RPA process error positioning and repairing scheme based on large language model

Abstract

The invention discloses an RPA process error positioning and repairing scheme based on a large language model, which is applied to an RPA process. Comprising the following steps of 1, capturing error events in RPA process execution in real time, collecting key information, and performing data structured processing on the collected key information to obtain structured error reporting information; 2, introducing a problem positioning Agent big language model, designing a problem positioning cue word, analyzing the context of the structured error report information, and obtaining an error positioning result; step 3, introducing a suggestion providing Agent large language model, designing suggestion providing cue words, analyzing error reasons based on an error positioning result and the structured error information, and generating repair suggestions; and 4, analyzing and integrating the error positioning result and the repair suggestion, and feeding back to a user in a visual form. According to the invention, the error positioning of the RPA process can be realized, the repair suggestion can be provided, the repair use threshold of the error report of the RPA process is reduced, and the maintenance efficiency of the RPA process is improved.

Description

Technical Field

[0001] The field of the present invention is the field of robotic process automation technology, and specifically relates to an RPA process error location and repair solution based on a large language model. Background Art

[0002] Robotic process automation (RPA) is a technology that automates the execution of repetitive and rule-based tasks by simulating human operations through software robots. During the development and operation of RPA processes, various errors will inevitably occur. Users usually only see the error message of a certain line, and this message may not point to the initial position where the error occurred. Moreover, RPA processes are usually complex, and fixing errors requires understanding the global context, not just correcting errors in local lines. And error location often requires debugging the entire process code through cumbersome and complex steps, which is time-consuming and laborious. In addition, in the face of error messages, especially for novice users who are not familiar with programming or RPA rules, it is difficult to directly infer effective solution steps from the error prompts, resulting in problems that cannot be repaired in a timely and effective manner. Traditional error messages only provide simple error cause prompts, such as syntax errors, undefined variables, type mismatches, etc., and do not give detailed repair suggestions. Summary of the Invention

[0003] The purpose of the present invention is to provide an RPA process error location and repair solution based on a large language model. The present invention can achieve error location of the RPA process and provide repair suggestions, reducing the repair usage threshold of RPA process errors and improving the maintenance efficiency of RPA processes.

[0004] The technical solution of the present invention: An RPA process error location and repair solution based on a large language model, comprising the following steps:

[0005] Step 1: Real-time capture error events during the execution of the RPA process, collect key information, and perform data structuring on the collected key information to obtain structured error information;

[0006] Step 2: Introduce a problem location Agent large language model, design problem location prompt words to drive the problem location Agent large language model to analyze the context of the structured error information, and obtain an error location result;

[0007] Step 3: Introduce a suggestion providing Agent large language model, design suggestion providing prompt words, and based on the error location result and the structured error information, guide the suggestion providing Agent large language model to analyze the error cause and generate repair suggestions;

[0008] Step 4: Parse and integrate the error location result and the repair suggestions, and feedback them to the user in a visual form.

[0009] In the aforementioned RPA process error reporting location and repair solution based on a large language model, in step 1, the key information includes the RPA instruction source code, error message, and variable stack.

[0010] In the aforementioned RPA process error reporting location and repair solution based on a large language model, the RPA instruction source code contains the input parameters and return values required by the instruction, the instruction line number, and the process information to which the instruction belongs, facilitating the problem location Agent large language model to accurately locate the problem line.

[0011] In the aforementioned RPA process error reporting location and repair solution based on a large language model, the error message contains the basic error cause and the error line number.

[0012] In the aforementioned RPA process error reporting location and repair solution based on a large language model, the variable stack includes the names, types, and current values of all variables within the scope.

[0013] In the aforementioned RPA process error reporting location and repair solution based on a large language model, in step 2, the design process of the problem location prompt is as follows:

[0014] Step 1: Use XML tags to divide different parts to help the problem location Agent large language model parse different components of the problem location prompt and accurately grasp the task requirements;

[0015] Step 2: Define the role of the problem location Agent large language model as an "error reporting location expert";

[0016] Step 3: Clearly define the task of the problem location Agent large language model as "combining the information provided by the user to locate the initial instruction causing the error";

[0017] Step 4: Clearly define the target audience as "computer novices" and emphasize the "basic errors" that often occur in writing applications;

[0018] Step 5: Describe that the user input includes the instruction source code, variable stack, and error message;

[0019] Step 6: Require the problem location Agent large language model to output the error location result in json format.

