Audit object portrait construction method based on knowledge graph

Abstract

The invention relates to the technical field of knowledge maps, in particular to an audit object portrait construction method based on a knowledge map. The method comprises the following steps: acquiring audit files of different departments in an enterprise, and audit objects and named entities of the different audit files; performing entity identification and attribute value extraction on different audit files, and determining different named entities and attribute values; according to the numerical value change of any two attribute values in the same audit file, determining the change similarity degree of the two attribute values; according to the change similarity degree of any two attribute values in all the audit files, determining the attribute association degree of the two attribute values; determining entity association indexes of the named entities according to attribute association degrees of different named entities among all attribute values; and constructing an audit object portrait model of the enterprise in combination with the entity association indexes among different named entities. According to the method and the device, more hidden relation among audit objects can be disclosed, so that the correlation analysis effect of the overall audit object portrait model is improved.

Description

Technical Field

[0001] The present invention relates to the field of knowledge graph technology, and in particular to a method for constructing an audit object portrait based on a knowledge graph. Background Art

[0002] The audit process of an enterprise is a systematic internal control and risk management process; it mainly analyzes the enterprise's activities and the financial data that changes in the process to examine whether the enterprise is operating normally. It aims to evaluate whether the enterprise's financial data is accurate, whether the operating process is efficient, and whether the risk control system is sound. Using artificial intelligence to construct a lot of audit data into more intuitive graph data to assist auditing is a relatively common audit method.

[0003] In related technologies, audit object portraits are constructed based on traditional direct association relationships of audit objects. Although these relationships can reveal the associations between various problem objects in the audit process, such associations are relatively superficial and direct for audit work, and it is easy to ignore the more hidden connections between these entity objects. There is a lack of analysis of the more hidden connections between audit objects, resulting in poor association analysis results for the overall audit object portrait model. Summary of the invention

[0004] In order to solve the technical problem that the more hidden connections between these entity objects are easily ignored in the related technology, the more hidden connections between the audit objects are lacking in analysis, resulting in poor association analysis effect of the overall audit object portrait model, the present invention provides a method for constructing an audit object portrait based on a knowledge graph, and the technical solution adopted is as follows: The present invention proposes a method for constructing an audit object portrait based on a knowledge graph, the method comprising: Obtain audit files of different departments in the enterprise, as well as audit objects and named entities of different audit files; perform entity recognition and attribute value extraction on different audit files to determine different named entities and attribute values; According to the numerical changes of any two attribute values ​​in the same audit file, the similarity of the changes of the two attribute values ​​is determined; according to the similarity of the changes of any two attribute values ​​in all audit files, the attribute correlation degree of the two attribute values ​​is determined; Select any two named entities as target entities, and determine the entity association index of the target entities according to the attribute association degree between all attribute values ​​of different target entities; Combine the entity association indicators between different named entities to build an enterprise's audit object portrait model.

[0005] Furthermore, the determining of the similarity of changes of any two attribute values ​​in the same audit file according to the numerical changes of the two attribute values ​​includes: According to the temporal changes of each attribute value in the audit file and word vector analysis, determine the temporal change vector of the attribute value; The degree of similarity of changes in the two attribute values ​​is determined based on the cosine similarity between the change vectors corresponding to any two attribute values.

[0006] Furthermore, the step of determining a time-series change vector of an attribute value according to the time-series change of each attribute value in the audit file and word vector analysis includes: Determine the numerical change of each attribute value in the same audit file in time series and obtain the change amount; perform word vector analysis on the attribute value and obtain the word vector direction of the attribute value; The change vector is constructed with the change amount as the change vector size and the word vector direction as the change vector direction.

[0007] Furthermore, determining the degree of similarity of changes in two attribute values ​​according to the cosine similarity between the change vectors corresponding to any two attribute values ​​includes: The cosine similarity of the change vectors corresponding to the two attribute values ​​is calculated, and the value of the cosine similarity is linearly normalized to obtain the change similarity of the two attribute values.

