Financial RPA rule generation method and device based on process template, equipment and storage medium

Abstract

The invention provides a financial RPA rule generation method and device based on a process template, equipment and a storage medium, and relates to the technical field of computers. The method comprises the following steps: constructing task data in a financial scene based on a predetermined process template; classifying the task data into a task logic extraction task and an action logic extraction task, analyzing task content by utilizing a text analysis model, and outputting an analysis result; carrying out secondary adjustment on the analysis result, and carrying out iterative optimization on the text analysis model by utilizing the analysis result subjected to secondary adjustment; extracting a financial RPA rule from the analysis result after the secondary adjustment, and constructing the financial RPA rule into a workflow capable of being automatically executed; and subsequently, directly calling the workflow to complete the financial RPA application. According to the method, on the basis of a traditional rule base, the capabilities of semantic comprehension, rule analysis, splicing and the like are added in combination with actual financial business requirements, and rapid storage and application of financial auditing requirements are achieved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method, device, equipment and storage medium for generating financial RPA rules based on a process template. Background Art

[0002] With the development of digitalization in all walks of life, the workload that users need to invest in the management system is also increasing. Among them, the proportion of work with clear rules, high repetitiveness and large business volume continues to rise. In theory, this type of work can be completely replaced by the management system itself through its own upgrades and iterations. However, the development cost of management system updates and iterations is high, and the implementation cycle is long. It is unable to keep up with the frequency and speed of policy adjustments and changes, and the cost performance is extremely low. If cross-system work is involved, the upgrade cost will be even higher. At the same time, it takes a lot of effort to convert business rules into easy-to-implement code logic. There is currently no complete implementation process. Therefore, under the premise of non-invasive management system, the system can be automated and the demand for process automation logic can be realized in a short time according to dynamic changes. Summary of the invention

[0003] Purpose of the invention: To propose a financial RPA rule generation method, device, equipment and storage medium based on process templates, which is suitable for process automation based on business rules in financial scenarios, and is used to solve the rapid conversion of business rules to process automation, improve the flexibility of process automation, and solve the above-mentioned problems existing in the prior art.

[0004] In a first aspect of the present invention, a method for generating financial RPA rules based on a process template is proposed, comprising the following steps:

[0005] Build task data in financial scenarios based on a predetermined process template;

[0006] Classifying the task data into task logic extraction tasks and action logic extraction tasks, parsing the task content using a text parsing model, and outputting parsing results;

[0007] Performing a secondary adjustment on the parsing result, and iteratively optimizing the text parsing model using the parsing result after the secondary adjustment;

[0008] The financial RPA rules are extracted from the parsing results after the second adjustment, and the financial RPA rules are constructed into an automatically executable workflow; the workflow is subsequently directly called to complete the financial RPA application.

[0009] In a further embodiment of the first aspect, task data T in a financial scenario is constructed based on a process template:

[0010] T=[t desc , tother ]

[0011] Where, t desc Indicates the requirement description; other Indicates other descriptions;

[0012] Requirement Description desc The expression is as follows:

[0013] t desc =∪x i

[0014] In the formula, x i It means to directly describe the task content, including the function and purpose to be achieved by the task;

[0015] Other descriptions other The expression is as follows:

[0016] t other =∪z i

[0017] In the formula, z i Represents other descriptions, including auxiliary information such as task scope, task classification, and task tags.

[0018] In a further embodiment of the first aspect, classifying and parsing tasks and performing content analysis based on the task data specifically includes:

[0019] The task data T in the financial scenario is classified into task logic extraction tasks T rule and action logic extraction task T action ;

[0020] Use the text parsing model to parse the task logic and extract the task T rule and action logic extraction task T action , output the analysis result R rule With R action .

[0021] In a further embodiment of the first aspect, the task logic extracts the task T rule The definition is as follows:

[0022] T rule =(t target , t judge , t type )

[0023] Where, t target is the audit field; judge is the judgment method; type To judge the type;

[0024] ttype For different types, rule Perform corresponding expansion operations;

[0025] t judge For different values ​​of t rule Perform corresponding expansion operations.