[0020] In the aforementioned RPA process error reporting location and repair solution based on a large language model, in step 3, the design process of the suggestion providing prompt is as follows:

[0021] Step 1: Use XML tags to divide different parts to help the suggestion providing Agent large language model parse different components of the suggestion providing prompt and accurately grasp the task requirements;

[0022] Step 2: Define the role of the Agent large language model for suggestion provision as "error analysis expert".

[0023] Step 3: Specify the task of the Agent large language model for suggestion provision as "informing the user of the error cause and how to fix it".

[0024] Step 4: Identify the target audience as "computer novices".

[0025] Step 5: Describe that the user input includes instruction source code, variable stack, error message, and error location result.

[0026] Step 6: Require the Agent large language model for suggestion provision to output the error cause and repair suggestions in json format.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] The present invention utilizes the semantic understanding ability of the problem location Agent large language model to extract richer context information from the process code, error message, and variable stack information, thereby conducting more in-depth error analysis and more accurately locating the root cause of the error. The present invention utilizes the Agent large language model for suggestion provision to not only analyze the error cause but also provide targeted repair suggestions based on the context information, greatly reducing the time for users to troubleshoot errors and find solutions by themselves. In addition, the present invention also parses and integrates the outputs of the problem location Agent large language model and the Agent large language model for suggestion provision, and then provides easy-to-understand error analysis and repair suggestions to help novice users quickly understand and solve problems, reducing the repair usage threshold of RPA errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is a flowchart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0030] The present invention will be further described below in conjunction with the drawings and embodiments, but it shall not be used as a basis for limiting the present invention.

[0031] Embodiment: An RPA process error location and repair solution based on a large language model, the specific process is as Figure 1 shown, applied to the RPA process, mainly including the following steps:

[0032] Step 1: When an error occurs during the execution of the RPA process, monitor and capture the error event in real time; after capturing the error event, automatically collect key information and perform data structuring processing on the collected key information to obtain structured error information.

[0033] In this embodiment, the key information includes the RPA instruction source code, error messages, and variable stacks.

[0034] The RPA instruction source code refers to the Python source code behind the RPA instruction. To prevent the large language model from not understanding the Python code after RPA application encapsulation, the instruction function description is written in the form of comments on the line above the Python code. The RPA instruction source code contains the input parameters and return values required by the instruction, the instruction line number, and the process information to which the instruction belongs, facilitating the problem-locating Agent large language model to accurately locate the problem line. The RPA instruction source code is shown in Table 1.

[0035] Table 1 RPA Instruction Source Code Table

[0036]

[0037]

[0038] The error message includes the basic error cause and the error line number. For example, "Error at line 4 in main process.flow: Value out of range of the list".

[0039] The variable stack is a snapshot of the variables at the time of the error. The variable stack includes the names, types, and current values of all variables within the scope (the scope refers to the visibility and lifecycle range of variables, functions, or objects). For example, "{"web_page": "web page object", "web_data_table": [['U2021041417063', completed'], ['U2021041417064', completed'], ['U2021041417065', cancelled']]}". The variable stack is crucial for locating data-related errors.

[0040] In this embodiment, the collected error messages are processed for data structuring, and the structured data is as follows:

[0041] {

[0042] "main process.flow": "Python source code with instruction description",

[0043] "sub process.flow": "Python source code with instruction description",

[0044] "error_msg": "Error at line 6 in main process.flow>sub process.flow: Basic error message",

[0045] "variable": {"variable name 1": variable value 1, "variable name 2": variable value 2}

[0046] }

[0047] Step 2: Introduce the problem localization Agent large language model, design problem localization prompts to drive the problem localization Agent large language model to analyze the context of the structured error message, and obtain the error localization result;

[0048] In this embodiment, in order to more effectively utilize the large language model, the present invention introduces the problem localization Agent large language model. The Agent large language model (Agent-based Large Language Model) is an AI system that combines intelligent agent technology with the large language model (LLM), aiming to complete complex goals through autonomous decision-making, task planning, and environmental interaction. The working principle of the Agent is to utilize the powerful capabilities of the large language model and drive it to perform specific tasks through carefully designed problem localization prompts (Prompts). The process of designing the problem localization prompts for the problem localization Agent large language model is shown in Table 2.