[0008] Furthermore, determining the attribute correlation degree of two attribute values ​​according to the similarity of changes of any two attribute values ​​in all audit files includes: The audit files containing the same two attribute values ​​are used as the files to be analyzed, and the mean of the similarity of the changes of the two attribute values ​​in all the files to be analyzed is used as the first correlation coefficient; Determine a second correlation coefficient according to the numerical fluctuation of the similarity of the changes of the two attribute values ​​in all the files to be analyzed, wherein the smaller the numerical fluctuation, the larger the second correlation coefficient; The product of the first correlation coefficient and the second correlation coefficient is calculated, and maximum and minimum value normalization processing is performed to obtain the attribute correlation degree.

[0009] Furthermore, determining the second correlation coefficient according to the numerical fluctuation of the similarity of the change of the two attribute values ​​in all the files to be analyzed includes: The standard deviation of the similarity of the changes of the two attribute values ​​in all the files to be analyzed is calculated, and the inverse of the standard deviation is obtained to obtain the second correlation coefficient.

[0010] Furthermore, determining the entity association index of the target entity according to the attribute association degree between all attribute values ​​of different target entities includes: Take any two target entities as comparison entities; take all attribute values ​​under the comparison entities as comparison attribute values; Determine the attribute correlation degree between two contrast attribute values ​​belonging to two contrast entities, which is the contrast correlation degree; The entity association index of the two compared entities is determined according to the numerical distribution of the comparison association degree of all comparison attribute value combinations under the compared entities.

[0011] Furthermore, determining the entity association index of two compared entities according to the numerical distribution of the comparison association degree of all comparison attribute value combinations under the compared entities includes: The average of all the comparison association degrees is calculated, and the product of the average and the maximum value of all the comparison association degrees is used as the entity association index of the two comparison entities.

[0012] Furthermore, the above-mentioned construction of the audit object portrait model of the enterprise by combining the attribute association degree between different attribute values ​​and the entity association index between different named entities includes: Obtaining an initial audit object portrait model based on all the named entities and all the attribute values; The connection lengths of different named entities in the initial audit object portrait model are adjusted based on the entity association index between different named entities, wherein the larger the value of the entity association index, the smaller the connection length between the corresponding two named entities.

[0013] Furthermore, the entity recognition and attribute value extraction of different audit files to determine different named entities and attribute values ​​include: Perform entity recognition based on the Transformer-CRF model to obtain named entities; Attribute values ​​are extracted based on the MoEtransformer model to obtain attribute values.

[0014] The present invention has the following beneficial effects: The present invention obtains the audit files of different departments in an enterprise, as well as the audit objects and named entities of different audit files; performs entity recognition and attribute value extraction on different audit files to determine different named entities and attribute values; determines the similarity of the changes of any two attribute values ​​according to the numerical changes of the two attribute values ​​in the same audit file; determines the attribute association degree of the two attribute values ​​according to the similarity of the changes of any two attribute values ​​in all audit files; selects any two named entities as target entities, and determines the entity association index of the target entity according to the attribute association degree between all attribute values ​​of different target entities; and constructs the audit object portrait model of the enterprise in combination with the entity association indexes between different named entities. Thus, it is possible to conduct a specific analysis of the entity association index in combination with the change relationship between the attributes, and construct the audit object portrait model of the enterprise based on the entity association index, so as to avoid constructing only the directly associated audit object portrait model, improve the analysis of the invisible association, and at the same time, visualize the association between different audit objects, so as to reveal the more hidden connection between the audit objects, so as to improve the association analysis effect of the overall audit object portrait model. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 A flow chart of a method for constructing an audit object portrait based on a knowledge graph provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of a method for constructing an audit object portrait based on a knowledge graph proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0018] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0019] The following is a detailed description of a specific method for constructing an audit object portrait based on a knowledge graph provided by the present invention in conjunction with the accompanying drawings.

[0020] See also Figure 1 , which shows a flow chart of a method for constructing an audit object portrait based on a knowledge graph provided by an embodiment of the present invention, the method comprising: S101: Acquire audit files of different departments in the enterprise, as well as audit objects and named entities of different audit files; perform entity recognition and attribute value extraction on different audit files to determine different named entities and attribute values.

[0021] It is understandable that an enterprise has multiple different departments and organizational structures, and some departments have direct connections, such as the procurement department and the logistics department. The logistics department needs to report the required materials to the procurement department, and the procurement department and the finance department also have direct connections to report the purchase amount. Under this logical chain, the logistics department and the finance department do not have a direct connection, and the relationship between the two cannot be reflected in the conventional audit object portrait. In fact, they have an implicit relationship, that is, the amount of materials required for logistics is directly counted by the finance department. This part of the implicit relationship cannot be made clear in the conventional audit object portrait.

