Business process automatic configuration method and device based on large language model

Abstract

The invention discloses a business process automatic configuration method and device based on a large language model, and the method comprises the steps: obtaining a user demand text, and obtaining a context constraint corresponding to the user demand text according to a business scene; inputting the user demand text and the context constraint into a first large language model to obtain preliminary standard data; analyzing the preliminary standard data to obtain a business process, and judging whether the business process meets an expected business requirement, if not, obtaining dialogue content data, and performing cyclic dialogue on the first large language model according to the dialogue content data and the preliminary standard data to obtain a second large language model; and the obtained business process meets the expected business requirement. The business requirements can be fully understood by using the strong natural language understanding capability of the large language model, and the time of requirement arrangement and business abstraction is saved, so that the learning cost and the technical threshold of process configuration personnel are reduced.

Description

Technical Field

[0001] The present invention relates to the field of Internet of Things application technology, and in particular to a business process automation configuration method and device based on a large language model. Background Art

[0002] At present, business process automation is usually implemented based on current process engines, and Business Process Modeling Notation (BPMN) is generally used to store and represent business processes. Users configure various complex processes through the visual interface provided by various process engine manufacturers and then convert them into BPMN standard data for storage. For example, to implement a relatively complex business process configuration, it is generally necessary to go through the steps of demand collection, business combing, process node combing, process node serial combing, and process visualization configuration to finally complete the business process configuration.

[0003] However, with the development of business, the complexity of business processes is increasing. The problem that follows is that the BPMN standard is becoming more and more complex, and the visual process configuration engine presented to the owner is becoming more and more complex, which leads to an increase in the threshold for general users to use it. For example, in the process of implementing a complex business process, the configuration personnel need to first summarize the accurate business process from the customer's description, and perform structured conversion to form concrete process nodes and processes, and this process also places high demands on the configuration personnel. In addition, in scenarios with a large number of process nodes, each node needs to be configured with different business processing logic, and in the process of configuring business processes with complex branches, even if a powerful visual configuration interface is provided, the configuration personnel need to spend a lot of time to configure the process nodes and process branches, and also need to spend a lot of time to make necessary notes. Before the configuration is completed, the configuration personnel need to spend a lot of time on path coverage testing to ensure the correctness of the process. Therefore, the existing methods have major problems in the ease of use and convenience of process configuration. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method and device for automatic configuration of business processes based on a large language model, thereby reducing the difficulty of configuration of configuration personnel and saving time investment.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A business process automation configuration method based on a large language model, comprising:

[0007] Obtaining a user requirement text, and obtaining a context constraint corresponding to the user requirement text according to a business scenario;

[0008] Inputting the user requirement text and context constraints into a first language model to obtain preliminary standard data;

[0009] The preliminary standard data is parsed to obtain a business process, and it is determined whether the business process meets the expected business requirements. If not, the conversation content data is obtained, and a cyclic conversation is performed on the first language model based on the conversation content data and the preliminary standard data until the obtained business process meets the expected business requirements.

[0010] In order to solve the above technical problems, another technical solution adopted by the present invention is:

[0011] A business process automation configuration device based on a large language model includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, each step in the above-mentioned business process automation configuration method based on a large language model is implemented.

[0012] The beneficial effects of the present invention are as follows: preliminary standard data is obtained by a large language model according to user demand text and context constraints, and the business process is continuously improved in combination with the conversation content data and the preliminary standard data updated in each round of conversation, thereby obtaining a business process that meets business needs; the powerful natural language understanding ability of the large language model can fully understand business needs, save time for demand organization and business abstraction, thereby reducing the learning cost and technical threshold of process configuration personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A flowchart of the steps of a method for automatic configuration of a business process based on a large language model in an embodiment of the present invention;

[0014] Figure 2 It is another step flow chart of a method for automatic configuration of business processes based on a large language model in an embodiment of the present invention;

[0015] Figure 3 A flowchart of the steps of automated test correction in a business process automated configuration method based on a large language model in an embodiment of the present invention;

[0016] Figure 4 It is a result schematic diagram of a business process automation configuration device based on a large language model in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to explain the technical content, achieved objectives and effects of the present invention in detail, the following is an explanation in combination with the implementation modes and the accompanying drawings.