[0026] In a further embodiment of the first aspect, t type For different types, rule Perform corresponding expansion operations, including:

[0027] When t type When it is "text", "number" or "field value", the "judgment condition" field is expanded and the field value is identified and supplemented;

[0028] When t type When it is "Date", expand the "Date" field to identify and supplement the calendar date;

[0029] When t type For "weekly", expand the field recognition supplement from Monday to Sunday;

[0030] When t type For "monthly", expand the field identification supplement of 1 to 31;

[0031] When t type For "Every Year", expand the fields to identify the months from January to December;

[0032] When t type When the date is "Current Date", "Current Month", "Current Year", "Last Year", "Recent Week", "Recent Month", "Recent Year", "Recent 2 Years" or "Recent 3 Years", no new fields are added.

[0033] In a further embodiment of the first aspect, t judge For different values ​​of t rule Perform corresponding expansion operations, including:

[0034] t judge When it is "Interval", the "Audit Field", "Interval Quantity" and "Interval Unit" fields are expanded to identify and supplement;

[0035] t judge When it is "contains", "contained in", "does not contain", "is not equal to", "greater than", "greater than or equal to", "equal to", "less than", "less than", the new field is not expanded;

[0036] t judge When the value is "Empty", "True" or "False", delete the "Judgement Type" field.

[0037] In a further embodiment of the first aspect, the action logic extraction task T action The definition is as follows:

[0038] T action =(t target , t action )

[0039] Where, t target is the audit field; action For operation actions;

[0040] t action When T action Perform the corresponding expansion operations:

[0041] t action When the action is "click", "hover", "check", or "right-click", the new field will not be expanded;

[0042] t action Expand the “target content” identification and supplement when “modify”, “add” and “select”.

[0043] In a further embodiment of the first aspect, the task logic extraction task T is parsed using a text parsing model rule and action logic extraction task T action , output the analysis result R rule With R action , specifically including:

[0044] Using the pre-trained model BERT as the text parsing model, the extraction task f(Q, K, V) of the text parsing model is defined as follows:

[0045]

[0046] Where, the query matrix of T is Q; rule The attention matrix is ​​K; T action The attention matrix is ​​V; QK T The query matrix is ​​multiplied by the attention matrix to obtain the attention weight value of Q on V. The output result is smoothed by softmax and limited to between 0 and 1.

[0047] Multi-task extraction multi (Q, K, V) are defined as follows:

[0048] f multi (Q, K, V) = ∪ i f(Q, K, V)

[0049] By splicing subtasks to achieve multi-task synchronous execution, the extraction result R is output ruleWith R action .

[0050] In a further embodiment of the first aspect, the analysis result R rule With R action Secondary adjustments are made, including:

[0051] Using word frequency to analyze R rule Perform a check:

[0052]

[0053] In the formula, Based on t type The effective content of the filter; is a binary classification function, with t judge Select the judgment content as the basis, and judge R rule Does it have a content that satisfies t target The effective content of the audit request;

[0054] like If the result is 0, it means that the currently extracted audit information and audit rules cannot be adapted, and a request is sent for secondary generation of the BERT model;

[0055] Using t action The specific business operations reflected in R action To perform a compliance check:

[0056]

[0057] In the formula, I is the compliance verification function which is a binary classification function; sort(*) is the operation order extraction function; if the result I is 0, it means that the current business behavior and the required business behavior cannot be adapted, and a request is sent for secondary generation of the BERT model.