[0049] Table 2 Design of Problem Localization Prompts

[0050]

[0051]

[0052] The specific steps for designing the problem localization prompts are as follows:

[0053] Step 1: Use XML tags to divide different parts, which helps the problem localization Agent large language model better parse different components of the problem localization prompts and accurately grasp the task requirements;

[0054] Step 2: <player>The label defines that the role of the problem - locating Agent large - language model is the "error - reporting location expert", which provides a clear role definition for the problem - locating Agent large - language model, enabling it to better play this role;

[0055] Step 3: <responsibility>The task of the Label Clear Problem Locating Agent large language model is to "locate the original instruction that caused the error in combination with the information provided by the user".

[0056] Step 4: <audience>The label clearly states that the target audience is "computer novices" and emphasizes the "elementary mistakes" that often occur in writing applications. It requires that when analyzing problems, the problem localization Agent large language model should think from the user's perspective and use simple and easy-to-understand language to explain, avoiding the use of overly professional terms;

[0057] Step 5: The <audience_input> label describes that the user input includes the instruction source code, variable stack, and error message;

[0058] Step 6: Require the problem localization Agent large language model to output the error localization result in json format for easy parsing and integration.

[0059] Then, input the structured error message obtained in Step 1 into the problem localization Agent large language model, call the problem localization Agent large language model, and obtain the error localization result. The formula is as follows:

[0060] Error localization result = problem localization agent (error message)

[0061] = {"position": "Line 4 of main process.flow"}.

[0062] Step 3: Introduce the suggestion providing Agent large language model, design the suggestion providing prompt words, and based on the error localization result and the structured error message, guide the suggestion providing Agent large language model to analyze the cause of the error and generate repair suggestions;

[0063] In this embodiment, in order to make more effective use of the large language model, the present invention introduces the suggestion providing Agent large language model. For the result output by the problem localization Agent large language model, it further analyzes the cause of the error and gives detailed repair suggestions. The design process of the suggestion providing prompt words of the suggestion providing Agent large language model is shown in Table 3.

[0064] Table 3 Design of Suggestion Providing Prompt Words

[0065]

[0066] The specific steps for designing the suggestion providing prompt words are as follows:

[0067] Step 1: Use XML tags to divide different parts, which helps the suggestion providing Agent large language model better parse different components of the suggestion providing prompt words and accurately grasp the task requirements;

[0068] Step 2: <player>The label definition suggests that the role of the Agent large language model is "error analysis expert";

[0069] Step 3: <responsibility>The label clearly recommends that the task of providing the Agent large language model is to "inform the user of the error reason and how to modify it";

[0070] Step 4: <audience>The label clearly states that the target audience is "computer novices", and it is required that the explanations provided by the Agent large language model be easy to understand, and the suggestions given be clear and executable.

[0071] Step 5: The <audience_input> label describes that the user input contains the instruction source code, variable stack, error message, and error location result.

[0072] Step 6: It is required that the Agent large language model provide error reasons and repair suggestions in json format.

[0073] Then, based on the structured error message and error location result, call the Agent large language model that provides suggestions to obtain repair suggestions. The formula is as follows:

[0074] Repair suggestions = Provide suggestions agent (error message error location result)

[0075] = {"reason": "The subscript exceeds the list length", "suggest": "Change the hard-coded index 9 to -1"}.

[0076] Step 4: Parse and integrate the error location result and repair suggestions, and then feedback them to the user in a visual form.

[0077] In this embodiment, the outputs of the problem location Agent large language model and the suggestion providing Agent large language model are parsed and integrated to obtain the final error location and repair suggestions. The integration result formula is as follows:

[0078]

[0079] In this embodiment, the final error location and repair suggestions are returned to the user in a clear and easy-to-understand form.