[0018] Please refer to Figure 1, a business process automation configuration method based on a large language model, comprising:

[0019] Obtaining a user requirement text, and obtaining a context constraint corresponding to the user requirement text according to a business scenario;

[0020] Inputting the user requirement text and context constraints into a first language model to obtain preliminary standard data;

[0021] The preliminary standard data is parsed to obtain a business process, and it is determined whether the business process meets the expected business requirements. If not, the conversation content data is obtained, and a cyclic conversation is performed on the first language model based on the conversation content data and the preliminary standard data until the obtained business process meets the expected business requirements.

[0022] From the above description, it can be seen that the beneficial effects of the present invention are: preliminary standard data is obtained according to user demand text and context constraints by a large language model, and the business process is continuously improved in combination with the conversation content data and the preliminary standard data updated in each round of conversation, so as to obtain a business process that meets business needs; the powerful natural language understanding ability of the large language model can fully understand business needs, save time for demand organization and business abstraction, thereby reducing the learning cost and technical threshold of process configuration personnel.

[0023] Furthermore, until the obtained business process meets the expected business requirements, it also includes:

[0024] Adding test case generation requirements to the business process and inputting them into the second language model in a preset format text;

[0025] The second largest language model obtains test cases covering the business process according to the test case generation requirement;

[0026] The first language model is tested using the test case.

[0027] From the above description, it can be seen that by automatically generating test cases covering the complete path based on the test case generation requirements through the second language model, the test work can be completed more quickly and the test efficiency can be improved.

[0028] Further, the testing the first language model through the test case includes:

[0029] Record the test results corresponding to the test cases to obtain execution logs;

[0030] Determine whether the test case fails to execute, and if so, input the execution log into a third language model;

[0031] The third language model obtains a correction plan according to the execution log, and repairs the test case according to the correction plan and continues to execute the test until the test passes.

[0032] From the above description, it can be seen that an execution log is generated by recording the test results of the test case, and when the test case fails to execute, the execution log can be automatically analyzed through the third language model to obtain the corresponding correction plan to repair the test case and continue to execute the test, thereby improving the accuracy and efficiency of the test.

[0033] Furthermore, the third language model obtains a correction scheme according to the execution log, including:

[0034] Obtaining a status code according to the execution log, and determining an error type according to the status code;

[0035] A solution corresponding to the error type is executed according to the error type.

[0036] From the above description, it can be seen that by setting different status codes for different execution errors, the third language model can quickly identify the error type of the test case according to the status code, thereby improving the efficiency of correcting the test case.

[0037] Further, the status code includes a first preset value;

[0038] If the status code is the first preset value, determine whether the execution log conforms to the use case result; if not, obtain the current test node configuration information, use case requirements and current execution results to correct the test configuration and re-execute the test case.

[0039] From the above description, it can be seen that the first preset value is used to indicate that the test case contains correct results, and the results of the process are certain after following each step, but there is an error in the process result; this error indicates that there is an error in the configuration, thereby prompting the need to check whether there are any abnormalities in all configurations of the node, and complete the correction of the test configuration based on the inspection results.

[0040] Further, the status code includes a second preset value;

[0041] If the status code is the second preset value, obtaining a corresponding test configuration description according to the status code;

[0042] Obtain the current node configuration information, use case requirements, current execution results, and configuration instructions to correct the test configuration and re-execute the test case.

[0043] From the above description, it can be seen that the second preset value represents the preset monitoring error that occurs during the test process; this type of error implements the preset code and protects the detailed repair instruction document, namely the test configuration instruction, so the large language model can be repaired by pulling the corresponding status code instruction document.

[0044] Further, the status code includes a third preset value;

[0045] If the status code is the third preset value, obtaining error information according to the status code;

[0046] Obtain the current node configuration information, use case requirements, current execution results, and error correction test configuration, and determine whether the change of the test configuration can be completed. If so, re-execute the test case; if not, notify manual intervention for correction.