[0022] This solution takes power companies as a specific scenario, but it should be noted that the technology of this solution is not limited to power companies, but can also be applied to most companies that need to be audited. In power companies, each activity requires the mobilization of many departments, and a large number of communication documents will be generated during the mobilization process. At the same time, during power production activities, the company will also generate corresponding order contracts and documents with many other companies. Because in the same time period (the time span is not far), the company's power generation method will not change, and the raw materials required for power generation and the cost of maintaining normal production are also relatively stable. Therefore, in the process of enterprise operation, the capital flow and capital flow size generated by each activity are relatively stable. However, considering that the means of production and the size of corporate team building activities will not remain stable, this will lead to fluctuations in the corresponding capital and financial flows. Based on the above factors, the implicit relationship between various audit objects can be derived, and the relationship between audit objects can be constructed accordingly.

[0023] According to the relevant audit regulations, the audit in power enterprises can be roughly divided into: financial audit, operational audit, risk management audit, internal control audit and many special audits. These have been carried out many times in the past audit process. Although the contents of the audit are different, the audit purposes are mostly the same, so the audit documents and audit objects are mostly similar. Therefore, the audit object in the audit document can be determined according to the preset audit object type. The audit object is the department and file type in the enterprise (such as financial statements, accounting records, etc.).

[0024] Since the audit files contain digital data and text data, it is necessary to analyze the named entities, mark the past audit files, and mark the entities (including attributes), the correlation between entities, and the corresponding entity categories. The correlation between entities and attributes is a relationship of time sequence, cause and effect, and hierarchy such as "belongs to", "therefore", and "instantiation". Entity categories include financial-related entities, organizational structure entities, personnel entities, geographic location entities, time-related entities, and so on. Therefore, named entity recognition and attribute value extraction can be performed.

[0025] Exemplarily, the attributes include department attributes, person in charge attributes, funding attributes and project attributes; for example, department attributes can be the number of department personnel and department budget items, etc., and can also be enterprise funding, historical audit results and the number of various audit issues, and person in charge attributes can specifically be the person in charge's personal resume and previous audit results, etc.; funding attributes can be funding name, funding type and historical audit results, etc.; project attributes can be project life cycle, project theme, project personnel, project content and project implementation status, etc.

[0026] Furthermore, in some embodiments of the present invention, entity recognition and attribute value extraction are performed on different audit files to determine different named entities and attribute values, including: performing entity recognition based on the Transformer-CRF model to obtain named entities; and extracting attribute values ​​based on the MoEtransformer model to obtain attribute values.

[0027] Of course, in other embodiments of the present invention, the BERT model or other deep learning models trained to perform entity recognition and entity classification can also be used to perform entity recognition to obtain named entities. Similarly, regular expressions and entity linking (EL algorithm) can be used to extract several named entities from numerous audit files associated with the audit object.

[0028] Use regular expressions to obtain named entities and several attribute values ​​corresponding to the named entities; then use deep learning-based methods (such as pre-trained Transformer models, BERT, and RoBERTa) to extract many attribute values ​​in the file data.

[0029] It should be noted that entity recognition and attribute value extraction are both common technologies for constructing audit object portraits, and will not be elaborated on here.

[0030] S102: Determine the similarity of changes of any two attribute values ​​in the same audit file according to the numerical changes of the two attribute values; determine the attribute correlation degree of the two attribute values ​​according to the similarity of changes of any two attribute values ​​in all audit files.

[0031] In order to analyze the implicit relationship between various audit objects, that is, the relationship that is not directly connected but is transmitted through multiple layers or caused by some indirect reasons, the correlation between two attribute values ​​is analyzed based on the changes in attribute values.

[0032] Furthermore, in some embodiments of the present invention, the degree of similarity of changes in any two attribute values ​​in the same audit file is determined based on the numerical changes in the two attribute values, including: determining the time series change vector of the attribute value based on the time series change and word vector analysis of each attribute value in the audit file; determining the degree of similarity of changes in the two attribute values ​​based on the cosine similarity between the change vectors corresponding to any two attribute values.

