Method for auditing multi-business type text, computer device and computer readable storage medium

Abstract

The application provides a multi-business type text auditing method, a computer device and a computer readable storage medium. The method comprises the following steps: setting a knowledge graph model of multiple business texts, setting a corresponding auditing rule model based on each type of business text, and determining an auditing rule under each auditing rule model; obtaining a target business text to be audited, applying a text classification model to determine the text type of the target business text, extracting key element information of the target business text, and constructing a small knowledge graph according to the key element information and the knowledge graph model; determining whether the auditing rule is a normal auditing rule or a script auditing rule according to the auditing rule of the business text, putting the graph variable into the rule engine, and obtaining an auditing result after executing the auditing rule. The application also provides a computer device and a computer readable storage medium for implementing the above method. The application can automatically audit different types of business texts.

Description

Technical Field

[0001] The present invention relates to the technical field of automated text review, and in particular to a method for automated review of various types of business texts, and also to a computer device and a computer-readable storage medium for implementing the method. Background Art

[0002] With the development of artificial intelligence (AI), automated text review has become a key area of ​​AI research. Since text is unstructured knowledge, effectively reviewing its content has always been a challenge. Automated text review involves numerous AI technologies, such as text classification, text information extraction, knowledge associations between multiple texts, and knowledge reasoning. The current research challenge is how to rationally combine these various AI technologies, particularly in every step and stage of automated text review.

[0003] As a component of artificial intelligence, knowledge graph technology possesses powerful semantic processing and knowledge association capabilities, providing a solution for automated text review. By storing textual knowledge in triples through knowledge graph technology and leveraging graph association and search techniques, it effectively addresses the issue of multi-text review.

[0004] However, because business documents have more auditable elements than general documents, and many business documents are interrelated, auditing a business document requires considering not only the document itself but also other related information. However, existing document auditing methods fail to consider information from other related documents and instead focus solely on the document itself, resulting in poor audit results. Summary of the Invention

[0005] The first object of the present invention is to provide a method for reviewing multiple business-type texts that can be associated with other texts for review.

[0006] A second object of the present invention is to provide a computer device for implementing the above-mentioned multi-business type text review method.

[0007] A third object of the present invention is to provide a computer-readable storage medium for implementing the above-mentioned multi-business type text review method.

[0008] To achieve the first object of the present application, the method for auditing multi-business type text provided by the present application comprises setting a knowledge graph model of multiple business texts, setting a corresponding auditing rule model based on each type of business text, and determining the auditing rule under each auditing rule model; obtaining a target business text to be audited, applying a text classification model to determine the text type of the target business text, and extracting key element information of the target business text, constructing a small knowledge graph from the key element information according to the knowledge graph model; determining whether the business field to which the target business text belongs needs to be combined with external information, if yes, fusing the constructed small knowledge graph with the external information to obtain a complete knowledge graph, and determining the auditing rule of the target business text, otherwise, directly obtaining the corresponding auditing rule of the target business text; determining whether the auditing rule of the business text is a normal auditing rule or a script auditing rule, if it is a normal auditing rule, reading all graph variables of the normal auditing rule, if it is a script auditing rule, reading all graph variables of the script auditing rule, putting the graph variables into a rule engine and executing the auditing rule to obtain an auditing result.

[0009] As can be seen from the above scheme, when auditing the business text, the type of the business text is identified, and the corresponding auditing rule model is obtained, so that the business text of a specific type can be audited in a targeted manner. Moreover, whether the small knowledge graph needs to be fused with external information is determined according to the type of the business text, that is, the target business text is audited by using other business texts and external information, which can improve the auditing quality of the business text.

[0010] A preferred scheme is that determining the auditing rule under each auditing rule model comprises setting a graph model of the auditing rule under each auditing rule model, each graph model comprising at least one graph variable, and each graph variable comprising an entity and an entity element.

[0011] As can be seen from the above scheme, the auditing rule of the business text can be flexibly defined by using the expression mode of the graph variable, so that the auditing of the business text is more efficient.

[0012] A further scheme is that the graph variable comprises at least one jump relationship, and the entity and the entity element are objects having the jump relationship.

