0052] S4, based on the system knowledge graph constructed in S3, builds a suspected clue acquisition model, conducts automatic system audit to obtain audit clues, and feeds back the obtained audit clues to the auditors;

[0053] S5, auditors complete the automatic audit work based on the feedback audit results.

[0054] The entity types in S1 above include system, unit, clause and audit tag, among which the attributes of the system entity include system name, system issuance number, system issuance time, system summary and system text, and the attributes of the unit entity include unit name.

[0055] The relationship types in S1 above include the abolition and dependency relationships between systems, the relationship between system issuing units, and the superior-subordinate relationship and inclusion relationship between units.

[0056] The above-mentioned inclusion relationships include the inclusion relationship between system and clauses, the inclusion relationship between system and labels, and the inclusion relationship between clauses and labels.

[0057] The specific steps for extracting entity knowledge and institutional relationships in the institutional knowledge graph model in S2 above are:

[0058] S11, establish a field extraction and alignment module, use the BMESO annotation set to annotate the system, extract the name, issuance number, issuance time field and issuing unit field of the system through the BILSTM+CRF extraction model, and establish a system library. After the field extraction is completed, align the issuing unit to complete the entity knowledge extraction of the system and the unit, as well as the extraction of the issuing unit relationship and subordinate relationship;

[0059] S12, establish an institutional relationship mining module, through which the institutional relationship mining module is used to extract institutional relationships of abolition and dependency between institutions. The institutional summary contains institutions that have dependency and abolition relationships with the target institution. The BILSTM+CRF extraction model is used to extract the name and document number of the dependent or abolished institution for the dependent and abolished institution, where the document number is directly located after the institution name; the institutional entities of the dependent or abolished institution and the institutional entities in the institutional database are aligned, and the institutions in the institutional database are matched by the institutional names and document numbers of the dependent or abolished institutions. At the same time, the relationship between the target institution and the associated institution is judged based on the key trigger words; the key trigger words for the dependency relationship include "according to", "based on", "based on" and "according to", and the key trigger words for the abolition relationship include "abolition", "invalidation" and "repealing";

[0060] S13, establish a clause parsing module, and complete the extraction of the entity knowledge of the clauses and the extraction of the relationship between the system and the clauses through the clause parsing module; parse the system summary file and the system text file through the clause parsing module to obtain the basic information and clause information of the system; the basic information of the system includes the system name, system issuance number, system issuance unit, system issuance time, system summary and system text file, and the system clause information includes each clause in the system text. The system clause is a collection of several continuous natural paragraphs in the system text, and the system clause is the semantic unit with the smallest strength;

[0061] S14, establish a label module, through which the entity knowledge of the audit label and the relationship between the system and the label, as well as the relationship between the clause and the label, are extracted, and the audit label is extracted through the entity extraction model; the sentences or contents containing the audit label are used as audit items, and the audit items are audited for audit scenarios with inconsistent values ​​and audit scenarios with inconsistent scope and process.

[0062] The extraction of the issuing time and issuing unit in the above S11 is performed by limiting the extraction and filtering of the two fields of issuing time and issuing unit in the last several short text contents of the summary file. The relationship between the system name and the issuing unit is located in the header short text or the header table of the summary file. The extraction of the relationship between the system name and the issuing unit is completed by limiting the relative position of the fields of the relationship between the system name and the issuing unit.

[0063] The audit label in the audit scenario with inconsistent values ​​in the above S14 is the specific system extracted by the entity extraction model as the audit subject. For the clauses containing the same audit label, the candidate audit item statements are located through the trigger words with inconsistent values, and the sentences in the candidate audit item statements whose subsequent words contain numerical features are used as audit item statements; the audit label in the audit scenario with inconsistent scope processes is the specific scope process name extracted by the entity extraction model, and the number of steps in the sub-process contained in the scope class clause is used as the audit item pointing to the relationship attribute of the audit subject.

[0064] The above-mentioned numerical inconsistency trigger words include "less than", "more than", "at most", "at least", "lower than", "higher than", "no less than", "not higher than", "not higher than" and "not lower than"; the said valid range clauses include multiple consecutive steps and the sequence numbers at the beginning of the steps are increasing.