[0033] Among many audit files, audit files involving different attribute values ​​(specific items, such as transaction amount, transaction item, etc.) focus on the impact of the attribute in different aspects. That is, for the same attribute value, under different named entities, although there is still a correlation with other associated attribute values, the specific causal relationship or numerical relationship may be different. These are reflected in the corresponding files, and the same data changes have different impacts. Based on this, the specific correlation between the two attribute values ​​is calculated.

[0034] In an embodiment of the present invention, in order to facilitate the analysis of all attribute values, it is necessary to unify all attribute values. The present invention uses word vectors to achieve the unification of attribute values, that is, the changes of attribute values ​​in the same audit file are represented by word vectors to obtain change vectors.

[0035] Among them, the change of attribute value in the same audit file can be specifically the change of the numerical value of the attribute value in time series, such as the change of transaction amount in January and October. It should be noted that this change process requires the change of the numerical value of the attribute value in the same time period in order to facilitate specific analysis, thereby performing analysis of the change vector.

[0036] Furthermore, in some embodiments of the present invention, based on the temporal changes of each attribute value in the audit file and word vector analysis, the temporal change vector of the attribute value is determined, including: determining the numerical change of each attribute value in the same audit file in time to obtain the change amount; performing word vector analysis on the attribute value to obtain the word vector direction of the attribute value; and constructing a change vector with the change amount as the change vector size and the word vector direction as the change vector direction.

[0037] Specifically, the word2vec model can be used to analyze word vectors, determine the direction of word vectors, and use the amount of change as the size to obtain a change vector.

[0038] After determining the change vector, a similarity analysis can be performed on the change vector. Furthermore, in some embodiments of the present invention, the degree of similarity of the changes of the two attribute values ​​is determined based on the cosine similarity between the change vectors corresponding to any two attribute values, including: calculating the cosine similarity of the change vectors corresponding to the two attribute values, linearly normalizing the value of the cosine similarity, and obtaining the degree of similarity of the changes of the two attribute values.

[0039] Among them, cosine similarity is a similarity calculation method well known to those skilled in the art, which will not be described in detail. By performing linear normalization processing on the cosine similarity value, the degree of similarity of changes in two attribute values ​​is obtained.

[0040] For example, in some embodiments of the present invention, the calculation formula for the degree of similarity of changes in two attribute values ​​may be specifically, for example, as follows: Representation attributes With attributes The similarity of the changes of two attributes; used to describe the correlation characteristics of the changes of two attributes. Representation attributes The change vector of Representation attributes The change vector of Indicates the cosine similarity, and since the value range of cosine similarity is [-1,1], we can use , limiting the value range to [0,1] to achieve a linear normalization effect.

[0041] S103: Select any two named entities as target entities, and determine entity association indexes of the target entities according to the attribute association degrees between all attribute values ​​of different target entities.

[0042] When an enterprise conducts an audit, there are many audit contents and different audit projects have different purposes. In addition, in many audit projects, the same named entity may be used in multiple audit projects. That is, the same named entity may be used in multiple different audit objects.

[0043] In addition, in addition to the causal, temporal, hierarchical and other relationships implied in the texts, the relationships between the elements corresponding to the named entities in the enterprise also involve mutual traction between the corresponding data changes, that is, the change of one will cause the other related data to change together, and these changes will comply with the causal, temporal and other relationships between the two. Therefore, it is necessary to analyze the entity relationship between the target entities.

[0044] Among them, there are certain associations between the target entities. The closest association is direct association, that is, direct business connection is generated in the enterprise architecture, but there are still some implicit associations that need to be specifically analyzed. In order to facilitate the analysis of implicit associations between named entities, association analysis is uniformly performed on any two named entities regardless of whether they are directly associated or not, that is, the entity association indicators of any two target entities are determined.

[0045] Furthermore, in some embodiments of the present invention, the degree of attribute correlation between two attribute values ​​is determined based on the degree of similarity of changes between any two attribute values ​​in all audit files, including: taking audit files containing the same two attribute values ​​as files to be analyzed, and taking the average of the degree of similarity of changes between the two attribute values ​​in all files to be analyzed as a first correlation coefficient; determining a second correlation coefficient based on the numerical fluctuation of the degree of similarity of changes between the two attribute values ​​in all files to be analyzed, wherein the smaller the numerical fluctuation, the larger the second correlation coefficient; calculating the product of the first correlation coefficient and the second correlation coefficient, and performing maximum and minimum value normalization processing to obtain the attribute correlation degree.