[0047] From the above description, it can be seen that the third preset value indicates a program running error during the test; when this error occurs, the execution log will be attached with detailed error stack information and program running process record information, which means that an unexpected program error has occurred. The large language model parses the error and running information according to the detailed description of the interface, and prioritizes self-repair by modifying the configuration and parameters according to the implementation goals; when the configuration and parameter repair cannot be performed, a product code upgrade suggestion is made, which is subsequently repaired manually.

[0048] Furthermore, after parsing the preliminary standard data to obtain the business process, the process further includes:

[0049] The business process is visualized and displayed on a visualization interface.

[0050] From the above description, it can be seen that by visually analyzing the preliminary standard data and displaying the business process on the visual interface, the configuration personnel can understand the specific business process more intuitively through the visual interface, which facilitates the optimization of the generated business process.

[0051] Furthermore, until the obtained business process meets the expected business requirements, it also includes:

[0052] Obtain optimization instructions, and optimize the business process on the visualization interface according to the optimization instructions.

[0053] From the above description, it can be seen that by obtaining the optimization instructions issued by the configuration personnel based on the visual interface, the business process can be further optimized. In the process of adjusting the visual interface, if major problems are found or regeneration is required, the model dialogue function of the process visualization interface can be called again to further communicate with the large language model to improve the process, thereby regenerating standard data and obtaining a business process that meets business needs.

[0054] Another embodiment of the present invention provides a business process automation configuration device based on a large language model, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, each step in the above-mentioned business process automation configuration method based on a large language model is implemented.

[0055] The method and device for automatic configuration of business processes based on a large language model provided by the present invention can be applied to the scenario of automatically generating business processes, which is described below through specific implementation methods:

[0056] Embodiment 1

[0057] Please refer to Figure 1 as well as Figure 2 , a business process automation configuration method based on a large language model, comprising:

[0058] S1. Obtain the user demand text, and obtain the context constraints corresponding to the user demand text according to the business scenario. Specifically: collect the user's needs and organize them into text to obtain the user demand text, and then confirm the necessary large language model context constraints required for the business scenario; the context constraints include but are not limited to: specific service interface descriptions based on the implementation of this system, such as: sql execution interface description (including system database design description, used to generate test library tables and test data), various business processing interface descriptions (used for large language model to obtain system, users, roles, organizations, system dictionaries and various general interfaces for services in the process), key design and implementation descriptions of the process visualization engine (used for large language model to organize business processing), and supplementary information about other systems, etc.

[0059] At the same time, the above interface description uses a dynamic interface to obtain directly from the program itself, rather than manual maintenance; when the program starts, the code structure and annotation information are parsed to generate key interface information, which is automatically updated as the program is updated. The library table design uses the method of directly reading the database structure to achieve real-time library table design information acquisition, avoiding manual maintenance and version inconsistency problems. Other functional configuration instructions or design implementation instructions change with the version change, and the data volume is large, there is a problem that the context length of the large language model may exceed the problem. Even if it does not exceed, it increases the calculation pressure of the large language model. After storing this part of information in the product design description library according to the product version classification, the vectorization technology is used to vectorize the data. When the large language model needs some design or configuration case information, the vectorization query interface is called to find the most matching N pieces of information as the context of the large language model for further large language model interaction. At the same time, as the system continues to run, the results of each configuration will be recorded and parsed; the system updates the basic weight information recorded in the vector library according to the results of the analysis to improve the accuracy of vector information retrieval and ultimately improve configuration efficiency.

[0060] S2. Input the user demand text and context constraints into the first language model to obtain preliminary standard data. Specifically: the user demand text and context constraints can be input into the first language model through the model dialogue interface of the process visualization interface. The first language model generates preliminary BPMN standard data according to the user demand text and context constraints; among them, the data that needs to be dynamically read will be merged into the BPMN standard data after data reading by various large language model agents, and finally output in a preset format as required. At the same time, the first language model can generate standard BPMN process data as required based on the additional information in the context constraints, and can also generate business processing configurations for each process node in combination with various interface instructions and system design implementation instructions provided by this system, such as specifying node handlers, calling SMS sending under specified conditions, setting node timeouts, changing data status, and generating and executing specific business scripts. In addition, the additional content is constantly improved with the iteration and upgrade of the system, and some content can also be added or deleted in the visualization interface according to the needs of users.