[0058] In a further embodiment of the first aspect, a financial RPA rule Rule is extracted from the parsing result after secondary adjustment:

[0059] Rule={M t_group , M condition , M h_group}

[0060] Where M t_group Represents the horizontal combination of regular splicing units; M condition Indicates the connection condition; M h_group It represents the vertical combination of regular splicing units;

[0061] Regular splicing unit horizontal combination M t_groupThe definition is as follows: a list consisting of multiple rules of the same level, each rule of the same level is spliced ​​through the variable logic of "and" and "or", and the final result obtained by reviewing each rule of the same level is then obtained through the variable logic of the same level to obtain the final review result of the list of rules of the same level;

[0062] Connection Condition M condition including or operating, and operating;

[0063] Regular splicing unit vertical combination M h_group It is defined as a tree list of child rules: each rule contains 1 child rule, and the child rule contains its own equal-level rules and child rules, and the child rules are infinitely nested; the review result of the rule and its own child rules are spliced ​​through the variable logic of "and" and "or", and the splicing order of the review result starts from the leaf node and continues to recursively splice toward the root. The final result obtained by the review of each parent-child rule is then obtained through the variable logic to obtain the final review result of the rule tree list.

[0064] A second aspect of the present invention provides a financial RPA rule generation device, which includes a task definition module, a rule base module, and a rule parsing module.

[0065] The task definition module builds task data in financial scenarios based on a predetermined process template.

[0066] The rule base module is used to classify the task data into task logic extraction tasks and action logic extraction tasks, parse the task content using a text parsing model, and output the parsing results.

[0067] The rule parsing module is used to make a secondary adjustment to the parsing results, and iteratively optimize the text parsing model using the parsing results after the secondary adjustment; extract the financial RPA rules from the parsing results after the secondary adjustment, and construct the financial RPA rules into an automatically executable workflow; and subsequently directly call the workflow to complete the financial RPA application.

[0068] A third aspect of the present invention proposes an electronic device, comprising: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the method for generating financial RPA rules based on process templates as described in the first aspect is implemented.

[0069] In a fourth aspect of the present invention, a computer-readable storage medium is proposed, which stores at least one executable instruction. When the executable instruction is executed on an electronic device, the electronic device executes the financial RPA rule generation method based on the process template as described in the first aspect.

[0070] Beneficial effects:

[0071] (1) The present invention designs a business logic requirement template for financial scenarios to solve the problems of unclear business logic description and difficulty in semantic understanding, thereby improving the accuracy of rule parsing.

[0072] (2) The present invention designs a process automation solution for the financial field to achieve rapid response to demand / business changes - a financial business rule library.

[0073] (3) Based on the traditional rule base, the present invention combines the actual financial business needs to increase the capabilities of semantic understanding, rule parsing, and splicing, so as to achieve rapid storage and application of financial audit requirements.

[0074] (4) The present invention combines the rule base with process automation technology to achieve accurate conversion between financial logic language and process automation language, thereby improving the design efficiency of process automation scenarios in the financial field. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 It is an overall architecture diagram of the financial RPA rule generation device in an embodiment of the present invention.

[0076] Figure 2 It is a flowchart of a financial RPA rule generation method based on a process template in an embodiment of the present invention.

[0077] Figure 3 It is a schematic diagram of regular splicing sub-units, a horizontal combination of regular splicing units, and a vertical combination of regular splicing units in an embodiment of the present invention. DETAILED DESCRIPTION

[0078] In the following description, a large number of specific details are provided to provide a more thorough understanding of the present invention. However, it is apparent to those skilled in the art that the present invention can be implemented without one or more of these details. In other examples, in order to avoid confusion with the present invention, some technical features known in the art are not described.

[0079] Figure 1 This is the overall architecture of the financial RPA rule generation device, which includes three parts: task definition module, rule base module, and rule parsing module.

[0080] Task definition module: This module is used to standardize the way of providing requirements and business rules in the financial field; sort out the business logic sequence by designing process templates to reduce the threshold of semantic understanding; apply this module to clearly divide the business rules in various scenarios within finance to facilitate subsequent task analysis and other operations.

[0081] Rule base module: This module is mainly used to generate, adjust and store financial business rules according to the content of task definition. By analyzing the task data, it can automatically generate the key points of the rules that meet the requirements of the rule base according to the template and semantics and make structured records. After manual secondary confirmation and adjustment, the generation model is optimized, and finally the full process of rule automatic tuning is realized.