[0046] Among them, when two attribute values ​​to be analyzed appear in the audit file at the same time, it can be called a file to be analyzed; when it only contains any one attribute value to be analyzed, or when it does not contain both attribute values ​​to be analyzed, it is not a file to be analyzed.

[0047] The first correlation coefficient represents the mean value of the similarity of the changes of the two attribute values ​​in the files to be analyzed. The larger the value is, the more similar the changes of the two attribute values ​​in all the files to be analyzed are, that is, the two attribute values ​​have a stronger correlation as a whole.

[0048] The second correlation coefficient represents the numerical fluctuation of the similarity of the changes of the two attribute values. The smaller the numerical fluctuation, the more consistent the similarity of the changes of the two attribute values ​​in different files to be analyzed. Considering that the changes of the two attributes in not every file to be analyzed are caused by the changes of each other, these files have negative interference in examining the correlation characteristics of the two attribute values, so the second correlation coefficient is calculated based on the fluctuation of the connection between the changes of the two.

[0049] Furthermore, in some embodiments of the present invention, a second correlation coefficient is determined based on the numerical fluctuations of the similarity of the changes in the two attribute values ​​in all the files to be analyzed, including: calculating the standard deviation of the similarity of the changes in the two attribute values ​​in all the files to be analyzed, and obtaining the second correlation coefficient by taking the opposite of the standard deviation.

[0050] Among them, the larger the value of the second correlation coefficient is, the more consistent the similarity of the changes in the two attribute values ​​in all the files to be analyzed is, that is, the greater the correlation between the causes of the changes in the two attribute values ​​in each file to be analyzed, and the changes in the two attributes are more likely to be caused by changes in each other. At this time, even if it is not directly associated with the target entity, the implicit change association of the attribute value quality inspection can be analyzed.

[0051] The above obtains the relationship between each attribute value (specific goal of the audit) and the distance between the relationships; however, during the audit, due to different audit goals and focus points, the same object (department name) may be a named entity in some audit processes, but in other audit processes it only appears as an attribute value. Therefore, the calculation method of the relationship between named entities and the relationship between attribute values ​​is almost the same. However, as a named entity, when constructing the knowledge graph, it is the superior of the attribute value and has an additional attribute value difference. Therefore, it is necessary to perform statistical analysis on all attribute values ​​under the named entity to determine the entity association index of the named entity.

[0052] Furthermore, in some embodiments of the present invention, the entity association index of the target entity is determined according to the attribute association degree between all attribute values ​​of different target entities, including: taking any two target entities as comparison entities; taking all attribute values ​​under the comparison entities as comparison attribute values; determining the attribute association degree between two comparison attribute values ​​belonging to two comparison entities as the comparison association degree; and determining the entity association index of the two comparison entities according to the numerical distribution of the comparison association degree of all comparison attribute value combinations under the comparison entities.

[0053] In the embodiment of the present invention, one target entity may correspond to multiple attribute values. Therefore, in order to analyze the association between two target entities, it is necessary to traverse the combination of all attribute values.

[0054] For example, target entity 1 corresponds to attribute values ​​a, b and c, and target entity 2 corresponds to attribute values ​​d and e. There may be different corresponding attribute value combinations, such as the combination of (a, d), (a, e), (b, d), (b, e), (c, d), and (c, e). Therefore, a and d can be selected as comparison attribute values, and the attribute correlation degree between a and d is the comparison correlation degree. Therefore, by changing different attribute value combinations, we can obtain the comparison correlation degrees corresponding to each combination of (a, d), (a, e), (b, d), (b, e), (c, d), and (c, e).

[0055] Furthermore, in some embodiments of the present invention, after all contrast association degrees are determined, the entity association index of the two contrast entities is determined based on the numerical distribution of the contrast association degrees of all contrast attribute value combinations under the contrast entities, including: calculating the mean of all contrast association degrees, and taking the product of the mean and the maximum value of all contrast association degrees as the entity association index of the two contrast entities.