[0013] In this way, the user can define multiple different graph variables in advance, and the auditing of different types of business texts can be realized by setting appropriate jump relationships.

[0014] A further scheme is that the graph variable comprises two or more jump relationships, and the number of entities contained in the graph variable is two or more, and there is a jump relationship between adjacent two entities.

[0015] It can be seen from this that by setting multi-level jump relationships and using one graph variable, multiple elements that need to be reviewed can be determined, making the review of business texts more flexible.

[0016] In a preferred solution, determining whether the business domain of the target business text needs to be combined with external information includes: obtaining system configuration information, and determining whether the business domain of the target business text needs to be combined with external information according to the system configuration information.

[0017] It can be seen that users can pre-set system configuration information to quickly determine whether the review of the target business text needs to be combined with external information, thereby improving the review efficiency of the business text.

[0018] A further solution is to put the graph variables into the rule engine and execute the audit rules to obtain the audit results, including: searching for the value of each graph variable in the complete knowledge graph through the graph depth search algorithm, putting the value of the graph variable into the rule engine and executing the audit rules.

[0019] It can be seen that applying the graph depth search algorithm to search the value of each graph variable and applying the value of the graph variable to review can quickly review business texts.

[0020] A further solution is that the logic complexity of the script audit rules is greater than the logic complexity of the ordinary audit rules.

[0021] A further solution is that the logic of ordinary audit rules includes: equals, contains; the logic of script audit rules includes: conditional branches, loops.

[0022] It can be seen from this that by dividing audit rules into different types, ordinary audit rules can be used to audit simple business texts, thereby improving the audit efficiency of business texts; and script audit rules can be used to audit complex business texts, which can improve the quality of business text audits.

[0023] To achieve the second objective, the present invention provides a computer device comprising a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, each step of the multi-business type text review method is implemented.

[0024] To achieve the third objective, the present invention provides a computer-readable storage medium storing a computer program, which implements the steps of the multi-business-type text review method when executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flow chart of establishing audit rules in an embodiment of the multi-business type text audit method of the present invention.

[0026] Figure 2It is a schematic diagram of the knowledge graph model in the embodiment of the multi-business type text review method of the present invention.

[0027] Figure 3 This is the first part of the flow chart of auditing target business texts in the embodiment of the multi-business type text auditing method of the present invention.

[0028] Figure 4 This is the second part of the flow chart of reviewing target business texts in the embodiment of the multi-business type text review method of the present invention.

[0029] The present invention will be further described below with reference to the accompanying drawings and embodiments. DETAILED DESCRIPTION

[0030] The multi-business document review method of the present invention is primarily used for automated review of various types of business documents and can be applied to computer devices such as desktop computers and laptop computers. The computer device provided by the present invention includes a processor and memory, and the processor can execute a computer program to implement the multi-business document review method described above.

[0031] Example of a method for reviewing texts of multiple business types:

[0032] This embodiment includes two major steps. The first major step is the system initialization process, that is, setting the audit rule model and setting the audit rules under each audit rule model to audit various types of business documents; the second major step is to obtain the business document to be audited and audit the business document to be audited using the audit rules set in the first major step.

[0033] During system initialization, see Figure 1 First, execute step S1 to set up knowledge graph models for various types of business texts. Since the business field involves a large number of different business areas, each with its own distinct business text types, and each type of business text requires different key extraction elements, dynamically supporting knowledge graph models for different business areas when building a knowledge graph model can improve system scalability, greatly facilitating the review of business texts.

[0034] Because there are many types of business documents, different types of business documents require different key points for review. Therefore, it is necessary to set a corresponding review rule model for each type of business document. To this end, it is necessary to set multiple types of business documents in step S1. Taking bidding as an example, business documents include three types: bidding documents, tender documents, and contract documents. In this embodiment, the entities and entity attributes are set for each type of business document. Figure 2In the figure, the ellipse represents the entity, the circle represents the entity attribute, and the connecting line represents the relationship between entities or the relationship between entities and entity attributes.