[0061] S3. Analyze the preliminary standard data to obtain the business process, and determine whether the business process meets the expected business requirements. If not, obtain the conversation content data, and conduct a circular conversation with the first language model based on the conversation content data and the preliminary standard data until the obtained business process meets the expected business requirements. Specifically:

[0062] S31. After receiving the BPMN standard data output by the first language model, the process visualization engine performs visual analysis to obtain the business process and displays the business process on the visualization interface. At this time, in addition to completing the business process diagram configuration, each node of the system has completed the basic business processing configuration.

[0063] S32. Users can view the content displayed by the process visualization engine. If the business process does not meet the expected business needs, such as omissions, errors, etc., the user can further communicate with the large language model through the model dialogue function of the process visualization interface, so that after the system obtains the optimization instruction, it optimizes the business process on the visualization interface through the optimization instruction, that is, the process visualization engine inputs the BPMN standard data generated in the previous step plus multiple rounds of dialogue content into the first large language model, and the first large language model will regenerate new BPMN standard data and output and visualize it; repeat this step until there are no obvious problems with the business process and meet the expected business needs; at the same time, according to the basic business process that has been generated, the optimization and details of the display are adjusted through the visualization interface until the needs are met; during the adjustment of the visualization interface, if major problems are found or regeneration is required, the model dialogue function of the process visualization interface can be called again to further communicate with the large language model to improve the process and regenerate the BPMN standard data.

[0064] S4. Generate test cases to test the first language model: attach test case generation requirements to the business process, and input them into the second language model in a preset format text, so that the second language model obtains test cases covering the business process according to the test case generation requirements, and then tests the first language model through the test cases; that is, after manually confirming that the configuration content is basically correct through the above steps, attach the test case generation requirements to the generated process configuration information, and input it into the second language model in a specific format text, and the second language model generates test cases covering the complete path according to the requirements; wherein the test case at least includes the original requirements, corresponding configuration items, input cases, correct outputs, and other implementation process descriptions, and the visual process engine parses the case text information into a visual interface and displays it to the user, and the user can further adjust the test case through the visual interface or confirm that the test case enters the process automation test; wherein the specific testing method is as follows:

[0065] Automated testing is implemented using a large language model agent. A Python-based automated testing agent is customized according to the visual process. The agent implements browser opening of the test address, specific user login, simulated manual operations and interface calls according to the test cases. During the process, the output of the interface and console is collected to determine whether the execution of the entire process is successful, and a report on the entire test process is output. If an error occurs during the process, according to the user's choice, the agent can call the large language model to achieve self-correction and test again until the test passes, or it can interrupt the test and return to the step S3 visual interface for manual intervention and modification. Among them, the agent can call the large language model to achieve self-correction, which does not refer to the correction of the large language model, but refers to the correction of the process configuration or the configured dynamic script code fragment. Specifically:

[0066] S41. Record the test results corresponding to the test cases and obtain the execution logs. During the test process, the process is discovered and the test cases are successfully executed or errors occur. Each execution of the test case will generate a UUID log number. During the test process, all execution logs will be recorded on the server with this UUID as the log number.

[0067] S42. Determine whether the test case fails to execute. If so, input the execution log into the third language model. After the execution of this test case fails, the Agent will call the server log acquisition interface to obtain the log information of the entire test process according to the UUID log number. The log contains the server execution process of the entire test case, interface parameters, environment variables when executing the interface, the execution time of each interface or key steps, and even the values ​​of some key intermediate parameter variables.

[0068] S43. The third language model obtains a correction plan according to the execution log, and repairs the test case according to the correction plan and continues to execute the test until the test passes. Specifically, the status code is obtained according to the execution log, and the error type is determined according to the status code. Then, the plan corresponding to the error type is executed according to the error type, that is, all execution process logs of the use case are sent to the large language model to determine the problem point:

[0069] 1. If it is a configuration error, further product configuration instructions will be obtained and then corrected directly.