[0065] like Figure 2 As shown, the construction of the suspected clue acquisition model in S4 above specifically includes the following steps:

[0066] S21, establish a system pair screening module, and screen the current systems through system relationships. The specific screening conditions are as follows: the system pairs are systems that are formulated by the superior and subordinate units and have a dependent relationship, and the two clauses from the two systems contain the same audit tag; the two clause pairs containing the same audit tag serve as audit clues to assist auditors in quickly locating audit clues; audit item clues are screened through the system pair screening module; among them, the abolished system is not used as the basis for system relationship screening;

[0067] S22, establish a suspected numerical inconsistency audit module, extract specific numerical values ​​from two audit items according to the numerical inconsistency trigger word, and judge whether the numerical values ​​are the same; if the numerical values ​​are different, the two audit items are used as audit clues; if the numerical values ​​are the same, the two audit items are not used as audit clues;

[0068] S23, establish a scope process inconsistency audit module to determine whether the number of steps of the two audit item attributes is the same. When the number of steps is the same, it is not used as an audit clue; when the number of steps is different, the Jaccard similarity of the two clauses after word segmentation and the similarity of the semantic vectors after Bert encoding are calculated, and then the final similarity value is generated by weighting. When the final similarity value is higher than 0.8, the two audit items are used as audit clues; the formula for calculating the similarity of clause pairs is as follows, where,

[0069] sim(i,j)=α*Jaccard(i,j)+(1-α)Bert_sim(i,j), i and j represent two terms, s i is the set of participles of clause i;

[0070] Bert sim(i，j) = cosine(V i , V j ), V i is the sentence vector of clause i after Bert encoding, where α is 0.8.

[0071] A knowledge graph-based system automatic audit system is implemented based on any of the above-mentioned knowledge graph-based system automatic audit methods, including the following:

[0072] The institutional knowledge graph module is used to establish an institutional knowledge graph model, which includes entity type definitions and relationship type definitions, and extract entity knowledge and institutional relationships from the institutional knowledge graph model;

[0073] The institutional knowledge graph module is used to construct the institutional knowledge graph based on the entity knowledge and institutional relations extracted from the institutional knowledge graph model;

[0074] The suspected clue acquisition module is used to build a suspected clue acquisition model based on the system knowledge graph, conduct automatic system audit to obtain audit clues, and feed back the obtained audit clues to the auditors;

[0075] The audit clue sending module is used to feed back the obtained audit clues to the auditors, and the auditors complete the automatic audit work according to the fed-back audit clue results.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, ordinary technicians in the field should understand that the specific implementation methods of the invention can still be modified or some technical features can be replaced by equivalents without departing from the spirit of the technical solution of the present invention, which should be included in the scope of the technical solution for protection of the present invention.

Claims

1. A knowledge graph-based automatic audit method for systems. It is characterized in that The method specifically comprises the following steps: S1, define the institutional knowledge graph and establish an institutional knowledge graph model, which includes entity type definition and relationship type definition; S2, extracts entity knowledge and institutional relations in the institutional knowledge graph model in S1; The specific steps of extracting entity knowledge and institutional relationships in the institutional knowledge graph model in S2 are: S11, establish a field extraction and alignment module, use the BMESO annotation set to annotate the system, extract the name, issuance number, issuance time field and issuing unit field of the system through the BILSTM+CRF extraction model, and establish a system library. After the field extraction is completed, align the issuing unit to complete the entity knowledge extraction of the system and the unit, as well as the extraction of the issuing unit relationship and subordinate relationship; S12, establish an institutional relationship mining module, and use the institutional relationship mining module to complete the institutional relationship extraction of the abolition and dependency relationship between institutions. The institutional summary contains institutions that have a dependency and abolition relationship with the target institution. The BILSTM+CRF extraction model is used to extract the name and document number of the dependent or abolished institution for the dependent and abolished institution, where the document number is directly located after the institution name; the institutional entities of the dependent or abolished institution and the institutional entities in the institution library are aligned, and the institutions in the institution library are matched by the institutional name and document number of the dependent or abolished institution, and the relationship between the target institution and the related institution is judged according to the key trigger words; the key trigger words of the dependency relationship include "according to", "based on", "based on" and "according to", and the key trigger words of the abolition relationship include "abolition", "invalidation" and "repealed"; S13, establish a clause parsing module, and complete the extraction of the entity knowledge of the clauses and the extraction of the relationship between the system and the clauses through the clause parsing module; parse the system summary file and the system text file through the clause parsing module to obtain the basic information and clause information of the system; the basic information of the system includes the system name, system issuance number, system issuance unit, system issuance time, system summary and system text file, and the system clause information includes each clause in the system text. The system clause is a collection of several continuous natural paragraphs in the system text, and the system clause is the semantic unit with the smallest strength; S14, establish a label module, through which the entity knowledge of the audit label and the relationship between the system and the label, as well as the relationship between the clause and the label, are extracted, and the audit label is extracted through the entity extraction model; the sentence or content containing the audit label is used as the audit item, and the audit item is audited for the audit scenario with inconsistent values ​​and the audit scenario with inconsistent scope and process; S3, based on S2, constructs an institutional knowledge graph; S4, based on the system knowledge graph constructed in S3, builds a suspected clue acquisition model, conducts automatic system audit to obtain audit clues, and feeds back the obtained audit clues to the auditors; The construction of the suspected clue acquisition model in S4 specifically includes the following steps: S21, establish a system pair screening module, and screen the current systems through system relationships. The specific screening conditions are as follows: the system pairs are systems that are formulated by the superior and subordinate units and have a dependent relationship, and the two clauses from the two systems contain the same audit tag; the two clause pairs containing the same audit tag serve as audit clues to assist auditors in quickly locating audit clues; audit item clues are screened through the system pair screening module; among them, the abolished system is not used as the basis for system relationship screening; S22, establish a suspected numerical inconsistency audit module, extract specific numerical values ​​from two audit items according to the numerical inconsistency trigger word, and judge whether the numerical values ​​are the same; if the numerical values ​​are different, the two audit items are used as audit clues; if the numerical values ​​are the same, the two audit items are not used as audit clues; S23, establish a scope process inconsistency audit module to determine whether the number of steps of the two audit item attributes is the same. When the number of steps is the same, it is not used as an audit clue; when the number of steps is different, the Jaccard similarity of the two clauses after word segmentation and the similarity of the semantic vectors after Bert encoding are calculated, and then the final similarity value is generated by weighting. When the final similarity value is higher than 0.8, the two audit items are used as audit clues; the formula for calculating the similarity of clause pairs is as follows, where, , , i and j represent two terms, is the set of participles of clause i; , is the sentence vector of clause i after Bert encoding, where Take 0.8; S5, auditors complete the automatic audit work based on the feedback audit results.