[0056] Among them, the mean calculation is performed directly. The larger the mean, the greater the correlation under all attribute value combinations, that is, the greater the overall correlation between the contrasting entities. At the same time, the maximum value of all contrasting correlation degrees is used for further weighting. The larger the maximum value of all contrasting correlation degrees, the greater the correlation between the two contrasting entities in a certain aspect. Therefore, the product of the mean and the maximum value of all contrasting correlation degrees is used as the entity correlation index of the two contrasting entities.

[0057] And adjust different comparison entities at the same time to obtain entity association indicators between all target entities.

[0058] S104: Build an audit object portrait model for the enterprise by combining entity association indicators between different named entities.

[0059] In an embodiment of the present invention, the entity association index between different named entities represents the correlation between the overall attribute changes of different named entities. Therefore, the relationship between audit objects can be analyzed based on the entity association index, so as to determine the close or distant relationship between audit objects and add the close or distant relationship to the traditional audit object portrait model.

[0060] Furthermore, in some embodiments of the present invention, an audit object portrait model of an enterprise is constructed in combination with the degree of attribute association between different attribute values ​​and the entity association index between different named entities, including: obtaining an initial audit object portrait model based on all named entities and all attribute values; adjusting the connection lengths of different named entities in the initial audit object portrait model based on the entity association index between different named entities, wherein the larger the numerical value of the entity association index, the smaller the connection length between the corresponding two named entities.

[0061] Among them, the initial audit object portrait model is the traditional audit object portrait model. First, the portrait framework is constructed based on the vertical relationship between various audit objects, named entities and attributes. Considering the audit process of an enterprise, the number of documents involved is extremely large, and some departments or projects are the common focus of multiple audit tasks, which leads to the situation of cross-connections when constructing the graph. Therefore, we choose to rely on conventional network visualization layout algorithms, such as the Fruchterman-Reingold method and the Yifan Hu layout algorithm, to construct the portrait graph of the audit object.

[0062] Then, according to the relationship between entities and entity association indicators, the relationship lines between entities are constructed. The entity association indicators are used to control the length of the lines. The larger the association indicator, the shorter the line. In an embodiment of the present invention, the opposite number of the entity association indicator can be directly calculated, and the opposite number can be normalized to the maximum and minimum values ​​as the connection length between the corresponding two named entities.

[0063] The present invention obtains the audit files of different departments in the enterprise, as well as the audit objects and named entities of different audit files; performs entity recognition and attribute value extraction on different audit files to determine different named entities and attribute values; determines the similarity of the changes of two attribute values ​​according to the numerical changes of any two attribute values ​​in the same audit file; determines the attribute association degree of the two attribute values ​​according to the similarity of the changes of any two attribute values ​​in all audit files; selects any two named entities as target entities, and determines the entity association index of the target entity according to the attribute association degree between all attribute values ​​of different target entities; and constructs the audit object portrait model of the enterprise in combination with the entity association indexes between different named entities. Thus, it is possible to conduct a specific analysis of the entity association index in combination with the change relationship between the attributes, and construct the audit object portrait model of the enterprise based on the entity association index, so as to avoid constructing only the directly associated audit object portrait model, improve the analysis of the invisible association, and at the same time, visualize the association of different audit objects, so as to reveal the more hidden connection between entity objects, so as to improve the association analysis effect of the overall audit object portrait model.

[0064] It should be noted that the sequence of the above embodiments of the present invention is for description only and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0065] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A method for constructing an audit object portrait based on a knowledge graph, characterized in that: The method comprises: Obtain audit files of different departments in the enterprise, as well as audit objects and named entities of different audit files; perform entity recognition and attribute value extraction on different audit files to determine different named entities and attribute values; According to the numerical changes of any two attribute values ​​in the same audit file, the similarity of the changes of the two attribute values ​​is determined; according to the similarity of the changes of any two attribute values ​​in all audit files, the attribute correlation degree of the two attribute values ​​is determined; Select any two named entities as target entities, and determine the entity association index of the target entities according to the attribute association degree between all attribute values ​​of different target entities; Combine the entity association indicators between different named entities to build an enterprise's audit object portrait model.