[0070] 2. If it is a program error, the corresponding source code location and version in the system interface description can be used to pull the corresponding relevant code from the source code library for further analysis; if it can be solved by modifying the configuration or modifying the dynamic script fragment, the Agent can directly repair it according to the correction suggestions given by the large language model and continue testing.

[0071] 3. If the problem cannot be solved by modifying the configuration or the dynamic script fragment, correction suggestions will be given and manual takeover will be carried out.

[0072] At the same time, after the problem is finally solved, the entire solution process is refined and summarized through the large language model and then vectorized and stored in the vector database. When similar problems are encountered in future self-testing and repair, the repair steps in the vector database are matched first to achieve rapid repair; external memory is used to supplement the knowledge blind spots of the general large language model in the system, avoiding excessive cost for multiple model fine-tuning or even training.

[0073] Please refer to Figure 3 The interface returns an additional status code, which is generally divided into three categories: the status code includes the first preset value (200), the second preset value (custom error code above 10000) and the third preset value (500); for example, the status of successful process execution is 200 and the specific content of the process execution result is attached; an unexpected error occurs in the process execution, the status code is 500 and the error message is attached; the preset monitoring error code occurs in the process execution, and the status starts with 10000, such as 10001 for no permission to submit the process, 10002 for the executor of the next node cannot be found, 10003 for the process judgment condition error..., 20000 for the process dynamic script execution error, etc. These status codes have detailed explanations and solutions in our product library, as the product continues to improve.

[0074] Category 1: If the status code is the first preset value, determine whether the execution log is consistent with the use case result. If not, obtain the current test node configuration information, use case requirements, and current execution results to correct the test configuration and re-execute the test case. That is, the status code returned by the interface after the operation is normal (such as status code 200), but the process result is wrong, indicating that there is an error in the configuration. At this time, it is necessary to check whether all configurations of the node are abnormal.

[0075] The second category: If the status code is the second preset value, obtain the corresponding test configuration instructions according to the status code; obtain the current node configuration information, use case requirements, current execution results and configuration instructions to correct the test configuration, and re-execute the test case. That is, at this time, a status code above 10000 is returned; and these status codes are pre-set codes, containing detailed repair instructions documents, and the large language model implements the repair by pulling the corresponding status code description documents. Among them, the server-side interface execution process of the process engine will record sufficient execution process information and provide it to the large language model as reference information for modification.

[0076] Category 3: If the status code is the third preset value, then the error information is obtained according to the status code; the current node configuration information, use case requirements, current execution results, and error correction test configuration are obtained, and it is determined whether the test configuration change can be completed. If so, the test case is re-executed; if not, manual intervention is notified for correction. That is, when the interface returns a status code of 500, it represents a program running error, and detailed error stack information and program running process record information will be attached. At this time, it is an unexpected program error. The large language model analyzes the detailed description of the interface combined with the error and running information, and prioritizes self-repair by modifying the configuration, parameters, etc. according to the implementation goals; if the configuration and parameter repair cannot be made, the product code upgrade suggestion is proposed, and it will be repaired manually later.

[0077] Among them, the first language model, the second language model and the third language model in this embodiment are general language models, but they need to be further enhanced by training; for example, the basic development framework based on springboot and vue encapsulation, the general interface, method, sample configuration and design materials provided are used as relatively fixed content to supplement the training in the large language model to form a large language model enhanced in specific aspects; that is, the three large language models provided in this embodiment are the same model, but the context content provided and the encapsulated Agent are different. For example, the basic process of the Agent used to judge the problem point is: when the interface is not returned according to the regulations, the large language model parses the problem that the status code is 10001, and outputs it to the Agent in a specific format. The Agent calls the interface to obtain the document information of the status code, and the large language model gives correction suggestions for the document processing. The Agent then calls the configuration interface to implement the configuration update, and then enters the automated testing link. The main purpose of supplementary training is to enable the large language model to better provide solutions within the development framework and design system.