2. According to the method of automatic system audit based on knowledge graph according to claim 1, It is characterized in that The entity types in S1 include system, unit, clause and audit tag, wherein the attributes of the system entity include system name, system issuance number, system issuance time, system summary and system text, and the attributes of the unit entity include unit name.

3. According to the method of automatic system audit based on knowledge graph in claim 1, It is characterized in that The relationship types in S1 include the abolition and dependency relationships between systems, the relationship between system issuing units, and the superior-subordinate relationship and inclusion relationship between units.

4. According to the method of automatic system audit based on knowledge graph according to claim 3, It is characterized in that The inclusion relationship includes the inclusion relationship between the system and the clause, the inclusion relationship between the system and the label, and the inclusion relationship between the clause and the label.

5. According to the method of automatic system audit based on knowledge graph in claim 1, It is characterized in that The extraction of the issuing time and issuing unit in S11 is performed by filtering the two fields of issuing time and issuing unit by limiting the extraction of the content of several short texts at the end of the summary file. The relationship between the system name and the issuing unit is located in the header short text or the header table of the summary file. The extraction of the relationship between the system name and the issuing unit is completed by limiting the relative position of the fields of the relationship between the system name and the issuing unit.

6. According to the knowledge graph-based automatic audit method of claim 1, It is characterized in that The audit label in the audit scenario with inconsistent values ​​in S14 is the specific system extracted by the entity extraction model as the audit subject. For the clauses containing the same audit label, the candidate audit item statements are located through the trigger words with inconsistent values, and the sentences in the candidate audit item statements whose subsequent words contain numerical features are used as audit item statements; the audit label in the audit scenario with inconsistent scope processes is the specific scope process name extracted by the entity extraction model, and the number of steps in the sub-process contained in the scope class clause is used as the audit item pointing to the relationship attribute of the audit subject.

7. According to the method of automatic system audit based on knowledge graph in claim 6, It is characterized in that The numerical inconsistency trigger words include "less than", "more than", "at most", "at least", "lower than", "higher than", "no less than", "not higher than", "not higher than" and "not lower than"; the valid range clauses include multiple consecutive steps with increasing sequence numbers at the beginning of the steps.

8. An automatic audit system based on knowledge graph, It is characterized in that The automatic audit method of the system based on any one of the knowledge graphs described in claims 1 to 7 is implemented, including the following: The institutional knowledge graph module is used to establish an institutional knowledge graph model, which includes entity type definitions and relationship type definitions, and extract entity knowledge and institutional relationships from the institutional knowledge graph model; The institutional knowledge graph module is used to construct the institutional knowledge graph based on the entity knowledge and institutional relations extracted from the institutional knowledge graph model; The suspected clue acquisition module is used to build a suspected clue acquisition model based on the system knowledge graph, conduct automatic system audit to obtain audit clues, and feed back the obtained audit clues to the auditors; The audit clue sending module is used to feed back the obtained audit clues to the auditors, and the auditors complete the automatic audit work according to the fed-back audit clue results.

Images