[0035] from Figure 2 It can be seen that entities include tender text, bidding text, project, contract text, etc., and the entity attributes of the tender text include document title, project name, bidding amount, bidding time, bidding company, etc. The entity attributes of the bidding text include document name, bidding company, bidding amount, project name, etc. The entity attributes of the contract text include document name, contract amount in uppercase, contract amount in lowercase, tax rate, tax amount, signing time, signing place, project name, Party A name, Party A contact person, Party A address, Party A legal representative, Party A bank account number, Party B name, Party B contact person, Party B address, Party B legal representative, Party B bank account number, etc. The information of the above entities and entity attributes can be set in the system in advance and formed as follows: Figure 2 The knowledge graph model shown.

[0036] Then, step S2 is executed to set the audit rule model corresponding to each type of business text based on different types of business texts. Figure 2 Taking the knowledge graph model as an example, it mainly includes three types of entities, namely bidding text, bid text and contract text. In step S2, the corresponding review rule model is set for the business text of each model.

[0037] When setting each audit rule model, step S3 needs to be executed to set the audit rules under each audit rule model. Taking the contract text as an example, an entity attribute of the merged text is the contract amount, and the rule of the contract amount needs to meet the condition of "contract amount == bid amount in the bid document". This condition is an audit rule.

[0038] In step S3, a graph model for the audit rules under each audit rule model needs to be set. Each graph model includes at least one graph variable, and each graph variable includes entities and entity elements. For example, the graph variable for the contract amount is represented as "contract@amount bid document bid amount", where the content before the first @ is the entity, and the content after the last @ is the entity element. Of course, the above graph variable has only one "@", which is called a jump relationship. A jump relationship is a relationship between two entities or between an entity and its attributes.

[0039] Of course, based on the characteristics of graphs, graph variables can also be expanded to have multiple jump relationships. For example, "Bidding documents@Bidding company@Executives does not include bidding documents@Bidding company@Executives". In this graph variable, there are multiple "@", and two adjacent "@"s are between two entities or entities and entity attributes. Among them, "Bidding documents@Bidding company@Executives" has two jump relationships, that is, "Bidding documents" is the entity, "@Bidding company" is the first jump relationship, and "@Executives" is the second jump relationship. Similarly, "Bidding documents@Bidding company@Executives" also contains two jump relationships, that is, "Bidding documents" is the entity, "@Bidding company" is the first jump relationship, and "@Executives" is the second jump relationship. The definition of the above review rules is to prohibit affiliated companies with close executives of the bidding company from bidding.

[0040] As can be seen, this embodiment leverages the strong correlation of graph variables to flexibly define multiple rules. Users can customize rules based on their actual needs. Furthermore, when setting a certain type of review rule, associated text can also be applied. For example, if the contract amount must be equal to the bid amount, this can be achieved by associating the contract text with the bid text, making the review of business documents more intelligent.

[0041] After setting the review rules, the specific business text will be reviewed, where the business text to be reviewed is the target business text. Figure 3 and Figure 4 When reviewing a target business document, step S11 is first performed to obtain the target business document to be reviewed. For example, a user uploads a target business document in the system. The target business document may be a contract, a bidding document, or a tender document.

[0042] Then, step S12 is executed to determine the type of the target business insurance document based on the business text classification model. The business text classification model is pre-set, for example, a model trained in advance using a machine learning algorithm based on the business domain to be processed. This model is used to classify the target business document and determine its type, i.e., whether it is a contract document, a bidding document, or a tender document.

[0043] Next, step S13 is executed to extract key elements of the target business text based on a unified information extraction framework. The unified information extraction framework is a machine learning-based text processing model that, through pre-annotated data and model training, can identify local information within business texts, such as the names of Party A and Party B in a contract, and the contract amount.

[0044] Then, step S14 is executed. Based on the key element information obtained in step S13 and the knowledge graph model of the business domain designed above, the key element information is constructed into a small knowledge graph according to the corresponding graph model. This embodiment uses this small knowledge graph as the basis for the review of the target business document.

[0045] Next, step S15 is executed to determine whether the target business document needs to be integrated with external information. Specifically, the system configuration information is read, the business domain of the target business document is determined, and a determination is made as to whether the rules corresponding to the business domain to which the target business document belongs require the small knowledge graph to be combined with external information for review. The system configuration information can be pre-set in the system. If the result of step S15 is yes, step S16 is executed; otherwise, step S17 is executed.