[0078] Embodiment 2

[0079] Please refer to Figure 3 , a business process automation configuration device based on a large language model, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that when the processor executes the computer program, each step in a business process automation configuration method based on a large language model as described in Example 1 is implemented.

[0080] In summary, the business process automation configuration method and device based on the large language model provided by the present invention can easily understand large-scale business requirements based on the powerful natural language understanding of the large language model, thereby saving the configuration personnel's time for requirement organization and business abstraction, and reducing the learning cost and technical threshold of process configuration personnel; at the same time, the large language model combined with constraints can output standard BPMN standard data according to business requirements, and the process visualization engine parses the BPMN standard data to generate a visualization interface to achieve convenient human-computer interaction; in addition, through the context understanding ability of the large language model, multiple rounds of large language model human-computer interaction can be achieved, and the configuration and adjustment of the process can be easily completed through simpler natural language; the BPMN standard data formed after the configuration is completed can be easily understood and summarized by the large language model, and finally a test case covering the complete path is generated, which is combined with the powerful large language model Agent technology to achieve faster and more complete process testing.

[0081] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent transformations made using the contents of the present invention's specification and drawings, or directly or indirectly applied in related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A business process automation configuration method based on a large language model, characterized in that: include: Obtaining a user requirement text, and obtaining a context constraint corresponding to the user requirement text according to a business scenario; Inputting the user requirement text and context constraints into a first language model to obtain preliminary standard data; The preliminary standard data is parsed to obtain a business process, and it is determined whether the business process meets the expected business requirements. If not, the conversation content data is obtained, and a cyclic conversation is performed on the first language model based on the conversation content data and the preliminary standard data until the obtained business process meets the expected business requirements.

2. The method for automatic configuration of business processes based on a large language model according to claim 1, characterized in that: After the business process obtained meets the expected business requirements, the following steps are further included: Adding test case generation requirements to the business process and inputting them into the second language model in a preset format text; The second largest language model obtains test cases covering the business process according to the test case generation requirement; The first language model is tested using the test case.

3. The method for automatic configuration of business processes based on a large language model according to claim 2, characterized in that: The testing of the first language model by using the test case includes: Record the test results corresponding to the test cases to obtain execution logs; Determine whether the test case fails to execute, and if so, input the execution log into a third language model; The third language model obtains a correction plan according to the execution log, and repairs the test case according to the correction plan and continues to execute the test until the test passes.

4. The method for automating business process configuration based on a large language model according to claim 3, characterized in that: The third language model is corrected according to the execution log, including: Obtaining a status code according to the execution log, and determining an error type according to the status code; A solution corresponding to the error type is executed according to the error type.

5. The method for automatic configuration of business processes based on a large language model according to claim 4, characterized in that: The status code includes a first preset value; If the status code is the first preset value, determine whether the execution log conforms to the use case result; if not, obtain the current test node configuration information, use case requirements and current execution results to correct the test configuration and re-execute the test case.

6. The method for automatic configuration of business processes based on a large language model according to claim 4, characterized in that: The status code includes a second preset value; If the status code is the second preset value, obtaining a corresponding test configuration description according to the status code; Obtain the current node configuration information, use case requirements, current execution results, and configuration instructions to correct the test configuration and re-execute the test case.

7. The method for automatic configuration of business processes based on a large language model according to claim 4, characterized in that: The status code includes a third preset value; If the status code is the third preset value, obtaining error information according to the status code; Obtain current node configuration information, use case requirements, current execution results, and error correction test configuration, and determine whether the test configuration can be changed. If so, re-execute the test case; If not, notify human intervention for correction.

8. The method for automatic configuration of business processes based on a large language model according to claim 1, characterized in that: After parsing the preliminary standard data to obtain the business process, the following further includes: The business process is visualized and displayed on a visualization interface.

9. The method for automatic configuration of business processes based on a large language model according to claim 8, characterized in that: After the business process obtained meets the expected business requirements, the following steps are further included: Obtain optimization instructions, and optimize the business process on the visualization interface according to the optimization instructions.

10. A business process automation configuration device based on a large language model, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, each step of the method for automatic configuration of a business process based on a large language model as described in any one of claims 1 to 9 is implemented.

Images