[0082] Rule parsing module: This module is used to convert the rules in the financial rule base into process automation language to quickly adapt to the needs of adding, modifying, and deleting financial rules.

[0083] The process of generating, transforming and applying financial RPA rules based on process templates is shown in Figure 2 , the following are the detailed steps of the technical solution:

[0084] Step S1: Build task data in financial scenarios based on process templates

[0085] Step S1-1: Define the financial requirement rule template T as follows:

[0086] T=[t desc , t other ]

[0087] where t desc is the requirement description, t other For other descriptions

[0088] Step S1-1-1: Demand description desc Definition, directly describe the task content, including the function and purpose to be achieved by the task.

[0089] t desc =∪x i

[0090] Step S1-1-2: Other descriptions other Definition, provide auxiliary information for the task. The auxiliary information mainly includes task scope limitation, task classification, and task tag, which is used to expand the rules in the secondary adjustment stage.

[0091] t other =∪z i

[0092] Step S2: Classify and analyze tasks based on task data and perform content analysis

[0093] Step S2-1: Perform task classification on the input T constructed in step S1, i.e., extract task T from task logic rule and action logic extraction task T action .

[0094] Step S2-1-1: Task logic extracts task T rule

[0095] Step S2-1-1-1: T rule The definition is as follows:

[0096] T rule =(t target , t judge , t type )

[0097] where t target is the audit field, t judge is the judgment method, t type To judge the type.

[0098] Step S2-1-1-2: type When the type is different, T rule Perform expansion operations.

[0099] Step S2-1-1-2-1: When t type When it is "Text", "Number" or "Field Value", the "Judgment Condition" field will be expanded and the field value will be identified and supplemented.

[0100] Step S2-1-1-2-2: When t type When it is "Date", the "Date" field will be expanded to identify and supplement the calendar date.

[0101] Step S2-1-1-2-3: When t type For "weekly", the field recognition supplement for Monday to Sunday will be expanded.

[0102] Step S2-1-1-2-4: When t type For "monthly", the fields for identification of dates 1 to 31 will be expanded.

[0103] Step S2-1-1-2-5: When t type For "Every Year", additional fields will be added to identify the months from January to December.

[0104] Step S2-1-1-2-6: When t type For "Current Date", "Current Month", "Current Year", "Last Year", "Recent Week", "Recent Month", "Recent Year", "Recent 2 Years" and "Recent 3 Years", no new fields will be expanded.

[0105] Step S2-1-1-3: t judge When t is different, rule Perform expansion operations.

[0106] Step S2-1-1-3-1: t judgeWhen it is "Interval", the "Audit Field", "Interval Quantity" and "Interval Unit" fields will be expanded for identification supplement.

[0107] Step S2-1-1-3-2: t judge When the value is "contains", "is contained in", "does not contain", "is not equal to", "is greater than", "is greater than or equal to", "is equal to", "is less than", or "is less than or equal to", no new field will be expanded.

[0108] Step S2-1-1-3-3: t judge When the value is "Empty", "True", or "False", the "Judgement Type" field will be deleted.

[0109] Step S2-1-2: Action logic extraction task T action

[0110] Step S2-1-2-1: T action The definition is as follows:

[0111] T action =(t target , t action )

[0112] where t target is the audit field, t action For operation actions.

[0113] Step S2-1-2-2: t action When it is a different value, T action Perform expansion operations.

[0114] Step S2-1-2-2-1: t action When the action is "click", "hover", "check", "right-click", etc., the new field will not be expanded.

[0115] Step S2-1-2-2-2: t action When "edit", "add", "select", etc., the "target content" will be expanded to identify and supplement.

[0116] Step S2-2: Apply the text parsing model to implement the task logic extraction task T rule and action logic extraction task T action

[0117] Step S2-2-1: Implement the extraction task based on the pre-trained model BERT

[0118] Step S2-2-1-1: The extraction task is defined as follows:

[0119]

[0120] The query matrix of T is Q, Trule The attention matrix is ​​K, T action The attention matrix is ​​V, QK T The query matrix is ​​multiplied by the attention matrix to get the attention weight value of Q on V. The output result is smoothed by softmax and limited to between 0 and 1.