2. A method for constructing an audit object portrait based on a knowledge graph as claimed in claim 1, characterized in that: Determining the similarity of changes in any two attribute values ​​according to the numerical changes in the two attribute values ​​in the same audit file includes: According to the temporal changes of each attribute value in the audit file and word vector analysis, determine the temporal change vector of the attribute value; The degree of similarity of changes in the two attribute values ​​is determined based on the cosine similarity between the change vectors corresponding to any two attribute values.

3. A method for constructing an audit object portrait based on a knowledge graph as claimed in claim 2, characterized in that: The step of determining the time-series change vector of the attribute value according to the time-series change of each attribute value in the audit file and the word vector analysis includes: Determine the numerical change of each attribute value in the same audit file in time series and obtain the change amount; perform word vector analysis on the attribute value and obtain the word vector direction of the attribute value; The change vector is constructed with the change amount as the change vector size and the word vector direction as the change vector direction.

4. A method for constructing an audit object portrait based on a knowledge graph as claimed in claim 3, characterized in that: Determining the degree of similarity of changes in two attribute values ​​according to the cosine similarity between the change vectors corresponding to any two attribute values ​​includes: The cosine similarity of the change vectors corresponding to the two attribute values ​​is calculated, and the value of the cosine similarity is linearly normalized to obtain the change similarity of the two attribute values.

5. The method for constructing an audit object portrait based on a knowledge graph as claimed in claim 1, characterized in that: Determining the attribute correlation degree of two attribute values ​​according to the similarity of changes of any two attribute values ​​in all audit files includes: The audit files containing the same two attribute values ​​are used as the files to be analyzed, and the mean of the similarity of the changes of the two attribute values ​​in all the files to be analyzed is used as the first correlation coefficient; Determine a second correlation coefficient according to the numerical fluctuation of the similarity of the changes of the two attribute values ​​in all the files to be analyzed, wherein the smaller the numerical fluctuation, the larger the second correlation coefficient; The product of the first correlation coefficient and the second correlation coefficient is calculated, and maximum and minimum value normalization processing is performed to obtain the attribute correlation degree.

6. A method for constructing an audit object portrait based on a knowledge graph as claimed in claim 5, characterized in that: Determining the second correlation coefficient according to the numerical fluctuation of the similarity of the change of the two attribute values ​​in all the files to be analyzed includes: The standard deviation of the similarity of the changes of the two attribute values ​​in all the files to be analyzed is calculated, and the inverse of the standard deviation is obtained to obtain the second correlation coefficient.

7. The method for constructing an audit object portrait based on a knowledge graph as claimed in claim 1, characterized in that: Determining the entity association index of the target entity according to the attribute association degree between all attribute values ​​of different target entities includes: Take any two target entities as comparison entities; take all attribute values ​​under the comparison entities as comparison attribute values; Determine the attribute correlation degree between two contrast attribute values ​​belonging to two contrast entities, which is the contrast correlation degree; The entity association index of the two compared entities is determined according to the numerical distribution of the comparison association degree of all comparison attribute value combinations under the compared entities.

8. A method for constructing an audit object portrait based on a knowledge graph as claimed in claim 7, characterized in that: Determining the entity association index of two compared entities according to the numerical distribution of the comparison association degree of all comparison attribute value combinations under the compared entities includes: The average of all the comparison association degrees is calculated, and the product of the average and the maximum value of all the comparison association degrees is used as the entity association index of the two comparison entities.

9. The method for constructing an audit object portrait based on a knowledge graph as claimed in claim 1, characterized in that: The audit object portrait model of the enterprise is constructed by combining the attribute association degree between different attribute values ​​and the entity association index between different named entities, including: Obtaining an initial audit object portrait model based on all the named entities and all the attribute values; The connection lengths of different named entities in the initial audit object portrait model are adjusted based on the entity association index between different named entities, wherein the larger the value of the entity association index, the smaller the connection length between the corresponding two named entities.

10. The method for constructing an audit object portrait based on a knowledge graph according to claim 1, characterized in that: The entity recognition and attribute value extraction of different audit files to determine different named entities and attribute values ​​include: Perform entity recognition based on the Transformer-CRF model to obtain named entities; Attribute values ​​are extracted based on the MoEtransformer model to obtain attribute values.

Images