[0080] The present invention utilizes the semantic understanding ability of the problem location Agent large language model to extract richer context information from the process code, error message, and variable stack information, so as to conduct more in-depth error analysis and more accurately locate the root cause of the error. The present invention utilizes the suggestion providing Agent large language model not only to analyze the error reason, but also to provide targeted repair suggestions according to the context information, greatly reducing the time for users to troubleshoot errors and find solutions by themselves. In addition, the present invention also parses and integrates the outputs of the problem location Agent large language model and the suggestion providing Agent large language model, and then provides easy-to-understand error analysis and repair suggestions to help novice users quickly understand and solve problems, reducing the repair and use threshold of RPA errors. The method of the present invention can automatically perform error analysis and repair, reducing the time for manual error troubleshooting and improving the maintenance efficiency of the RPA process.

[0081] In summary, the present invention can achieve error location of the RPA process and provide repair suggestions, reducing the repair usage threshold of RPA process errors and improving the maintenance efficiency of the RPA process.< / audience> < / responsibility> < / player> < / audience> < / responsibility> < / player>

Claims

1. An RPA process error location and repair solution based on a large language model, applied to the RPA process; characterized in that, It includes the following steps: Step 1: Real-time capture of error events during the execution of the RPA process, collection of key information, and data structuring of the collected key information to obtain structured error messages; Step 2: Introduce the problem localization Agent large language model, design problem localization prompt words to drive the problem localization Agent large language model to analyze the context of the structured error message, and obtain the error localization result; Step 3: Introduce the suggestion providing Agent large language model, design suggestion providing prompt words, and based on the error localization result and the structured error message, guide the suggestion providing Agent large language model to analyze the error cause and generate repair suggestions; Step 4: Parse and integrate the error localization result and the repair suggestions and feedback them to the user in a visual form.

2. The RPA process error reporting location and repair solution based on the large language model according to claim 1, characterized in that: In Step 1, the key information includes the RPA instruction source code, error message, and variable stack.

3. The RPA process error location and repair solution based on the large language model according to claim 2, characterized in that: The RPA instruction source code contains the input parameters and return values required by the instruction, the instruction line number, and the process information to which the instruction belongs, facilitating the problem localization Agent large language model to accurately locate the problem line.

4. The RPA process error location and repair solution based on the large language model according to claim 2, characterized in that: The error message contains the basic error cause and the error line number.

5. The RPA process error reporting location and repair solution based on the large language model according to claim 2, wherein: The variable stack includes the names, types, and current values of all variables within the scope.

6. The RPA process error location and repair solution based on the large language model according to claim 1, characterized in that: In Step 2, the design process of the problem localization prompt words is as follows: Step 1: Use XML tags to divide different parts to help the problem localization Agent large language model parse the different components of the problem localization prompt words and accurately grasp the task requirements; Step 2: Define the role of the problem localization Agent large language model as an "error localization expert"; Step 3: Clearly define the task of the problem localization Agent large language model as "combining the information provided by the user to locate the initial instruction that caused the error"; Step 4: Clearly define the target audience as "computer novices" and emphasize the "low-level errors" that often occur in application writing; Step 5: Describe that the user input includes the instruction source code, variable stack, and error message; Step 6: Require the problem localization Agent large language model to output the error localization result in json format.

7. The RPA process error reporting location and repair solution based on the large language model according to claim 1, characterized in that: In Step 3, the design process of the suggestion providing prompt words is as follows: Step 1: Use XML tags to divide different parts to help the suggestion providing Agent large language model parse the different components of the suggestion providing prompt words and accurately grasp the task requirements; Step 2: Define the role of the suggestion providing Agent large language model as an "error analysis expert"; Step 3: Clearly define the task of the suggestion providing Agent large language model as "informing the user of the error cause and how to modify it"; Step 4: Clearly define the target audience as "computer novices"; Step 5: Describe that the user input includes the instruction source code, variable stack, error message, and error localization result; Step 6: Require the suggestion providing Agent large language model to output the error cause and repair suggestions in json format.

Images