[0121] Step S2-2-1-2: Multi-task extraction is defined as follows:

[0122] f multi (Q, K, V) = ∪ i f(Q, K, V)

[0123] Multi-task synchronous execution is achieved by splicing subtasks.

[0124] Step S2-2-2: Output the extraction result R rule With R action

[0125] Step S3: Extract the result R of step S2-2-2 rule With R action Make secondary adjustments and iteratively optimize the text parsing model

[0126] Step S3-1: Using word frequency and t type and t judge The business situation for R rule Perform a check

[0127]

[0128] in Based on t in step S2-I-1-2 type The effective content of the filtering is mainly through t type Supplementary information and business rules are used for regular matching. is a binary classification function, whose main function is to use t judge Select the judgment content as the basis, and judge R rule Does it have a content that satisfies t target The effective content of the audit request. If the result is 0, it means that the currently extracted audit information and audit rules cannot be adapted, and a request is sent for secondary generation of the BERT model.

[0129] Step S3-2: Using t action The specific business operations reflected in R action To perform compliance verification, the specific steps are as follows:

[0130]

[0131] Among them, I is the compliance verification function, which is a binary classification function. Its specific function is that if R action The extracted valid content can be obtained by inputting information T and preset operation set t action In the action The logical operations extracted in are consistent with the operation order described by T, and sort(.) is the operation order extraction function. If the result of I is 0, it means that the current business behavior and the required business behavior cannot be adapted, and a request is sent for secondary generation of the BERT model.

[0132] Step S3-3: Manual intervention, rule verification and secondary adjustment of the output of step S2 according to the process template to obtain R′ rule and R′ action .

[0133] Step S3-3-1: Modify and confirm the “Business Audit Phase”.

[0134] Step S3-3-2: Modify and confirm the “Business Scope of Use”.

[0135] Step S3-3-3: Modify and confirm “Business Applicability Objects”.

[0136] Step S3-3-4: Modify and confirm the “Business Applicable Scenarios”.

[0137] Step S3-3-5: Modify and confirm the “business transfer logic”.

[0138] Step S3-4: Adjust the result R' rule and R′ action Perform model tuning.

[0139] Step S3-4-1: Repeat step S2-2 with the adjusted result

[0140] Step S3-4-2: The evaluation criteria are as follows:

[0141]

[0142] Among them, TP, TN, FP, and FN are divided into true positive examples, true negative examples, false positive examples, and false negative examples of predicted results and actual results.

[0143] Step S3-4-3: When Acc reaches the preset requirement Acc r Stop the tuning training.

[0144] Step S4: Analyze the rules to apply to the process automation link to complete the rule application

[0145] Step S4-1: Data splicing to complete the combination of business and logic, including rule splicing subunit Munit , Regular splicing unit horizontal combination M t_group , Regular splicing unit vertical combination M h_group And the connection condition M condition .

[0146] Step S4-1-1: M unit The definition is as follows: The sub-unit rule is the simplest rule that can be split, and is composed of basic fields, such as "contract amount is greater than 10,000 yuan", "reimbursement period is equal to this month", "remarks include meal expenses", etc. This rule is the simplest judgment unit; there are peer relationships and sub-level relationships between rules.

[0147] Step S4-1-2: M t_group The definition is as follows: a list consisting of multiple rules of the same level. The rules of the same level can be spliced ​​through the variable logic of "and" and "or". The final result obtained by reviewing each rule of the same level is then passed through the variable logic of the same level to obtain the final review result of the list of rules of the same level.

[0148] Step S4-1-3: M h_group The definition is as follows: a tree list of sub-level rules. Each rule can contain 1 sub-level rule, and a sub-level rule can contain its own equal-level rules and sub-level rules. In theory, sub-level rules can be nested infinitely. The audit results of a rule and its own sub-level rules can be spliced ​​through the variable logic of "and" and "or". The splicing order of the audit results starts from the leaf node and continues to recursively splice toward the root. The final result obtained by the audit of each parent-child rule is then passed through the variable logic to obtain the final audit result of the rule tree list.