[0046] If the target business document is a tender document, then according to pre-set rules, when reviewing the target business document, it is necessary to integrate it with external information, and then step S16 is executed to obtain the executive information of the tendering company. For example, the executive information of the tendering company is obtained from a pre-set company executive information database.

[0047] Furthermore, based on the small knowledge graph obtained in step S14, information such as the names of Party A and Party B in the contract can be obtained, as well as information such as the bidding company and the bidding company in the tender documents. In step S16, the information in the tender documents and the bidding documents needs to be integrated with the information in the contract to obtain the connections between the executives of the bidding company and the bidding company, thereby obtaining a more complete knowledge graph. Then, step S17 is executed to determine the review rules for the target business document based on the complete knowledge graph.

[0048] If the result of step S15 is negative, step S17 is directly executed to obtain the review rules for the target business document. For example, based on the type of target business document determined in step S12, the review rule model for that type of business document set in step S2, and the review rules within the review rule model set in step S3, the review rules required for the target business document can be determined. Because the review rules for business documents can be manually adjusted by the user in the system configuration information, any changes to the review rules do not require system adjustments; the user only needs to adjust the corresponding review rules, allowing for flexible adaptation to different business document review requirements.

[0049] Then, step S18 is executed to determine whether the current target business document uses standard or scripted review rules based on the obtained review rules. In this embodiment, standard rules target relatively simple logic, such as simple logical calculations like "equal to" and "contains," while scripted rules target complex logic, such as "conditional branching" and "looping." Therefore, in step S18, the complexity of the obtained review rules is determined. If the complexity is low, standard review rules are used; if the complexity is high, scripted review rules are used.

[0050] If it is determined that the normal audit rules are to be used, step S19 is executed to obtain all graph variables for the normal audit rules. If it is determined that the script audit rules are to be used, step S20 is executed to obtain all graph variables for the script audit rules. In this embodiment, both script audit rules and normal audit rules are configured in step S3. In fact, each graph variable is a node or a path in the knowledge graph.

[0051] After reading all graph variables of the general audit rules or all graph variables of the script audit rules, execute step S21. Based on the meaning of the graph variables, use the graph depth search algorithm to search for the value of each graph variable in the small knowledge graph obtained in step S14 or the complete knowledge graph obtained in step S16. Then execute step S22, put the obtained graph variable value into the rule engine, and execute each audit rule to obtain the execution result of each audit rule, that is, execute step S23.

[0052] Finally, step S24 is executed, applying the result obtained in step S23 to determine the review result of the target business document. For example, if the result of step S23 is true, it means that the review rule is met; if the result is false, it means that the rule is not met. The target business document is considered to have passed the review only if it meets all the review rules.

[0053] It can be seen that, unlike the traditional text review for a single text type, this embodiment can perform associated review on the content of multiple types of related business texts, and the text review is not only based on the content of the target business document itself, but also based on the existing information system knowledge, and a comprehensive judgment is made by integrating the text content and the existing information system.

[0054] In addition, based on the definition of graph variables, this embodiment can flexibly define multi-related audit points. Combined with the advantages of knowledge graph path search, a code-free text audit rule function is proposed. In the face of complex and changeable audit points in business texts, the audit rules can be quickly and flexibly expanded.

[0055] Finally, the embodiment fuses multiple technologies, including text classification, natural language unified information extraction, knowledge graph construction, knowledge graph search, and graph rule engine, and through the combination of multiple technologies, the complex text auditing requirements can be met.

[0056] Computer device embodiment:

[0057] The computer device of the embodiment can be a desktop computer or a notebook computer, and can also be a server. The computer device has a processor, a memory, and a computer program stored in the memory and executable on the processor, such as an information processing program for implementing the above information processing method. When the processor executes the computer program, each step of the above multi-business type text auditing method is implemented.

[0058] For example, the computer program can be divided into one or more modules, one or more modules are stored in the memory and executed by the processor to complete each module of the application. One or more modules can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the terminal device.