[0149] Step S4-1-4: M condition Including OR operation and AND operation

[0150] Step S4-1-4-1: OR operation, that is, if there is a rule that is True, the final result is True. If all rules are False, the final result is False.

[0151] Step S4-1-4-2: AND operation, that is, if there is a rule that is False, the final result is False. If all rules are True, the final result is True.

[0152] Step S4-2: According to M t_group 、M h_group and M condition Conduct rule analysis and review.

[0153] Step S4-2-1: Rule review order: The final review result of a task, for each rule, is determined by first determining the review result of the rule combined with its child rules, and then combined with the review results of other rules at the same level.

[0154] Step S4-2-2: Verify header rules

[0155] Step S4-2-2-1: Recursively check the header rules M h_group , until the last level

[0156] Step S4-2-3: Continue to verify M t_group The next order rule M unit

[0157] Step S4-2-3-1: If you encounter M unit is False, and M condition If it is "and", the review will stop and the result will be returned as False directly.

[0158] Step S4-2-3-2: If you encounter M unit is True, and M condition If it is "or", the review will stop and the result will be returned as True directly.

[0159] Step S4-2-3-3: If you encounter M unit is False, and M condition If it is "or", continue.

[0160] Step S4-2-3-4: If you encounter M unit is True, and M condition If it is "and", continue.

[0161] Step S4-3: Convert the splicing rule combination into process automation language to complete process establishment or update.

[0162] Step S4-3-1: Combination of splicing rules

[0163] Rule={M t_group , M condition , M h_group}

[0164] Step S4-3-2: The process automation conversion module uses parameters and actions as configurable parameters and assigns parameters to the rule subset in the splicing rule combination Rule.

[0165] Step S4-3-2: Update the rules to the automated process to complete the process update action.

[0166] The present invention can be established using a financial rule library. The following only details the implementation of the business logic and action logic of the financial rule library:

[0167] Step S1: Based on the process template, construct task data in the financial scenario, referring to the task data of the process template, where t desc is the "rule description", and t other is the "rule name", "financial link", "province in use", "rule type", "economic event", "business scenario", "processing method when the rule passes", "processing method when the rule fails".

[0168] Step S2: Classify and parse the tasks based on the task data and perform content parsing

[0169] Step S2-1: The input T is described as follows: If "whether image scanning is involved" is "yes", transfer to manual operation.

[0170] Step S2-1-1: Classify as task logic to extract task T rule .

[0171] Step S2-1-2: Obtain the result T according to the model rule =(t target , t judge , t type ), where t target is "whether image scanning is involved", t judge is "is", and t type is "yes"

[0172] Step S2-1-3: t type is "yes"; "yes" is the judgment condition, and the judgment type is "text".

[0173] Step S2-1-4: t iudge is "is"; "is" is the judgment method and is matched as "equal to".

[0174] Step S2-1-5: This financial rule is T rule =("whether image scanning is involved", "equal to", "yes")

[0175] Step S2-2: The input T is described as follows: If the description of the basic information is empty, add 1, and for the basic information, whether it is urgent is no, whether it is split on the credit side is no, and whether it is a payment is yes.

[0176] Step S2-1-1: Classify as action logic to extract task T action .

[0177] Step S2-1-2: Obtain the result T according to the model action =(t target , taction ), where t target For basic information, whether to expedite, whether to split the credit, whether to pay, t action To add 1. Click.

[0178] Step S2-1-3: The above task can be divided into four action logics, namely (basic information, add 1), (whether to expedite, click no), (whether to split the creditor, click no), (whether to pay, click yes)

[0179] Step S3: Rule verification and secondary adjustment

[0180] Step S3-1: Due to the basic information, whether to expedite, whether to split the creditor is t type The type is "text" and t judge When the match is equal, consistent corresponding information can be found from the input information T, and the business rules are reasonable, so the Bert model extraction is correct.

[0181] Step S3-2: Since 1 is added, click No, click No, click Yes for T action The business actions are valid and the logic is correct in the input information T, so the Bert model is extracted correctly.

[0182] Step S3-3: Adjust the content of S2 output after manual verification

[0183] Step S3-3: Task logic extracts task T rule Secondary Adjustment

[0184] Step S3-3-1: For auxiliary information t other , adjust the "Financial link" to "Preliminary review", the "Using province" to "Gansu", the "Rule type" to "Non-general", the "Economic matter" to "Salary posting", the "Business scenario" to "Unlimited", the "Rule passing processing method" to "None", and the "Rule not passing processing method" to "Transfer".

[0185] Step S3-4: Action logic extraction task T action Secondary Adjustment

[0186] Step S3-4-1: For auxiliary information t other , extract "Financial link" as "Preliminary review", "Using province" as "Jiangsu", "Rule type" as "Non-general", "Economic matter" as "Salary posting", "Business scenario" as "Unlimited", "Rule passing processing method" as "None", and "Rule not passing processing method" as "None".

[0187] Step S4: Analyze the rules to apply to the process automation link to complete the rule application

[0188] Step S4-1: Regular splicing of subunits M unit For ("Is it related to image scanning", "Equal to", "Yes"), (Basic information, add 1), (Is it expedited, click No), (Is it split credit, click No), (Is it paid, click Yes)...

[0189] Step S4-2: Splicing unit horizontal combination M t_group ,{("Is it related to image scanning", "equal to", "yes"),(basic information, add 1)}

[0190] Step S4-3: Regular splicing unit vertical combination M h_group , {(basic information, add 1), (whether to expedite, click no)}

[0191] Step S4-4: The financial rule combination is as follows, namely, {("Is image scanning involved?", "Equal to", "Yes") or (Basic information, add 1)}, {(Basic information, add 1) and (Basic information, add 1) and (Is it expedited, click No) and (Is it split credit, click No) and (Is it paid, click Yes)}

[0192] Step S4-5: Update the rules to the automated process to complete the process update action.

[0193] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0194] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0195] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the present invention itself. Various changes may be made to it in form and detail without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A financial RPA rule generation method based on process template, characterized in that: The steps include: Build task data in financial scenarios based on a predetermined process template; Classifying the task data into task logic extraction tasks and action logic extraction tasks, parsing the task content using a text parsing model, and outputting parsing results; Performing a secondary adjustment on the parsing result, and iteratively optimizing the text parsing model using the parsing result after the secondary adjustment; The financial RPA rules are extracted from the parsing results after the second adjustment, and the financial RPA rules are constructed into an automatically executable workflow; the workflow is subsequently directly called to complete the financial RPA application.

2. The method for generating financial RPA rules based on process templates according to claim 1 is characterized in that: Based on the process template, construct the task data T in the financial scenario: T=[t desc ,t other ] In the formula, t desc Indicates the requirement description; other Indicates other descriptions; Requirement Description desc The expression is as follows: t desc =∪x i In the formula, x i It means to directly describe the task content, including the function and purpose to be achieved by the task; Other descriptions other The expression is as follows: t other =∪z i In the formula, z i Represents other descriptions, including auxiliary information such as task scope, task classification, and task tags.

3. The method for generating financial RPA rules based on process templates according to claim 2 is characterized in that: Based on the task data, the tasks are classified and analyzed and content analysis is performed, specifically including: The task data T in the financial scenario is classified into task logic extraction tasks T rule and action logic extraction task T action ; Use the text parsing model to parse the task logic and extract the task T rule and action logic extraction task T action , output the analysis result R rule With R action ; The task logic extracts the task T rule The definition is as follows: T rule =(t target ,t judge ,t type ) In the formula, t target is the audit field; judge is the judgment method; type To judge the type; t type For different types, rule Perform corresponding expansion operations; t judge For different values ​​of t rule Perform corresponding expansion operations.