[0059] Those skilled in the art can understand that the schematic diagram of the present application is only an example of the terminal device and does not constitute a limitation on the terminal device, which can include more or less components than the diagram, or combine certain components, or different components, for example, the terminal device can also include an input / output device, a network access device, a bus, etc.

[0060] The processor of the present application can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), ready programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The processor is the control center of the terminal device, which connects all parts of the terminal device through various interfaces and lines.

[0061] The memory can be used to store computer programs and / or modules. The processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and accessing the data stored in the memory. The memory can mainly include a program storage area and a data storage area. The program storage area can store the operating system, at least one application required for a function, etc.; the data storage area can store data created based on the use of the mobile phone, etc. In addition, the memory can include high-speed random access memory and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card (FlashCard), at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0062] Computer readable storage medium:

[0063] If a computer program stored in a computer device is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention can also implement all or part of the process of the above-mentioned embodiment method by instructing the relevant hardware through a computer program. This computer program can be stored in a computer-readable storage medium. When executed by a processor, this computer program can implement each step of the above-mentioned multi-business document review method.

[0064] Among them, computer programs include computer program code, which can be in source code form, object code form, executable file, or some intermediate form. Computer-readable media may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, mobile hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signals, telecommunications signals, and software distribution media. It should be noted that the content contained in computer-readable media can be appropriately increased or decreased based on the requirements of legislation and patent practice within a jurisdiction. For example, in some jurisdictions, based on legislation and patent practice, computer-readable media does not include electric carrier signals and telecommunications signals.

[0065] Finally, it should be emphasized that the present invention is not limited to the above-mentioned embodiments. For example, changes in the audit rule settings or changes in the classification of business text types should also be included in the scope of protection of the claims of the present invention.

Claims

1. A multi-business type text review method, characterized in that: include: Setting up knowledge graph models for various business texts, setting up corresponding review rule models based on each type of business text, and determining review rules under each review rule model; Obtain a target business text to be reviewed, apply a text classification model to determine the text type of the target business text, extract key element information of the target business text, and construct a small knowledge graph based on the key element information according to the knowledge graph model; Determine whether the business field to which the target business text belongs requires the integration of external information. If so, fuse the constructed small knowledge graph with the external information to obtain a complete knowledge graph and determine the review rules for the target business text. Otherwise, directly obtain the review rules corresponding to the target business text. Determine, based on the review rule of the business document, whether the review rule is a common review rule or a script review rule. If it is a common review rule, read all graph variables of the common review rule. If it is a script review rule, read all graph variables of the script review rule, put the graph variables into the rule engine, execute the review rule, and obtain the review result. Determining whether the business field described in the target business text needs to be combined with external information includes: obtaining system configuration information, and determining whether the business field described in the target business text needs to be combined with external information according to the system configuration information; The step of obtaining an audit result after the graph variable is placed in the rule engine and the audit rule is executed includes: searching for the value of each graph variable in the complete knowledge graph through a graph depth search algorithm, placing the value of the graph variable in the rule engine and executing the audit rule; The logic complexity of the script audit rule is greater than the logic complexity of the common audit rule.

2. The multi-business type text review method according to claim 1, characterized in that: Determining the audit rules under each of the audit rule models includes: setting a graph model of the audit rules under each of the audit rule models, each of the graph models including at least one graph variable, and each of the graph variables including an entity and an entity element.

3. The multi-business type text review method according to claim 2, characterized in that: The graph variable includes at least one jump relationship, and the entity and the entity element are objects having a jump relationship.

4. The multi-business type text review method according to claim 3, characterized in that: The graph variable includes two or more jump relationships; The number of entities included in the graph variable is two or more, and there is a jump relationship between two adjacent entities.

5. The multi-business type text review method according to claim 1, characterized in that: The logic of the general audit rules includes: equal to, contains; The logic of the script review rules includes: conditional branches and loops.

6. A computer device, characterized in that The method comprises a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the method implements the steps of the multi-business type text review method according to any one of claims 1 to 5.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the multi-business type text review method according to any one of claims 1 to 5 are implemented.

Images