4. The method for generating financial RPA rules based on process templates according to claim 3 is characterized in that: The action logic extraction task T action The definition is as follows: T action =(t target ,t action ) In the formula, t target is the audit field; action For operation actions; t action When T action Perform the corresponding expansion operations: t action When it is "click", "hover", "check", or "right-click", the new field will not be expanded; t action Expand the "target content" identification and supplement when "modify", "add" and "select".

5. The method for generating financial RPA rules based on process templates according to claim 3 is characterized in that: Use the text parsing model to parse the task logic and extract the task T rule and action logic extraction task T action , output the analysis result R rule With R action , specifically including: Using the pre-trained model BERT as the text parsing model, the extraction task f(Q, K, V) of the text parsing model is defined as follows: Where, the query matrix of T is Q; rule The attention matrix is ​​K; T action The attention matrix is ​​V; QK T The query matrix is ​​multiplied by the attention matrix to obtain the attention weight value of Q on V. The output result is smoothed by softmax and limited to between 0 and 1. Multi-task extraction multi (Q, K, V) are defined as follows: f multi (Q,K,V)=∪ i f(Q,K,V) By splicing subtasks to achieve multi-task synchronous execution, the extraction result R is output rule With R action .

6. The method for generating financial RPA rules based on process templates according to claim 5 is characterized in that: The analytical result R rule With R action Secondary adjustments are made, including: Using word frequency to analyze R rule Perform a check: In the formula, Based on t type The effective content of the filter; is a binary classification function, with t judge Select the judgment content as the basis, and judge R rule Does it have a content that satisfies t target The effective content of the audit request; like If the result is 0, it means that the currently extracted audit information and audit rules cannot be adapted, and a request is sent for secondary generation of the BERT model; Using t action The specific business operations reflected in R action To perform a compliance check: In the formula, I is the compliance verification function which is a binary classification function; sort(*) is the operation order extraction function; if the result I is 0, it means that the current business behavior and the required business behavior cannot be adapted, and a request is sent for secondary generation of the BERT model.

7. The method for generating financial RPA rules based on process templates according to claim 1 is characterized in that: Extract the financial RPA rule from the parsing result after secondary adjustment: Rule={M t_group ,M condition ,M h_group } Where M t_group Represents the horizontal combination of regular splicing units; M condition Indicates the connection condition; M h_group It represents the vertical combination of regular splicing units; Regular splicing unit horizontal combination M t_group The definition is as follows: a list consisting of multiple rules of the same level, each of which is connected through the variable logic of "and" and "or". The final result of each rule of the same level is obtained through the review and then the final review result of the list of rules of the same level is obtained through the variable logic of the same level. Connection Condition M condition including or operating, and operating; Regular splicing unit vertical combination M h_group It is defined as a tree list of child rules: each rule contains 1 child rule, and the child rule contains its own equal-level rules and child rules, and the child rules are infinitely nested; the review result of the rule and its own child rules are spliced ​​through the variable logic of "and" and "or", and the splicing order of the review result starts from the leaf node and continues to recursively splice toward the root. The final result obtained by the review of each parent-child rule is then obtained through the variable logic to obtain the final review result of the rule tree list.

8. A financial RPA rule generation device, characterized in that: include: Task definition module; The task definition module constructs task data in a financial scenario based on a predetermined process template; Rule base module; The rule base module is used to classify the task data into task logic extraction tasks and action logic extraction tasks, parse the task content using a text parsing model, and output the parsing results; Rule parsing module; the rule parsing module is used to make a secondary adjustment to the parsing results, and iteratively optimize the text parsing model using the parsing results after the secondary adjustment; extract the financial RPA rules from the parsing results after the secondary adjustment, and construct the financial RPA rules into an automatically executable workflow; and subsequently directly call the workflow to complete the financial RPA application.

9. An electronic device, characterized in that: The device includes: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the method for generating financial RPA rules based on process templates as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The storage medium stores at least one executable instruction. When the executable instruction is executed on the electronic device, the electronic device executes the financial RPA rule generation method based on the process template as described in any one of claims 1 to 7.

Images